34results about How to "Solve the generalization problem" patented technology

The invention discloses an urban water-logging analysis method and an urban water-logging analysis system in order to solve the problem in the existing water-logging analysis method that inaccurate results are caused as there is no accurate data of underground drainage facilities and a lot of variables in the model are simulated data. The method comprises the following steps: generating an urban space model containing all depressions according to urban space information; getting the water depth of each depression according to the acquired predicted total rainfall within a period of time T, the area of the depression and the underground pipeline information; judging whether the water depth is greater than a set threshold; and generating alarm information characterizing depression water-logging if the water depth is greater than the set threshold. According to the technical scheme, a rainfall model, a confluence model and an improved drainage model form a water model, the generalization problem of urban drainage system facilities in the traditional rainstorm water-logging mathematical model is solved, and the accuracy of water depth prediction is improved.

The invention discloses a photovoltaic generating capacity prediction method based on an RBF (Radial Basis Function) neural network. The photovoltaic generating capacity prediction method based on anRBF neural network includes the steps: constructing a training sample according to photovoltaic generating capacity and historical data of influence factors to be selected; based on the constructed training sample, selecting influence factors of photovoltaic generating capacity by using an improved genetic algorithm, and training the RBF neural network to obtain influence factors of photovoltaic generating capacity and a trained RBF neural network; and inputting the data of a day to be predicted of the influence factors of photovoltaic generating capacity into the trained RBF neural network toobtain a predicted value of the photovoltaic generating capacity. The photovoltaic generating capacity prediction method based on an RBF neural network can preferably solve the generalization problemof the RBF neural network, and can improve accuracy of the photovoltaic generating capacity prediction result.

The invention discloses a fouling experimental device, provides a particulate fouling resistance prediction method for an arc-tube heat exchanger, and further discloses a corresponding prediction system by utilizing the object-oriented high-level language Delphi. The prediction method comprises establishment of the fouling experimental device, determination of experimental tubular products, determination of geometric dimensions, installation of a measurement and control unit, measurement and processing of parameters, establishment of prediction models, optimization of important parameters, and establishment and application of judgment models and the prediction system. According to the particulate fouling experimental device, the prediction method and the prediction system for the arc-tube heat exchanger, the defect of a local minimum of a neural network and other conventional methods is overcome, the phenomena of under-learning and over-learning are effectively restrained, the problem of generalization in the machine learning theory is solved, the calculated amount is small, the model optimization speed is high, and online monitoring of fouling resistance of a heat-exchange device can be achieved. The most prominent advantage is that small samples can be used for training models. Due to the fact that the fouling characteristics of the heat exchanger are predicted by the models in terms of temperature, flow speed and other parameters which are easy to measure, a lot of manpower and material resources are saved, and a new method is provided for designing a cooling water system under a known water quality condition afterwards and predicting the fouling characteristics.

The invention discloses a construction machinery hidden danger detection method for a power transmission line. The construction machinery hidden danger detection method comprises the following steps:acquiring images in a power transmission line channel and around the power transmission line channel in real time; making a data set of large construction machinery existing in a power transmission line channel, and randomly distributing according to a ratio of a training set to a test set of 4: 1; performing image preprocessing on the training set; processing the training set by adopting a multi-sample image synthesis method to obtain new training set sample data; training the new training set sample data by using a Faster R-CNN + FPN model to obtain a detection model of the hidden danger ofthe power transmission line channel; carrying out image target detection on the test set, updating detection model parameters, and carrying out secondary verification on the test set; and detecting the image acquired in real time by using the updated detection model, and detecting whether a large construction machinery hidden danger exists in the power transmission line channel or not. According to the construction machinery hidden danger detection method, the generalization problem of the detection model is solved, and the false alarm rate and the missing report rate are reduced.

The invention provides an antimony ore grade soft-measurement method based on the selective fusion of a heterogeneous classifier. The method comprises the step of together forming a feature space based on the pretreatment of antimony flotation froth image feature data and production data related to the grade of the antimony ore. According to the method, firstly, some feathers are randomly selected to form a plurality of sub-sample spaces. Secondly, a plurality of different sub-samples in each sub-sample space are sampled through the bootstrap sampling process. At the same time, the PCA analysis is conducted on each sub-sample to obtain key features that are high in sensitivity to grade change and free of / weak in dependency. Thirdly, two KELMs are conducted respectively for each sub-sample set to construct a candidate sub-model, based on an RBF kernal of better learning ability and a polynomial kernel type KELM of better generalization ability. Fourthly, each candidate sub-model is endowed with a weight based on the method of information entropy. Finally, all candidate sub-models are sorted from small to large based on the RMSE, and then an optimal weighted sub-model combination is selected as a final model for the prediction on the grade of the antimony ore.

The invention discloses a building energy consumption prediction method based on an RBF neural network. The method comprises the steps of: selecting historical data of to-be-selected influencing factors of building energy consumption to generate an input vector and the history data of the corresponding building energy consumption value as the output, to obtain training samples; selecting the influencing factors of the building energy consumption by using the improved genetic algorithm; selecting the data of the to-be-predicted date of the building energy consumption influencing factors and inputting into the RBF neural network, to obtain the predicted value of the building energy consumption. The problem of generalization of the RBF neural network is solved; the building energy consumptionprediction is realized by using the RBF neural network based on an L-GEM, and the accuracy of the neural network prediction result is improved.

The invention discloses a test method and a system for IMU automatic calibration. The method comprises the following steps of: generating a calibration instruction and a test instruction according toa setting of a user interface, wherein the calibration instruction controls the working state of a turntable; collecting specified data in the test instruction according to the working state; callinga control algorithm to calculate a calibration parameter and a calibration error according to the specified data; and performing error compensation according to the calibration parameter and the calibration error, and sending the updated calibration parameter to the IMU and simultaneously outputting the test result. The invention also discloses a system which belongs to the same concept as the method, and the calibration intermediate process does not need people to participate, the universality is good, the automation level is high, so that the calibration and the testing have high efficiencyand high precision, and meanwhile, and the data integration management, the online control and the monitoring are achieved.

A belt conveyer with deep semi-circular channel and big inclination up to 40 deg is composed of a machine frame, a drum group, a group of supporting rollers including central supporting rollers and side supporting rollers for rolling the conveying belt to become a deep semi-circular channel, a conveying patterned rubber belt, a carriage-type tension unit for the pull drum, and cleaning unit with brush and vibration cleaner.

The invention discloses a speech emotion recognition method and system based on semi-supervised adversarial variation self-encoding, and the method comprises the steps: S1, constructing a generative adversarial network, and constructing a speech emotion recognition model through the combination of a semi-supervised variation self-encoding model and the generative adversarial network, wherein data with emotion labels in input data and corresponding emotion labels are used as input, data without emotion labels in the input data are used as emotion label attribute missing types for processing, feature probability distribution of the input data in a hidden layer is learned through the generative adversarial network, and an SSAVAE model is constructed; S2, training the constructed SSAVAE model by using a training set; and S3, inputting to-be-processed speech emotion data, and inputting the to-be-processed speech emotion data into the trained SSAVAE model to obtain an emotion recognition result. The method has the advantages of being simple in implementation method, high in recognition precision, good in generalization ability and data disturbance resistance and the like.

The embodiment of the invention belongs to the technical field of voice processing in artificial intelligence, and relates to an address recognition method and device based on named entity recognition, computer equipment and a storage medium. In addition, the invention also relates to a blockchain technology, and the question and answer audio data of the user can be stored in the blockchain. According to the named entity identification-based address identification method provided by the invention, the extension text vector is combined with the following word group feature information of the token and the feature information of the token, so that the extension text vector can solve the generalization ability of entity extraction of a model in a suffix of a specific range, and fitting of a large amount of data is not needed; therefore, the model training cost is reduced, and the model recognition capability is improved.

The invention belongs to the technical field of machinery, and discloses a numerical control machine tool cutter remaining service life prediction method and system and application. The numerical control machine tool cutter remaining service life prediction method comprises the steps of collecting controller signals and sensor signals in the working process of a numerical control machine tool, and carrying out the preprocessing, feature extraction and feature selection of the signals; excavating information related to cutter wear in various signals; and establishing a cutter remaining service life prediction model by using a long-short term memory network and an attention mechanism to realize the remaining service life prediction of the numerical control machine tool cutter. The controller signal and the sensor signal in the working process of the numerical control machine tool are collected, the cutter remaining service life prediction model is established through multi-source information, the cutter abrasion condition reflected by different types of signals is fully considered, the limitation of establishing the prediction model through a single signal in the prior art is effectively overcome, and therefore, the generalization ability of the cutter remaining service life prediction model is improved.

The invention discloses a dynamic modeling method for the combustion process of a circulating fluidized bed boiler. The dynamic modeling method comprises the following steps: the operation parametersof the combustion process of the boiler, which mainly affect the thermal efficiency of the boiler and the emission concentration of nitrogen oxides, are adjusted and recorded as input data and outputdata; Firstly, the input weights and the hidden layer thresholds of the sample incremental quantum neural network are determined according to the quantum computation rules. Then, based on the input data and the output data, the output layer weights and the weight matrix between the input layer and the output layer are calculated, i.e., the initialization model of the boiler thermal efficiency andNOx emission concentration is established. Based on the initialization model, the boiler operation parameters are collected on line, and the sample increment is calculated. The model parameters of thesample increment quantum neural network are updated in real time, including input weights and hidden layer thresholds, output weights and weights between input layer and output layer. Thus, the on-line models of thermal efficiency and NOx emission concentration are established, and the real-time modeling of boiler operating parameters is realized.

The invention provides a few-sample target detection method based on singular value decomposition feature enhancement. The problem that a few-sample target detection method is poor in generalization and discrimination is solved. The method comprises the steps of obtaining a target detection image data set; performing feature extraction on the training sample set image; constructing a feature enhancement module to enhance the extracted features; enabling the RPN module to generate a candidate frame area and carrying out RoI alignment; fusing the two feature maps to form a feature fusion layer; positioning and classifying the frame of the target object; carrying out training on the improved Faster R-CNN network; and performing target detection on the to-be-detected image. According to the method, three parts of a feature enhancement module, a feature fusion layer and an Lkl loss function are provided, more essential features of an image and discrimination information in a high-dimensional space are learned, the features have good generalization and discrimination, the positioning and classification precision of few-sample target detection is effectively improved, and the method can be used in the fields of robot navigation, intelligent video monitoring and the like.

The invention discloses a webpage text extraction method based on deep learning. The method comprises the following steps: 1) preparing a data set from a root DOM node to a leaf DOM node; 2) constructing a data set from a root DOM node to a leaf DOM node; 3) labeling data in a data set from the root DOM node to the leaf DOM node; 4) utilizing Fasttext to carry out pre-training and encoding on the label of the path; 5) training an LSTM classification model of the label path text; 6) enabling the LSTM model to predict the label path text; and 7) restoring the extracted webpage text. The invention belongs to the technical field of the Internet, and particularly relates to a webpage text extraction method based on deep learning, which improves the accuracy of resume webpage text extraction.

The invention discloses a logistics scheduling planning method based on a graph neural network and reinforcement learning. The method comprises the following steps: step 1, constructing a complete solution of a vehicle path planning problem instance; step 2, selecting a disturbance controller or a lifting controller by the element controller; after the lifting controller is selected, the lifting operator set forms an action space of the lifting controller; training the graph neural network in the action space; step 3, carrying out de-lifting; step 4, if the meta controller selects a disturbance controller, the disturbance controller randomly selects a disturbance operator to disrupt and reconstruct a feasible solution, and then iterative lifting is carried out to find an optimal solution; and step 5, selecting a solution with the minimum total path length in all visited feasible solutions in the lifting and disturbance process as an optimal solution and a final output solution of the whole algorithm. Compared with the prior art, the optimal solution of the given problem can be efficiently searched, and the method has practical significance for planning problems such as logistics and order distribution.

The invention discloses a target marking method and a target recognition method for building facade damage detection, and a target recognition model can be quickly and finely adjusted through the target marking method. In the target recognition method, the target recognition model finely adjusted by the target marking method is adopted, so that the target recognition model adopted in the building facade damage detection process is more matched with the target building, and the accuracy and adaptability of the recognition process are improved; the problems that the difference of building facades of different buildings is too large, and the generalization ability and accuracy of a target recognition model are difficult to consider at the same time are solved.

The invention provides a universal output method for the digital quantity of a simulation platform. According to the method, firstly, the description of initialize settings is defined, wherein the initialize settings are composed of the sending mode, the pulse width, the pulse number, the level state, the signal triggering delay time and external signal synchronous setting parameters. After that,whether a sent enabling or synchronizing signal is valid or not is judged. A corresponding high-low level output, a positive and negative single pulse output, a limited number of pulse output and a continuous pulse output are generated according to the parameter configuration information after the corresponding time delay time is waited. According to the invention, the unified description of digital quantity output signals is achieved. The defect that a traditional digital quantity output method is poor in universality, long in development period and high in cost can be overcome. The method has the advantages of normative performance, high reusability, short development period and low cost.

The invention provides a satellite peripheral interface simulation system and method. Satellite-borne software and CPU simulation software, the satellite peripheral interface simulation system, a single machine and a power module are connected in sequence. The satellite-borne software, the CPU simulation software and the power module are external software, and the single machine is a fixed codingmodule. The satellite peripheral interface simulation system realizes a corresponding reading function and a corresponding writing function of the register according to an address. The satellite peripheral interface simulation system comprises a configurable module and a configuration file, the configurable module comprises a register address module and other peripheral interface configuration modules, the register address module is used for storing register addresses, and the other peripheral interface configuration modules are used for storing related design parameters. The configuration file is respectively connected with the register address module and other configuration modules of the peripheral interface, and the configuration file is used for initializing the register address module and other configuration modules of the peripheral interface according to configuration.

The present invention provides a method for fault identification and source tracing of complex electromechanical equipment, which includes the following steps: S1, obtaining state monitoring information of electromechanical equipment through multi-dimensional sensors; S2, bringing state monitoring information of electromechanical equipment into a graph convolutional neural network to obtain Diagnostic evidence information; S3, integrate various diagnostic evidence information, and perform fault identification based on the principle of evidentiary reasoning; S4, according to the fault identification result, trace the source of the fault through the fault propagation model, and obtain the root cause location of the fault. The invention makes up for the problems of lack of original data processing level and intelligence level.

The invention relates to an intelligent cabin product automatic detection device which comprises an assembly line and a carrier plate used for supporting and fixing products, at least one detection station is arranged on the assembly line, and the assembly line comprises a rack and a conveying mechanism installed on the rack and used for conveying the carrier plate. Jacking and positioning devices in one-to-one correspondence with the detection stations are arranged on the rack at intervals in the flow direction of the carrier plates, each detection station is provided with a carrier plate blocking device, the conveying mechanism, the jacking and positioning devices and the carrier plate blocking devices are all electrically connected with the assembly line control system, and connectors are arranged on the upper end faces of positioning plates of the jacking and positioning devices of the detection stations. The positioning plate is used for being connected with a first connector arranged on the carrier plate in a matched mode to achieve electric connection between the positioning plate and the carrier plate, the connector on the positioning plate is electrically connected with the intelligent cabin product automatic detection system, and the carrier plate is provided with a second connector used for being electrically connected with a product. According to the invention, the problem of automatic detection of intelligent cabin products is solved, and the detection efficiency is improved.

The invention belongs to the technical field of circuit boards, and particularly relates to a circuit board fixing device. The circuit board fixing device comprises a main frame used for supporting a PCB, a clamping block used for clamping and fixing the PCB, a sliding block and a fast-assembling nut. According to the invention, the universalization of fixing devices for PCBs with different specifications and sizes is well realized, the PCB to be coated with conformal coating is quickly fixed, and the preparation time for preparing special fixing devices for different specifications and sizes before the conformal coating of the PCB is coated is reduced. The positioning precision of the initial positioning origin coordinates of the circuit board corresponding to the coating procedure is improved, and standardized operation is facilitated. According to the device, the problems of accumulation and mark leaving of coating on the edge of the circuit board are solved, the appearance quality of the circuit board is effectively improved, and the quality problems that three-proofing paint which is difficult to clean is left after an existing coating device is used for a long time, circuit board positioning is inaccurate and clamping stagnation is caused by accumulation of pollutants such as excessive paint liquid, and accurate coating cannot be achieved are solved.

The invention provides a system and a method for measuring the water content of ceramic paste. The method comprises the following steps: beta rays with stable energy are radiated, the positive and negative ions of the beta rays are ionized and collected through adopting an electron collecting module and are transmitted to an intelligent processing module, the intelligent processing modules carries out signal amplification treatment on a minimal voltage signal formed by the positive and negative ions, and the amplified voltage signal undergoes nonlinear fitting treatment by adopting a spline function neural network to obtain the water content of the ceramic paste. The measurement method adopting a spline interpolation function to realize discrete input of the nerve network in order to solve the problems of learning and generalization of a large input sample. The problem of low efficiency of present measurement of the water content of the ceramic paste is solved in the invention.

The invention relates to the technical field of neural networks, in particular to a skull fracture detection method and device and a storage medium. The method comprises the following steps: acquiring a to-be-detected head medical image; a target detection model is obtained, the target detection model comprises a random residual network, and a random convolution layer in the random residual network is used for randomly shielding output data of a part of convolution kernels; according to the head medical image, the target detection model is called to output and obtain a target detection result, and the target detection result is used for indicating the detection condition of skull fracture. According to the embodiment of the invention, by introducing the designed random residual network into the target detection model, the sensitivity of network parameters can be reduced, overfitting in a scene with limited training data is avoided, and the generalization ability of the model on a small data set is effectively improved, so that the detection effect of the model is improved.