195 results about "Echo state network" patented technology

The invention discloses a high-speed train tracking running curve optimization setting method. According to the characteristic of 'movable and dynamic length' of a tracking section of a high-speed train under movable blocking condition, the method establishes a high-speed train echo state network speed prediction model, a movable blocking based tracking running model, a line network and a tracking running curve multi-target setting model adopting innovative evaluation indexes based on line and high-speed train running data acquired in a site; an efficient multi-target particle swarm optimization is adopted to use an algorithm convergence condition as one of model setting constraints, and high-speed train tracking running curve optimization setting is performed based on the real-time data; finally, section operation efficiency and stability are used as the evaluation indexes of the setting method, and a group of optimal running curves are screened out, so that the high-speed train running process is safe and efficient, and meanwhile the high-speed train section operating efficiency and stability under the movable blocking condition are improved.

The invention discloses a lithium ion battery service life forecasting method based on an integrated model and relates to a lithium ion battery cycle life forecasting method. The lithium ion battery service life forecasting method is used for solving the problem that the existing lithium ion battery is low in service life forecasting adaptability and poor in stability. The lithium ion battery service life forecasting method includes: performing preprocessing on battery cycle charging and discharging test testing data; adopting a Bagging algorithm to perform secondary resampling on a Train database; building a monotonous echo state network model; initializing inner connection weights of a monotonous echo state network, and repeating for T times to obtain T untrained monotonous echo state network sub-models; setting a first free parameter set and a second free parameter set of the monotonous echo state network model; integrating output RULi of the monotonous echo state network model, adopting the Test database to drive the integrated monotonous echo state network model, and obtaining remaining service life of a lithium ion battery. The lithium ion battery service life forecasting method based on the integrated model is suitable for lithium ion battery service life forecasting.

The invention discloses an ultra-short-term power load forecasting and early warning method based on a Kalman filter and wavelet echo state network. In order to solve the problem that noise and the like are contained in power load data, a Kalman filtering method is adopted to conduct real-time estimating on 'collected data', with the help of a forgetting factor, the weight of old-fashioned data is weakened, and prediction accuracy is improved. Before ultra-short-term load forecasting is conducted, firstly, a principal component is used for analyzing and determining main working procedures for influencing the change of a power load, the main working procedures are used as the input of a power load capacity prediction model, afterwards, wavelets are used for decomposing the loads of different spectral characteristics (high frequency, follow-up and stability) of the power load, echo state network singe power loads are respectively established for predicting and modeling, various forecasting components are integrated to obtain a total load variation trend, and ultimately an early warning test is conducted on the prediction model specified by a user.

The invention relates to an indirect prediction method for the residual life of a lithium ion battery. The method comprises the following steps of collecting lithium ion battery monitoring data, a normalized battery capacity sequence and an equal-time voltage difference sequence; carrying out correlation analysis to extract health factors; constructing a lithium ion battery health state estimationmodel based on an echo state network; constructing a health factor prediction model based on a long short-term memory neural network, and calculating a health factor of a future cycle period; and calculating the true value of future cycle period capacity to finish the prediction of the residual life of the lithium ion battery. The invention further provides an indirect prediction device for the residual life of the lithium ion battery. According to the predication device and method, online prediction of the residual life of the lithium ion battery can be realized, and timely replacement and maintenance are carried out before the expiration of the residual life, so that normal operation of the lithium ion battery is ensured, and the prediction precision of the residual life of the lithiumion battery is improved.

The invention discloses a cache strategy method in a D2D network based on deep reinforcement learning. The method comprises the steps of acquiring position information of each user at a next moment via an echo state network algorithm by using the historical position information of each user in the cached and enabled D2D network as input data; acquiring content request information of each user at the next moment via the echo state network algorithm according to the position information of each user at the next moment in combination with the context information of each user at a current moment;caching the content request information into a cache space of the corresponding user; and acquiring an optimal strategy for delivering the content request information between the users in the cached and enabled D2D network via a deep reinforcement learning algorithm by minimizing the transmission power of the user transmitting the content request information and minimizing the delay of the user receiving the content request information as targets. According to the method provided by the invention, the problems that in the cached and enabled D2D network, the placement hit rate of the cached content is low and the consumed energy is large and the delay is long during a cache delivery process are solved.

The embodiment of the invention provides an echo state network based prediction method and a prediction device, and relates to the field of computers. The prediction method and prediction device can effectively improve the prediction performance of the ESN (Echo State Network), and increase the speed and the accuracy of prediction. The prediction method comprises the following steps: firstly, establishing an echo state network prediction model, wherein nerve cells in a dynamic pool of the echo state network prediction model serves as wavelet nerve cells; acquiring a training set according to the input data, utilizing the training set to carry out prediction training on the echo state network prediction model, and obtaining the trained prediction model; predicting according to the trained prediction model, so as to obtain output data. The echo state network based prediction method and the prediction device disclosed by the embodiment of the invention can be applied to prediction of nonlinear chaotic time series data.

The invention relates to a parallel modular neural network-based byproduct gas real-time prediction method. According to the method, according to the principle of state space segmentation of a neural network, Fuzzy c-means (FCM) clustering is adopted to divide sample data into a plurality of categories; each category is corresponding to the subspace (namely, module) of one state space; the data are reconstructed, so that a prediction model can be established; in a modeling process, an improved echo state network is provided, a modular method is adopted to segment the state space of the neural network into a plurality of independent sub spaces, wherein each subspace is a sub network; a reserve pool sharing method is used in combination, so that the training of all networks is completed in the same reserve pool, each sub space is corresponding to an output weight matrix, and therefore, the operation rules of a system can be better simulated; a network training problem is simplified into a parallel training problem of a plurality of small networks, so that the calculation process of the model can be accelerated; and a big data sample containing more useful information is introduced, so that the prediction precision of the model can be improved; and a Map Reduce computing framework is adopted to parallelize solution problems, so that a high speed-up ratio can be obtained, and real-time prediction of the metallurgical gas system can be realized.

The invention relates to photovoltaic power prediction, discloses a method for predicting online learning photovoltaic power of leaky integral echo state networks, and aims at providing a prediction model for analyzing influences, on photovoltaic power prediction performance, of parameters of leaky integral echo state networks (LIESN) and then obtaining photovoltaic power. A least squares online learning algorithm is utilized to train the model so as to obtain an online learning leaky integral echo state network prediction model and then finally realize leaky integral echo state network-basedonline learning photovoltaic power prediction. The method for predicting online learning photovoltaic power of leaky integral echo state networks comprises the following steps of: 1, importing a leakyintegral nerve cell; 2, setting parameters; 3, carrying out training by utilizing the online learning algorithm; and 4, predicting photovoltaic output power. The method is mainly applied to the photovoltaic power prediction occasions.

The invention relates to an unmanned-ship speed and uncertainty estimation system and a design method. According to the system, an echo state network can be applied to speed estimation of an unmannedship, the echo state network is utilized to approximate model uncertainty and environment disturbance to enable the system to obtain target speed observation values, and also approximate unknown dynamics generated by the uncertainty of model parameters, non-modeling of fluid dynamics, external interference caused by wind waves and ocean currents and the like, and the state observation problem containing the model uncertainty and the unknown environment disturbance is effectively solved. Introduction of the echo state network overcomes the problems of slow convergence, proneness of falling intolocal minimums, complicated training processes and the like brought by traditional neural networks based on a learning algorithm of gradient descending. According to the system, the neural network with a low-frequency learning link is adopted to approximate system uncertainty, high-frequency oscillation which may be caused by a high-gain learning rate is effectively filtered out, and steady stateobservation on a system with unknown dynamics is realized.

The invention discloses a time sequence prediction method based on a grey wolf optimization echo state network. According to the method, the output weight of the echo state network is adjusted throughthe grey wolf algorithm so that optimization of the echo state network can be realized, the problem of training difficulty of the echo state network prediction method can be effectively solved, the constructed echo state network prediction method has better adaptability, the prediction accuracy is obviously higher than that of the present echo state network prediction method and other network prediction methods and thus the application value is enabled to be higher.

The invention relates to a method and a system for controlling constant torque of a switched reluctance motor by use of composite control current. According to the method, linear control current of each phase is obtained on the basis of linear inductance model torque distribution function control, and an echo state network is adopted and outputs non-linear current according to current total torque of the switched reluctance motor, given total torque and output non-linear current feedback as well as switched reluctance motor Jacobian information calculated by an RBF (radial basis function) neural network through parameter learning; non-linear control current of each phase and the linear control current are superposed, and the composite control current is obtained and taken as a set value for a current hysteresis loop controller. Current, torque and positon sensors of the system are connected with a signal processor, the signal processor executes the module of the method and outputs the composite control current, a power converter of the motor is controlled through the current hysteresis loop controller, and torque ripple of the switched reluctance motor is remarkably and effectively inhibited.

The invention provides a radio frequency identification device (RFID) and Bluetooth network based monitoring positioning system and method. By means of a low complexity and self-networking manner, topological structure changes of existing networks can be in fast response, and a new network topological structure can be accurately and efficiently reconstructed; by means of the RFID technology and a received signal strength indicator (RSSI) model based echo state network positioning method, defects of large measurement errors and slow real-time response existing in traditional RSSI positioning methods are overcome, and problems that existing monitoring probes lack exclusive positioning systems, are large in positioning errors and difficult in real-time positioning, and the like are solved.

The embodiment of the invention provides a prediction method and a prediction device of an echo state network (ESN) and relates to the technical field of a communication service network so that the prediction performance of the ESN can be improved effectively and the prediction speed and the prediction precision can be improved. The method is as follows: through establishment of a prediction model of the ESN, wherein a dynamic pool of the prediction model is a mixed neuron dynamic pool which includes wavelet neurons, a training set is obtained from input data and prediction training is carried out on the prediction model according to the training set and then the prediction model which completes training is obtained; and through prediction input data, the prediction model which completes training is predicted and then prediction output data in a prediction process is obtained. The prediction method and the prediction device are used for prediction of nonlinear chaotic time sequence data.

The invention discloses a channel prediction system and method for an OFDM wireless communication system. The channel prediction system comprises a standard echo state network and a two-layer adaptiveelastic network. In the method, for a subcarrier of each pilot OFDM symbol, frequency domain channel information of each subcarrier obtained by channel estimation is used to train an echo state network. The trained echo state network can realize short-term prediction of frequency domain channel information. In order to overcome the defect that the output weight in the echo state network is easy to be ill-conditioned, the output weight in the echo state network is estimated by using a two-layer adaptive elastic network. According to the method, the defect that the channel information of the traditional channel estimation is easy to expire is overcome, and single-step prediction and multi-step prediction can be realized with high precision. According to the method, the short-term predictionof the channel information of the OFDM wireless communication system can be realized, and a guarantee is provided for realizing adaptive transmission, adaptive coding and the like of wireless communication.

The invention discloses a method for detecting the weak signal of the echo state network based on the improved teaching optimization algorithm in the chaotic background. The method adopts the improved teaching optimization algorithm (ITLBO) algorithm to optimize the parameters of the echo state network model. Determine the number of optimized parameters, the value of the reserve pool size N and the sparsity SD, and encode the rest of the parameters; secondly, find the optimal echo state network model parameter combination through the teaching stage, learning stage, and feedback stage of ITLBO and compare these Modeling, training and prediction of parameters, analysis of the single-step prediction error and judging whether there is a weak target signal in the chaotic background noise, using this method to simulate the Lorenz chaotic background and the actual sea clutter signal, accurately and quickly Detect weak signals; overcome the shortcomings of difficulty in selecting parameters of the echo state network model, and improve work efficiency.

The invention discloses an ESN neural network image classification processing method based on a memristor, and relates to the technical field of image processing. The unique memory characteristics andoperational capability of the memristor are utilized, an echo state network is given, an ESN neural network circuit based on the memristor is designed to meet the requirement for the storage capability in the image processing process, memory access operation of training data is reduced, and finally the purpose of improving the performance and efficiency of the overall neural network is achieved.According to the method, data storage and operation based on the memristor are fused; image data is used as a training object; an image preprocessing function is realized by utilizing convolution operation of the image. According to the method, basic logic operations required by image preprocessing are screened out, circuit design of the memristor is performed on the basic logic operations by referring to implicit circuits, so that a data storage and operation structure based on the memristor is completed, and memory access operation of training data is reduced by combining storage and operation of image data. The application of the invention can improve the performance of the whole neural network.

The invention provides a short-term load forecasting method based on an echo state network, which comprises the steps of collecting historical load data and information of load influencing factors; preprocessing the historical load data; screening out similar days which are similar to a day to be forecasted by using a fuzzy clustering analysis method based on the information of the load influencing factors; building an echo state network load forecasting model based on the preprocessed historical load data of the similar days; and performing load forecasting on the day to be forecasted based on the echo state network load forecasting model. According to the invention, the load influencing factors are considered, the historical similar days are screened out, and the data of the historical similar days is used as training samples, so that the forecasting accuracy of the forecasting model is greatly improved; and meanwhile, by the forecasting model is trained by adopting an L1 / 2 norm regularization method, the generalization ability of the forecasting model is enhanced, and the accuracy of the forecasting result is further improved. The invention further discloses a short-term load forecasting system based on the echo state network.

The invention discloses a method for diagnosing faults of an analog circuit based on synchronous optimization of an echo state network, and relates to a method for diagnosing faults of an analog circuit. The problem of lower diagnosis precision by using the traditional neural network to diagnose the faults of the analog circuit is resolved. The method comprises the following steps of: using a unit pulse signal to excite the analog circuit to work; obtaining a response signal to be diagnosed of the circuit; collecting a unit pulse response output signal of the analog circuit; using a wavelet transform method to process the unit pulse response output signal of the analog circuit; obtaining fault characteristics as a data sample; inputting the data sample in the echo state network; using a differential evolution algorithm to perform synchronous optimization selection of parameters and characteristics; establishing a model for diagnosing the faults of the analog circuit; using the wavelettransform method to process the response signal to be diagnosed of the circuit; obtaining fault data; inputting the fault data in the model for diagnosing the faults of the analog circuit; and obtaining and outputting a fault diagnosis result. The method disclosed by the invention is applicable for diagnosing the faults of the analog circuit.

The invention provides a construction method of an electric power material demand model. The construction method comprises the steps of S1, obtaining electric power material whole-process data; s2, constructing and forming a multi-level comprehensive electric power material demand prediction model by utilizing the electric power material whole process data obtained in the step S1 and adopting a mode of combining a least square support vector machine, an echo state network and a regularization extreme learning machine, and the method comprises the following steps: S21, establishing a sample database; s22, adopting a least square support vector machine to predict the electric power material demand; s23, adopting an echo state network to predict the electric power material demand; s24, adopting a regularization extreme learning machine to predict the electric power material demand; and S25, integrating and weighing the prediction results to obtain a final prediction result of the electricpower material demand. Meanwhile, the invention provides an electric power material demand prediction system. The system comprises the electric power material demand model constructed by the method.The method effectively improves the perspectiveness of material management of a power grid company, creates favorable conditions for an enterprise to extract overall resources, guarantees the operation reliability support of a power grid, and reduces the operation cost of a power grid enterprise.

Methods and systems for providing fast semantic proposals for image and video annotation including: extracting image planes from an input image; linearizing each of the image planes to generate a one-dimensional array to extract an input feature vector per image pixel for the image planes; abstracting features for a region of interest using a modified echo state network model, wherein a reservoir increases feature dimensions per pixel location to multiple dimensions followed by feature reduction to one dimension per pixel location, wherein the echo state network model includes both spatial and temporal state factors for reservoir nodes associated with each pixel vector, and wherein the echo state network model outputs a probability image; post-processing the probability image to form a segmented binary image mask; and applying the segmented binary image mask to the input image to segment the region of interest and form a semantic proposal image.

The invention discloses a base station flow prediction method based on an echo state network. The method comprises following steps of step 1: acquiring flow data from a wired network base station, carrying out data preprocessing and generating normalized sample data; step 2: initializing states of a reservation pool through the normalized sample data and generating the reservation pool; step 3: combining the reservation pool and the normalized sample data, carrying out weight initialization, determining weight distribution intervals and training an echo state network model; and step 4: testingthe trained echo state network. The method is advantageous in that accuracy of base station flow prediction can be effectively improved; based on the uncertainty and complexity of the base station flows, by use of the echo state network, relations between base station flow data can be well excavated; and accuracy of the base station flow prediction can be improved to a certain degree.

The invention discloses a convolutional echo state network based time series classification method. An echo state network has a time series core and an echo state property, wherein the time series core refers to that the echo state network maps inputted signals into a high-dimensional space of a reserve pool, and the echo state property refers to that the network has a short-term historical information memory capacity. In the convolutional neural network, multi-scale characteristics in the echo state network can be extracted through a multi-scale convolutional layer, and multi-scale time series invariance can be kept through maximal pooling in time direction. By combination of the echo state network and the convolutional neural network, a convolutional echo state network model is provided;by the model for operations including multi-scale convolution, maximal pooling in the time direction and the like of state represent information outputted by the echo state network, advantage complementation of the echo state network and the convolutional neural network is realized, and high efficiency of an echo state network learning mode is kept while advantages of the convolutional neural network in characteristic extraction are achieved.

The invention provides a joint prediction method of base station traffic. The problem that the traditional linear algorithm is bad in prediction performance when the traffic data is nonlinear and hasa sudden change value is solved. The method comprises the following steps: firstly collecting traffic data from a base station as a data set, performing data preprocessing on an abnormal value and a missing value; decomposing processed data by adopting wavelet transform, enabling the traffic data to be smooth and easy to predict; performing single reconstruction on a sequence obtained through decomposition, wherein a low-frequency signal is predicted by adopting an echo state network model, and a high-frequency signal performs prediction by adopting an autoregression integral sliding average model; and finally performing linear accumulation on the prediction numerical value of the single sequence to obtain a final result. Compared with the single model prediction, the joint model method disclosed by the invention can reach better prediction, the reduced average absolute percentage error can achieve 6%, and the normalization root mean square error is reduced to a certain degree; the traffic data prediction accuracy of the base station is improved, and the network resource reasonable allocation can be improved.

An analog circuit fault diagnosis method based on echo state network dynamic classification relates to an analog circuit fault diagnosis method. It solves the problem of low diagnostic accuracy of analog circuit fault diagnosis using traditional neural network. Its method: use the unit pulse signal to excite the analog circuit to work, obtain the response signal of the circuit to be diagnosed; collect the unit pulse response output signal of the analog circuit, and use the unit pulse response output signal as the fault data sample; input the fault data sample to the echo Training is carried out in the state network, and an analog circuit fault diagnosis model is established according to the training results; the obtained circuit to-be-diagnosed response signal is used as fault data, and input into the analog circuit fault diagnosis model, and the fault diagnosis result is obtained and output. The invention is suitable for analog circuit fault diagnosis.

The invention belongs to the technical field of information, relates to a resampling method, a Bootstrap estimation and Bayesian estimation method and an echo state network integration theory, and specifically relates to a metallurgy enterprise gas flow interval predicting method based on Bootstrap echo state network integration. The method comprises the steps of firstly, performing re-sampling processing on the flow data of each user of a gas system to construct an effective training sample by use of the existing historical data of a metallurgy enterprise site, secondly, establishing an interval prediction model based on the echo state network integration and predicting the gas system user flow within specified time length after a current time point, and finally, estimating the influence of the uncertainty of the model and the data on the prediction result based on the Bootstrap method and the Bayesian method, respectively, thereby constructing a confidence interval and a prediction interval. The metallurgy enterprise gas flow interval predicting method based on Bootstrap echo state network integration can be widely applied to other energy medium systems of the metallurgy enterprises.

The invention relates to an analog circuit fault diagnosis method based on a differential evolution algorithm and static classification of an echo state network, solving the problem of lower diagnosis precision in the methods for diagnosing analog circuit faults by adopting the traditional neural networks. The method comprises the following steps: adopting unit pulse signals to excite an analog circuit to work to obtain circuit-to-be-diagnosed response signals and acquiring unit pulse response output signals of the analog circuit; adopting a method of wavelet transform to process the acquiredunit pulse response output signals of the analog circuit, taking the obtained fault features as the data samples, inputting the fault features into the echo state network, adopting a differential evolution algorithm to train the fault features and building an analog circuit fault diagnosis model; and adopting the method of wavelet transform to process the circuit-to-be-diagnosed response signals to obtain fault data and inputting the fault data into the analog circuit fault diagnosis model to obtain and output the fault diagnosis results. The method is suitable for fault diagnosis of the analog circuit.

The embodiment of the invention provides a predication method and a device based on an echo state network (ESN), which relates to the communication field. Through generating a plurality of small-world and non-scale dynamic pool groups, the type and the topology structure of an original ESN dynamic pool structure are changed, and better predication effects can be generated for a nonlinear chaotic time system. The method comprises steps: a second dynamic pool is built, wherein the second dynamic pool is composed of N small-world dynamic pools and Y non-scale dynamic pools according to a complex network theory, and N and Y are larger than 0; a first dynamic pool replaces the second dynamic pool, and a lateral boundary suppression mechanism is used for associating the N small-world dynamic pools in the second dynamic pool with the N non-scale dynamic pools; a training set with a specified length is inputted to update the second dynamic pool, and a connection matrix Wout after updating is further obtained; and according to the Wout after updating, the following formula is used for predication: y(n+1)=f<out>(Wout(u(n+1), x(n+1), y(n))).

The invention discloses a medium and long term runoff prediction method based on secondary decomposition and an echo state network (ESN), and belongs to the technical field of runoff prediction. The medium and long term runoff prediction method comprises the following steps: 1, acquiring a runoff sequence x(t), and dividing the runoff sequence x(t) into a training sample and a test sample according to the data condition of the runoff sequence; 2, decomposing the runoff sequence into a plurality of intrinsic mode functions (IMF) and a trend term (Res) by using adaptive noise complete empiricalmode decomposition (CEEMDAN); 3, performing secondary decomposition on the IMF component with the highest frequency by using a variational mode decomposition (VMD) method to obtain a plurality of variational modes (Mode); and 4, respectively inputting the sub-sequences decomposed twice into the ESN for prediction, and reconstructing each prediction result to obtain a final prediction value. According to the medium and long term runoff prediction method, the problem that the prediction error of the high-frequency component in primary decomposition is large is solved, and the prediction precision of the ESN is further improved.

The invention discloses an uncalibrated visual servoing control method for estimating an image Jacobian matrix based on an echo state network facing mold protection, which comprises the following steps: 1) an image index region and a target image sample library of the mold are built; 2) feature extraction and dimensionality reduction are carried out; 3) a polynomial interpolation method is adopted for inverse kinematics planning in a space constraint condition; and 4) a pseudo inverse Jacobian matrix based on the echo state network is realized. The problem of small local part of the traditional BP neural network can be effectively solved, and by using characteristics of adaptivity and high computation efficiency of the ESN (echo state network) in the case of a constant weight, and the problem of poor instantaneity as a general dynamic network computes a network output weight online can be solved.

The invention relates to a polyester spinning process control method based on a local plasticity echo state network. The method comprises steps that the polyester spinning process parameter data at the t+1 time in the production process is collected and used as the input u(t+1) of the network; the local plasticity echo state network input layer is used to realize the input of the polyester spinning process parameter data u(t+1) at the t+1 time, and the predicted value of the next moment is calculated through the reserve pool state equation and the output layer state equation of the local plasticity echo state network; and according to the predicted value, the polyester spinning process parameters are adjusted. The local plasticity echo state network refers to a plasticity echo state network in which different neurons in the reserve pool are locally optimized through different plasticity rules. The method of the invention can further improve the prediction precision of the production process parameters, so that the prediction result is enabled to better guide the polyester fiber spinning process, and finally the output performance and quality of the precursor fiber can be improved.