91results about How to "Optimize weight" patented technology

The invention discloses a random forest-based non-invasive home appliance identification method, which aims to monitor a family internal power utilization condition without invading an interior of a family user, and is high in identification speed and high in accuracy. According to the adopted technical scheme, the method comprises the steps of establishing a load feature database; taking load features stored in the feature database as an original training set, and generating N training sub-sets in the original training set; combining N generated decision trees into a random forest; finishing a training process of the random forest through optimization of weights of different decision tree leaf nodes; detecting an electric appliance switching event by utilizing a secondary detection algorithm to obtain starting and ending time of event occurrence, and separating out current and voltage signals of a switched electric appliance from a bus signal, thereby obtaining the load features from separated data; and finally inputting the load features as input parameters to the trained random forest, and finishing electric appliance identification through voting.

The invention designs a blind channel balancing method based on an improved PSO (Particle Swarm Optimization) BP (Back Propagation) neural network. In the process of solving the blind balancing problem on the basis of a BP neural network, determination of an initial weight and a threshold of the BP neural network is lack of the theoretical basis and has the defects of low convergence speed, easiness for falling into a local minimal value and the like so as to cause a poor channel blind balancing effect. In order to overcome the defects of the BP neural network and improving the channel blind balancing effect, the invention discloses a blink balancing method based on the improved PSO-BP neural network. According to the method, firstly, defects of a basic particle swarm algorithm are overcome, parameters of the basic particle swarm are improved, and an inertia weight and a learning factor are adaptively regulated; secondly, the initial weight and the threshold of the neural network are optimized by utilizing the advantage of high global searching capacity of the improved particle swarm, and then more accurate searching is carried out in such local space by utilizing a BP algorithm soas to obtain an optimal connection weight and threshold of the neural network; and finally, blind balancing based on the the improved PSO-BP neural network is implemented.

The invention relates to a simultaneous localization and mapping method based on distributed edge unscented particle filter. First, a coordinate system is built and an environmental map is initialized; then subfilters are built for each landmark point with successful matching respectively; next, based on a robot motion model, a particle swarm is generated in each subfilter respectively, and the state vector and the variance of each particle are obtained; noise is introduced, particle state vectors after extension are calculated by utilization of unscented transformation, the particles after extension are updated and the particle swarms are optimized; then particle weights are calculated and normalization is carried out, and aggregated data of each subfilter are subjected to statistics and the data are sent to a master filter; next, global estimation and variance are calculated; then the effective sampling draw scale and sampling threshold of each subfiter are determined, the subfilters with severe particle degeneracy are subjected to resampling; then the state vectors and the variances of the robot are output, and stored in a map. Finally, landmark point states are updated by utilization of kalman filtering algorithm until the robot is no longer running.

The invention discloses a wind power prediction method based on a modified particle swarm optimization BP neural network. The method includes the following steps: 1. encoding weight values and threshold values of a BP neural network as particles, and initializing the particles; 2. computing each particle fitness value with the difference between the result obtained from BP neural network training and an anticipated value as a fitness function; 3. comparing the fitness value of each particle and individual optimal particle to obtain a global optimal particle; 4. updating the speed and position of the particle; 5. determining whether the global particle meets termination conditions, if the global particle meets termination conditions, terminating the computing and outputting an optimal weight threshold value, and if the global particle does not meet termination conditions, back to step 2 and carrying out iterative operation; and 6. Using the optimal weight threshold value that is acquired by step 5 to connect an input layer, a hidden layer and an output layer of the BP neural network, and obtaining the result of wind power prediction on the basis of the result of the BP neural network. The method has fast convergence speed, high precision, and is not easily trapped to local extremum.

The invention discloses a meta-learning algorithm based on stepwise gradient correction of a meta-learner, and the algorithm comprises the steps: firstly, obtaining training data with noise marks anda small amount of clean unbiased metadata sets; establishing a meta-learner, namely a teacher network, on the metadata set relative to a classifier, namely a student network established on the training data set; and carrying out united updating of student network parameters and teacher network parameters by using random gradient descent; obtaining a student network parameter gradient update function through a student network gradient descent format; feeding the network parameters back to the teacher network, and updating the teacher network parameters by using metadata to obtain a corrected student network parameter gradient format; and then updating the student network parameters by using the correction format. Accordingly, the student network parameters can achieve better learning in thecorrection direction, and the over-fitting problem of noise marks is weakened. The method has the characteristics of easiness in understanding, realization, interpretability and the like of a user, and can be robustly suitable for an actual data scene containing noise marks.

An image semantics classification method based on class-shared multiple kernel learning (MKL), which relates to the artificial intelligence field, is disclosed. The method is characterized by: a pretreatment stage: extracting a bottom layer characteristic of an image and calculating a multiple kernel matrix; a modeling stage: constructing a class-shared multiple kernel classifier model; a parameter learning stage: optimizing classifier parameters of multiple classes, basic kernel function weights and kernel function weights which are related to the classes in an uniform frame; an image classification stage: using the classifier with a good learning ability to carry out image classification to a sample to be classified. In the invention, on one hand, through sharing a group of basic kernelfunction weights, common implicit knowledge of each class in a kernel function space can be excavated; on the other hand, characteristics of the each class in the kernel function space can be considered for the different classes which possess class-related kernel function weights. According to a degree of training data, a kernel classification method is provided for the kernel function combination to achieve mutual independence, partial sharing or complete sharing in the classes.

The invention discloses a spinning quality prediction method based on a firework algorithm improved BP neural network. According to the method, a firework algorithm is introduced into a BP neural network, and an optimizing mechanism of the firework algorithm is utilized to optimize a network weight and a threshold value of a BP neural network model; input and output indicators are selected, and an FWA-BP-based spinning quality prediction model is constructed; a standardized data set is utilized to perform learning and training on the FWA-BP-based spinning quality prediction model established in the step 2; and finally prediction of spinning quality is completed. Through the method, the problem that spinning quality is difficult to predict precisely due to numerous yarn quality influence factors in a spinning system and mutual coupling is solved, a function mapping relation between fiber indicators and resultant yarn quality can be effectively established, prediction of yarn quality in spinning production is realized, and the method is beneficial for increasing the quality management level of a spinning workshop.

The invention relates to the field of information retrieval and provides a processing method for retrieving sentences by a user. The method comprises the following steps: establishing a sample database which is related to a word which is retrieved by the user and a resource database which is related to the word which is retrieved by the user; performing feature extraction on the word which is retrieved by the user; classifying the word, which is retrieved by the user, by virtue of a classifier and performing basic empowerment on the word which is retrieved by the user; after the basic empowerment, performing entity power call on the word which is retrieved by the user; and outputting the weight of the word which is retrieved by the user. The invention also provides a processing system for retrieving sentences by the user. By adopting the technical scheme of the invention, the accuracy of entity extraction is ensured, the dynamic weight is obtained and the problem that the weight is constant and unreasonable if only words which are counted offline are inquired is solved. Finally, the weight of the word which is retrieved by the user is further optimized by subordination identification, the weight of the core word which is retrieved by the user is highlighted and a practical and reasonable information support is provided for search engines.

The invention discloses a remote cooperative diagnosis task allocation method. Aiming at complex diagnosis tasks, an RCFD (Remote Collaboration Fault Diagnostic) center is used for decomposing the complex diagnosis tasks into a plurality of executable diagnosis tasks and allocating the executable diagnosis tasks for allowing each diagnosis resource participating in the cooperative diagnosis to execute; and each executable diagnosis task can also carry out diagnosis task allocation on the basis of a method for expanding a contract net. The remote cooperative diagnosis task allocation method disclosed by the invention comprises three links, namely diagnosis task model establishment, diagnosis task path planning and diagnosis resource configuration and has the advantages of modularity, favorable expansibility, wide application occasions and the like, wherein in the diagnosis task model establishment link, an intersection set of different types of models is paid attention to from different decomposition granularities and weights of the models are optimized by applying a Bayesian network method; in the diagnosis task path planning link, a unified key path planning algorithm is established on the basis of a D algorithm; and in the diagnosis resource configuration link, a configuration algorithm is established by fusing multi-constraint indicators from a view of service.

The invention provides a neural network short-term power load prediction method based on a squirrel and weed hybrid algorithm. The method comprises: forming a sample data set by historical power loads, meteorological factors and date types before the day to be predicted, conducting principal component analysis on meteorological factor data through SPSS software factor analysis, and extracting principal components to replace original meteorological factor variables to form a new sample data set; taking the normalized historical power load data as an output sample, and taking meteorological factors and date types as input samples; optimizing the weight and threshold of the BP neural network by applying a squirrel and weed hybrid algorithm to construct an SSIWO-BP neural network prediction model; and inputting the date type to be predicted and the meteorological factor data into the SSIWO-BP neural network prediction model to predict the power load value. According to the method, the global convergence of the squirrel weed hybrid algorithm and the stability in a high-dimensional space are considered, BP neural network parameters are optimized, the generalization ability of the neuralnetwork is enhanced, and the prediction precision of the model is improved.

The invention provides a PSO neural network-based engineering ceramic electrospark machining effect prediction method. The method comprises the steps of building a three-layer BP neural network prediction model of an insulated engineering ceramic line electrode discharge grinding process effect by utilizing a BP neural network; in combination with an adaptive variation particle swarm algorithm, optimizing connection weight and threshold of the BP neural network; building a particle swarm neural network-based engineering ceramic line electrode discharge grinding process effect prediction modelby utilizing Matlab programming; and verifying the reliability of the engineering ceramic line electrode discharge grinding process effect prediction model built by optimizing the BP neural network bythe adaptive position variation particle swarm algorithm. The method remarkably reduces an iterative frequency, has relatively high prediction precision, reliability and validity, has a certain practical value, and can be used for optimizing engineering ceramic line electrode discharge grinding process electric parameters to further improve surface quality of workpieces.

The invention discloses a photovoltaic power generation power short term prediction method based on a mind evolution Elman neural network. According to the good expandability, the good portability and the extremely strong global optimization capacity of the mind evolution algorithm, and the historical data state sensitivity and the strong self dynamic information processing capacity of the Elman neural network, a photovoltaic power generation power short term prediction algorithm by using the mind evolution algorithm to optimize the Elman neural network is brought forward. Through optimizing an Elman neural network weight and a threshold by the mind evolution algorithm, defects that the Elman neural network is likely to fall into local optimum and the like are overcome. The example result shows that the method is quick in convergence rate, strong in optimization capacity and convenient in dynamic information processing, and an important role is played in photovoltaic power short term prediction.

The invention discloses a prediction method for shearing parameters of Martian soil based on a GA-BP neural network, and the method comprises the steps: S1, determining a BP network weight and threshold, and using an improved genetic algorithm (GA) to optimize the BP neural network weight and thresholds: firstly coding the BP neural network weight and threshold, carrying out the selection, crossing and mutation operations of the weight and threshold according to a fitness function to obtain a new weight and a threshold until a training error reaches 0.001, otherwise continuing the optimizationto obtain the weight and threshold; S2, establishing a BP prediction model; S3, training the BP model to predict shearing parameters of Martian soil. The method improves a selection operator of the GA, maintains the diversity of populations while effectively selecting a better individual in the populations, thereby optimizing the weight and threshold of the BP neural network model and enabling the BP network to accurately estimate the shearing parameters of the Martian soil.

The invention discloses a data mining based hybrid filtering film recommendation method. The data mining based hybrid filtering film recommendation method comprises the steps that firstly, a film attribute matrix M and a preliminary user score matrix G are established; then film key words daily browsed by a user are mined, then user clustering is performed by utilizing an ant colony method, user score which is empty in the preliminary user score matrix G is simulated, the sparsity problem of the user score matrix G is solved, and a perfect user score matrix G is obtained; a user interesting matrix H is obtained according to the perfect user score matrix G and the film attribute matrix M; finally, the user interesting matrix H is combined with the film attribute matrix M to obtain a recommended film score matrix Q. The clustering effect is improved by utilizing mined data, meanwhile the accuracy and stability of the ant colony method are improved, and the weight of a user interesting vector in the user interesting migration problem is optimized.

The invention relates to a foundation pit displacement prediction method based on a particle swarm optimization BP neural network, and the method comprises the following steps: 1) selecting a first displacement monitoring point of a slope top of a foundation pit project and a plurality of adjacent monitoring points, and obtaining deformation monitoring data of the monitoring points in a past firsttime period as sample data; 2) performing preprocessing including data normalization on the sample data to obtain a training data set; 3) constructing a BP neural network; 4) optimizing the weight and threshold of the BP neural network by using a particle swarm algorithm, and outputting the optimal weight and threshold; 5) assigning the obtained optimal weight and threshold to a BP neural network, and then performing network training by adopting the training data set to obtain a trained BP neural network; and 6) predicting the displacement deformation of the first displacement monitoring point in the second time period by adopting the trained BP neural network. The method is beneficial to improving the accuracy and stability of foundation pit displacement prediction.

The invention provides a comprehensive monitoring and diagnosis system and method of state information of a mobile transformer. The comprehensive monitoring and diagnosis system of state information of a mobile transformer includes an intelligent wireless sensor which is used for collecting quantity of state of the transformer, an intelligent data exchange terminal which can obtain the data of theintelligent sensor in real time, a comprehensive monitoring system which selectively and deeply analyzes the reasons for equipment anomalies and searches equipment failure points and uploads the dataand the diagnostic result to a cloud platform expert diagnosis system, a cloud platform diagnosis service system which utilizes the intelligent data exchange terminal to movingly and comprehensivelymonitor the data uploaded by the equipment and comprehensively analyze operation state of the equipment, and evaluates the health of the transformer by combining failure case library so as to make anoverhauling strategy, and a background management system which logs in according to different user rights and can use different functions according to the rights, wherein the main function of the system is responsible for data management, data statistics, data analysis, and classified storage of monitoring data. The comprehensive monitoring and diagnosis system of state information of a mobile transformer can realize interaction between operation and maintenance staff and equipment management, and has practical significance for improving the monitoring efficiency of a substation and safe operation of the equipment.

The invention discloses a server fault monitoring method and system based on a neural network. The invention belongs to the field of equipment fault diagnosis analysis. The technical problem to be solved by the invention is how to realize server fault diagnosis based on a neural network and monitor a server state. According to the technical scheme, the method comprises the steps that server information is obtained through a BMC, whether a server breaks down or not is analyzed and predicted through a neural network, the fault is fed back to a webpage to be displayed, meanwhile, the state of theserver is monitored, and therefore the stability of the server is improved. The system comprises a data acquisition module, a data processing module, a fault prediction module based on a neural network and a fault early warning module. The fault prediction module based on the neural network applies the neural network which completes training and determines parameters of each node to a BMC system,the preprocessed data is used as input of the neural network, and output of the neural network is a predicted server fault type.

The invention brings forward a general airplane air-material demand prediction method based on an MPSO-BP network with the combination of a particle swarm algorithm and a BP network. The method includes: firstly, analysis and research of main influence factors of the general airplane air-material demand are conducted, then the basic principle of the BP network, the PSO algorithm, and the improved strategy of the PSO algorithm are introduced, and a prediction model for improving the particle swarm algorithm and optimizing the BP network is built. Five main influence factors of the general airplane air-material demand are regarded as the input of the network, the demand quantity of the general airplane air-material is regarded as the output, and a nonlinear mapping relation between the input and the output is established. According to the method, accurate prediction of the general airplane air-material demand quantity can be realized, the non-linear fitting capability and generalization capability are very good, the convergence efficiency is improved, the possibility of falling to a local minimum value is reduced, the prediction precision is higher, and the application effect in the general airplane air-material demand prediction is good.

The invention discloses a method of analyzing cotton content of a textile based on an improved DBN near infrared spectrum. The method comprises the following steps: (1) collecting a sample and gathering the near infrared spectrum and the physical and chemical data of the near infrared spectrum; (2) pre-processing the acquired spectrum data to acquire pre-processed data; (3) establishing a quantitative analysis model by employing an improved deep brief network (DBN), and adjusting the network by means of a global learning algorithm to obtain the optimum textile cotton content quantitative analysis model; and (3) introducing the predicted data into the optimum textile cotton content quantitative analysis model for predicted analysis of the cotton content of the textile. The method disclosed by the invention can quickly determine the cotton content of the sample without loss, and applies the deep neural network into the field of detecting the cotton content of the near infrared spectrum textile, so that the detecting accuracy and reliability are improved.

The invention provides a coal component analysis method based on coal spectroscopic data. The method comprises the following steps: collecting coal spectroscopic data; performing coal component prediction by using a coal component analysis model, and ensuring that the input of the model adopts the collected spectroscopic data and the output is a coal component. The method establishes the coal component analysis model through the spectroscopic data and coal industrial analytic determination results, obtains spectroscopic characteristic data through extraction of a convolutional neural network,outputs coal components corresponding to the coal spectroscopic data determined by coal industrial analysis through a spectroscopic characteristic data extreme learning machine, and adopts a weight and deviation value of an artificial bees colony optimized extreme learning machine during the predication process, so as to obtain an optimized coal component analysis model. The coal component analysis model is integrated with a spectrum technology and is applied to the coal industrial analysis field, and a novel, rapid and accurate coal component analysis method is provided.

The invention discloses a license plate recognition system and method based on a particle swarm optimization neural network. The license plate recognition system comprises a license plate positioningmodule, a character recognition preprocessing module and a license plate recognition module, wherein the license plate positioning module comprises a license plate character preliminary positioning unit, a real license plate acquisition unit, a license plate rivet removal unit and a character axis determination unit; the character recognition preprocessing module comprises a binarization unit, a tilt correction unit and a normalization processing unit; and the license plate recognition module comprises a particle swarm learning unit, a dynamic optimization neural network unit and a recognitionunit. According to the license plate recognition system and method based on the particle swarm optimization neural network disclosed by the invention, the weight of the neural network and the numberof hidden layer neurons are dynamically optimized by the particle swarm optimization neural network, thereby increasing the accuracy and speed of license plate recognition.

The invention provides a communication signal modulation identification method of a quantum elephant population mechanism evolution BP neural network. A weighted Myriad filter is designed in combination with a data set of designed characteristic parameters, and then the BP neural network is evolved by using the quantum elephant population mechanism to obtain optimal system parameters of the neuralnetwork. And the BP neural network with the optimal weight and threshold is used as a classifier to efficiently identify the communication signal modulation mode under an impact noise background. Thedesigned method can acquire optimal network parameters and classification and recognition effects in an impact noise environment, so that a relatively high recognition rate is obtained in severe environments like impact noise and a low mixed signal-to-noise ratio, and the application limit of existing neural network modulation and recognition is broken through.

The invention relates to a short-term load forecasting method and system based on feature selection of DFS and SVM. The method comprises the steps of obtaining a feature subset according to the F-score value influencing the feature of the short-term load forecasting, determineing the optimal feature subset according to the classification accuracy of the SVM classification model corresponding to the feature subset, training the short-term load forecasting model by using the historical data corresponding to the feature of the optimal feature subset, and forecasting the daily load value by usingthe short-term load forecasting model. The method of the invention realizes the effective dimensionality reduction of the input data, comprehensively considers the global searching ability and the local searching ability, realizes the optimization of the weight value and the threshold value of the neural network, avoids falling into the local optimum, and improves the precision of the load forecasting.

The invention provides a sewage treatment effluent parameter prediction method based on a grey neural network composition model, the method comprises the steps: a historical influent parameter sequence x (t) is acquired and is accumulated to form an accumulation sequence y(t); according to the accumulation sequence y(t), a grey neural network is established and is utilized to predict historical effluent parameters, and in the prediction process, an improved particle swarm optimization is adopted to optimize the grey neural network; according to the optimized neural network, present effluent parameter prediction is performed; a Markov chain is adopted to correct obtained present effluent parameters.

The invention provides a connection structure of a straddle-type monorail vehicle body and bogies and a straddle-type monorail vehicle. The connection structure of the straddle-type monorail vehicle body and the bogies comprises a first vehicle body, a bogie assembly and a second vehicle body which are sequentially arranged, wherein the rear end of a first vehicle body hinge part is arranged at the rear end of the first vehicle body, and a second vehicle body hinge part and a supporting part are arranged at the front end of the second vehicle body; the first vehicle body hinge part comprises afirst vehicle body upper heart disc assembly, and the second vehicle body hinge part comprises a second vehicle body lower heart disc assembly and a traction upper heart disc assembly; and a center traction pin of the bogie assembly is provided with a traction lower heart disc assembly. The straddle-type monorail vehicle applies the connection structure of the straddle-type monorail vehicle bodyand the bogies to the body layout connection. The number of bogies is drastically reduced, and the problem of large dead weight existing in a connection structure of bogies in the prior Art is solved.

The invention discloses an urban rainstorm disaster risk assessment method and system based on a GA optimized BP neural network, relates to the technical field of urban rainstorm disaster risk assessment, and aims to solve the problems that an assessment mechanism is not visual enough, the reliability of an assessment result is insufficient, and real-time dynamic risk assessment is lacked in an assessment method adopted at present. According to the technical scheme, the method includes: establishing a rainstorm disaster risk assessment system including disaster-inducing factor dangerousness, disaster-pregnancy environment sensitivity, disaster-bearing body vulnerability and disaster prevention and disaster resistance; generating a risk level label based on k-means clustering historical disaster damage data; according to the rainstorm disaster risk assessment system and the risk level label, constructing a GA optimization neural network rainstorm disaster risk assessment model; and inputting the real-time rainfall into the rainstorm disaster risk assessment model to obtain a risk level label in a specific time period. According to the method and system, the evaluation comprehensiveness and accuracy are improved.

The invention relates to a reclaimed water irrigation underground water pollution risk assessment method comprising the following steps: 1, gathering the water quality and water level of each monitoring point, and gathering positioning data and attribute data of land utilization types data; 2, building a DRASTIC model according to underground water depth, net supplement amount, aquifer medium, soil type and aeration zone medium data; scoring the underground water depth, the rainfall infiltration supplement, the aquifer medium, the soil type and the aeration zone medium influences, drafting a score graph, superposing and calculating and evaluating the underground water inherent antifouling property; 3, building a hierarchy analysis method according to the characteristic pollutant toxicity,mobility and degradability, and evaluating the underground water pollution load risks; 4, calculating the underground water values according to the aquifer characteristics, water volume, water quality, ecology and service functions; 5, calculating the reclaimed water irrigation underground water pollution risks according to the underground water inherent antifouling property, the underground waterpollution load risks, and the underground water values.

The invention provides a comprehensive geological borehole logging lithology identification method, which comprises the following steps of: refining borehole logging data to obtain a refined data set,including missing value filling, equalization processing and data set normalization processing on the borehole logging data; enabling refined data set to be subjected to dimensionality reduction processing according to a tSNE algorithm, improving and optimizing a BP neural network according to a PSO algorithm, obtaining the optimal initialization weight and threshold value of the network, establishing a network model, and carrying out training learning on the dimensionality reduction data set through the established network model.According to the method, drilling and logging data is refined,the problem that the final recognition rate is too low due to the fact that acquired drilling and logging data are missing, data sets are unbalanced and training data are not in a unified dimension range is solved, dimension reduction processing is conducted on the drilling and logging data sets according to the tSNE algorithm, data are simplified accordingly, a common BP neural network in the prior art is optimized through the PSO algorithm, and the identification accuracy and the identification rate are improved.

The invention discloses a spiral inductance optimization method based on HPSO algorithm and GA-BP algorithm. The spiral inductance optimization method comprises the steps that an on-chip spiral inductance data set is manufactured through electromagnetic simulation software HFSS, and data preprocessing is carried out; a genetic algorithm (GA) is adopted to optimize the BP neural network, then the optimized BP neural network is trained, and a GA-BP model is established; layout search is performed on the on-chip spiral inductor by adopting HPSO to obtain structure parameters meeting constraint conditions under a specific target inductance value; characterization parameter prediction is carried out on the structure parameters of the on-chip spiral inductor under the specific inductance value by using the GA-BP model. The method can help designers compromise and analyze the inductive performance and the inductive size, enables the circuit to have the optimal performance and the optimal size, not only guarantees the accuracy of the calculation result, but also can save time and cost, and has a good application prospect.