481 results about "Back propagation algorithm" patented technology

The invention provides an AI intelligent diagnosis method and system based on an intelligent system, and the method comprises the steps: collecting a building electromechanical equipment operation data flow containing multi-source heterogeneous data, such as a vibration spectrum, a thermal imaging characteristic and an energy consumption fluctuation parameter, through an edge calculation node, and carrying out the state time sequence coding to generate a coding sequence; and inputting the sequence into a fault prediction neural network to obtain a multi-dimensional fault feature distribution matrix. And then obtaining a standard diagnosis report of the expert knowledge base and an emergency diagnosis report of the field operation and maintenance record. And then a composite training objective function is constructed based on the result and equipment operation characteristic hidden space mapping, a fault prediction neural network parameter space is optimized through a back propagation algorithm, and a target fault prediction neural network is generated. And finally, fault prediction and diagnosis are carried out on the operation data flow of the building electromechanical equipment by using the network, so that comprehensive, accurate and intelligent fault diagnosis is realized.

The invention discloses a solid-state battery performance testing method and system based on data analysis, and relates to the field of battery performance testing, and the method comprises the steps: building a multi-physics field coupled digital twinborn model, collecting a strain signal and temperature field distribution of a failure region according to a three-dimensional physics field state map of the failure region, and when abnormally abruptly changed, carrying out the testing of the performance of a solid-state battery. Triggering multi-stage early warning, matching physical failure evidences with a historical database, extracting an optimal test parameter combination through a meta-learning framework, dynamically adjusting a charging and discharging strategy and monitoring frequency, generating a real-time data stream, and updating parameters of a multi-physics field coupled digital twin model through a back propagation algorithm. According to the method, the test parameters are dynamically optimized through the meta-learning framework, the performance of the solid-state battery is monitored and predicted, the safety and the reliability of the solid-state battery are improved, and a charging and discharging strategy is optimized to prolong the service life of the battery.

The invention is suitable for the field of spatio-temporal data prediction, and provides a CNS model soil humidity prediction method fusing dynamic spatio-temporal pruning and a WOA algorithm, and the method comprises the steps: building an initial CNS model based on a multiple convolution module, a BiLSTM module and an attention mechanism module; taking the soil time sequence data as the input of an initial CNS model, and performing preliminary training on the initial CNS model; performing dynamic pruning on each module of the preliminarily trained model by adopting a dynamic space-time pruning strategy; model parameters are updated through a back propagation algorithm, and the CNS model used for soil humidity prediction is obtained. The CNS model integrates a convolutional neural network CNN, a space-time diagram convolutional network STGCN, a bidirectional long short-term memory network BiLSTM and an attention mechanism layer Attention, multi-modal data from different data sources can be comprehensively processed and analyzed, and the CNS model shows higher precision and robustness in a soil humidity prediction task; through combination of dynamic pruning and WOA algorithm optimization, the model can maintain stable prediction performance.

The invention discloses a cross-system fault diagnosis method and system combined with multi-dimensional anomaly detection, and relates to the technical field of fault diagnos.The method comprises the steps that a graph neural network with a liquid time constant network unit as a node is constructed through a dynamic topology dependency relationship and a multi-dimensional key performance index flow; each unit describes state evolution through a coupled ordinary differential equation system, and a liquid state time constant can be adaptively adjusted. A time back propagation algorithm is adopted to train a model to learn a normal behavior track contour reference, and anomaly is detected through a dynamic time warping distance. And determining a fault propagation path and a root cause through anti-fact intervention and forward integral solution. And generating an optimal diagnosis action sequence in a liquid graph neural network simulation environment, and calculating a reward value based on execution efficiency, accuracy and a repair effect to carry out strategy optimization. The abnormal detection accuracy and the root cause positioning precision are improved, the fault repair time is shortened, the operation and maintenance cost is reduced, and an intelligent fault diagnosis solution is provided for a complex information technology system.

The invention relates to the technical field of diagnosis and treatment path planning, in particular to an intelligent diagnosis and treatment path dynamic planning system based on symptom ontology reasoning, which comprises a symptom specialized mapping module, a holographic panoramic navigation module, a path optimization core module and a real-time feedback module, wherein the symptom specialized mapping module generates a dynamic medical resource recommendation parameter set by fusing a symptom ontology feature vector, medical institution treatment efficiency data and a specialized association rule base, and dynamic medical resource recommendation parameters comprise a multi-dimensional matching degree tensor and a treatment efficiency quantification matrix; the holographic panoramic navigation module receives the multi-dimensional matching degree tensor and constructs a three-dimensional medical resource topological graph, wherein the three-dimensional medical resource topological graph comprises hospital nodes and department connecting lines; the path optimization core module analyzes hospital nodes in the three-dimensional medical resource topological graph, and generates a path with optimal cost performance based on a multi-objective optimization algorithm of a treatment efficiency quantification matrix; and the real-time feedback module acquires and updates the treatment efficiency quantization matrix through a back propagation algorithm to form a closed-loop optimization system.

The invention provides a lightweight visible light ship target detection method based on edge feature guidance, and relates to the technical field of ship detection image data processing, and the method comprises the steps: collecting remote sensing satellite images, and carrying out the random distribution of the images after screening and marking, and obtaining a training set and a verification set; the backbone network module comprises a plurality of Conv modules and C3k2 modules which are mutually stacked; the neck module comprises a detail-enhanced convolution module and a hierarchical pyramid module based on dynamic feature aggregation; in the head module, after the features of all detection layers are subjected to independent convolution processing, feature transformation is carried out through a multi-branch detail enhancement convolution module; performing data enhancement on the training set; and obtaining a trained ship target detection model through a back propagation algorithm and a gradient descent optimization method. According to the invention, the lightweight and precision improvement of the detection head are realized, the robustness of the model to the illumination change is enhanced, and the global semantic information and the local detail features are fused to balance the detection of the small target and the large target.

The invention provides an image segmentation method and device based on an SAM (Segmentation All Model), and the method comprises the steps: obtaining a to-be-segmented image and first segmentation prompt information; inputting the to-be-segmented image and the first segmentation prompt information into a segmentation cutting model SAM obtained by pre-training to output a first image segmentation result; wherein the SAM obtained through pre-training is obtained by taking a mask decoder as a learnable parameter of model training and updating the weight of the learnable parameter through a back propagation algorithm. Therefore, according to the embodiment of the invention, the mask decoder of the SAM is set as the learnable parameter and the mask decoder is finely adjusted through the back propagation algorithm, and the whole SAM does not need to be adjusted, so that the fine adjustment efficiency of the model is ensured, the accuracy of the model is improved, and the accuracy and efficiency of image segmentation are improved.

The disclosure provides a transformer fault detection model training method, a fault diagnosis method, and a related device. The method comprises: obtaining an initial voiceprint signal of a transformer, and a fault type corresponding to the initial voiceprint signal; preprocessing the initial voiceprint signal by using a wavelet packet analysis method, so as to obtain an input signal, and according to the input signal and the fault type, establishing an input signal data set; according to a preset feature extraction algorithm, performing feature extraction on a first input signal in a training data set to obtain a first voiceprint feature; using the first voiceprint feature and a first fault type corresponding to the first input signal to train an initial detection model, so as to obtain a first training result; according to the first training result and the fault type, determining a loss function; on the basis of the loss function, using a back propagation algorithm to iteratively adjust weight values of the initial detection model, until the loss function converges, so as to obtain a fault detection model. In the present disclosure, accurate transformer fault detection is realized by utilizing the obtained fault detection model.

The invention provides a PCB processing method, device and equipment based on technological parameter optimization, and the method comprises the steps: carrying out the feature extraction of PCB design data, and obtaining feature data; matching an adaptive initial process parameter combination according to the feature data; performing quality detection on the PCB which is processed for the first time based on the initial process parameter combination, and analyzing detection data by using a defect identification algorithm to obtain a quality evaluation result; based on a quality evaluation result, optimizing the initial process parameter combination by using an error back propagation algorithm to obtain an optimized process parameter combination; and controlling secondary processing of the PCB based on the optimized process parameter combination. According to the method, the initial process parameter combination is optimized according to the quality evaluation result of the circuit board, so that the defect that an existing processing method cannot perform self-adaptive optimization according to the actual condition of the circuit board is overcome.

The invention relates to the technical field of intelligent communication, and discloses an end-to-end optimization method, a signal constellation geometry and probability joint shaping optimization method in an end-to-end intelligent communication system, and a communication device based on a neural network. According to the method, a transmitter is constructed by using a neural network, and a received signal passes through a two-stage equalization processing system to form an end-to-end optimization framework without assistance of a channel model. The transmitter adopts a modular design, can realize joint optimization of constellation probability shaping (PS) and geometric shaping (GS), and performs joint updating on all trainable parameters of the system based on a total loss function containing an equilibrium error and a decoding error through a back propagation algorithm. According to the invention, computing resources are effectively saved; the two-stage equalization system can accurately compensate channel damage, the realized PS and GS joint optimization can significantly improve the performance of the communication system and reduce the bit error rate, and the method has good performance advantages and practical value in an actual optical fiber communication system.

The invention provides a federated learning method and system based on a hybrid expert model, and relates to the technical field of machine learning, and the method comprises the steps: carrying out the initialization of a local gating network of a current client through employing a global gating network, obtaining an initial local gating network # imgabs0 #, carrying out the feature extraction of a local data set, and obtaining a local data set; k target expert models Sk with the highest adaptation degree with the data features are screened out from the multiple expert models or a required target expert model # imgabs1 # is downloaded from a server side by using a Top-K-based adaptive expert selection mechanism to ensure that all the selected expert models can be used locally; and the client uses local data to perform forward calculation on the target expert model, obtains training feedback parameters of the target expert based on a prediction result and a real label, and optimizes the target expert model through a back propagation algorithm # imgabs2, and the client performs weighted summation on calculation results of the target expert models to obtain a comprehensive prediction result. And the generalization ability and performance of the model are improved.

The invention discloses a trajectory prediction method and system based on a CNN-LSTM-Attention neural network, belongs to the technical field of machine learning and data analysis, and solves the problem that the precision and reliability of trajectory prediction are reduced due to noise and accumulative errors existing in IMU (Inertial Measurement Unit) data. The method comprises the steps of collecting sensor data, performing preprocessing, dividing the sensor data into a training set and a test set, and performing standardization processing on input features and target output; constructing a hybrid neural network architecture, setting structures and parameters of each layer in the hybrid neural network architecture, and defining a loss function and an optimizer; training set data is adopted to train the hybrid neural network architecture, and a back propagation algorithm is adopted to optimize model parameters; calculating an error index between the predicted trajectory and the real trajectory, and evaluating the prediction performance of the hybrid neural network architecture according to the error index; and selecting the neural network with the best prediction performance to carry out real-time trajectory prediction. The method is suitable for application scenes such as intelligent traffic systems, unmanned driving and route planning.

A bidirectional autoencoder learns or approximates an identity mapping as it trains a single network with a version of the new bidirectional backpropagation algorithm. Ordinary unidirectional autoencoders find many uses in image processing and in large language models. But they use separate networks for encoding and decoding. Bidirectional autoencoders use the same synaptic weights for encoding and decoding. The forward pass encodes while the backward pass decodes. Bidirectional autoencoders improved network performance and significantly reduced memory usage and used fewer parameters. Simulations compared unidirectional with bidirectional autoencoders for image compression and denoising. The models trained on the MNIST handwritten-digit and CIFAR-10 image datasets. The performance measures were the peak signal-to-noise ratio and the index of structural similarity. Bidirectional autoencoders outperformed unidirectional autoencoders and still reduced the number of trainable synaptic parameters by about 50%.

The invention provides a PCB manufacturability intelligent analysis and early warning method and system based on artificial intelligence, and the method comprises the steps: collecting and marking the multi-source time sequence process parameter data in the PCB design and manufacturing process under working conditions, building a dynamic causal graph model with direction and time lag marks through a sliding window and standardization processing by applying a causal discovery algorithm, and carrying out the calculation of the dynamic causal graph model. Dynamic expression of causal relationships among process variables is realized; when manufacturing abnormity is detected, abnormity attribution is carried out by combining a Bayesian back propagation algorithm, high-contribution-degree root dependent variables are screened, the effectiveness of root causes is verified through virtual intervention simulation and statistical test, and finally verification results and a causal mode are stored in a knowledge base to support subsequent rapid matching and reasoning. According to the method, the accuracy, efficiency and interpretability of PCB manufacturing abnormity attribution are improved, and process optimization and preventive intervention are facilitated.

The invention discloses a generative defense method and system for resisting attacks, relates to the technical field of network security, and aims to solve the problems that an existing defense scheme is high in calculation overhead and poor in real-time performance, static defense is easy to bypass, and robustness and accuracy are difficult to balance. The method comprises the following steps: constructing an adversarial network model by taking a pre-trained target model as a discriminator and a generative model as a generator; constructing a total loss function by combining a defensive loss function and an accuracy loss function, generating a defensive benign sample by adding defensive disturbance into a benign sample, generating a defensive confrontation sample by adding defensive disturbance after generating a confrontation sample based on the benign sample, and inputting the three types of samples into a target model to obtain total loss; and training the generative model to convergence by using a back propagation algorithm to obtain a trained adversarial network model for classification of defense disturbance samples. Defense generation network training is completed in the training stage, only defense disturbance needs to be overlaid in the reasoning stage, the real-time requirement is met, the robustness and accuracy of the model can be balanced, and the method is suitable for various attack scenes.

The invention relates to a geological analysis technology, and discloses a multi-head attention model training method fused with geological rules, which is used for improving the scientificity and accuracy of geological body model prediction and improving the interpretability of a geological analysis process. According to the scheme, firstly, feature vectors of geological attribute data and geological coordinate data are spliced to obtain a fusion feature vector, then spatial locality and directivity are considered at the same time, geomorphic multi-head attention calculation is conducted on the fusion feature vector, and an attention feature sequence is obtained; then extracting multi-scale features from the input geological map, constructing a cross-fault weight mask matrix based on fault constraints, and obtaining comprehensive features through fusion; calculating a loss value by adopting a joint loss function containing prediction loss and fault non-penetrability constraint terms, and updating model parameters through a back propagation algorithm to complete model training; and finally, drawing an attention thermodynamic diagram for visual display, and superposing the attention thermodynamic diagram with the three-dimensional geologic model.

The invention discloses a rapid nondestructive testing method and system for the lipid content and deterioration degree of red pine nuts based on hyperspectral imaging and deep learning, and belongs to the technical field of nondestructive testing of the quality and safety of agricultural and forestry products. The method is provided for solving the problems that an existing method for detecting the lipid content and the oxidation degree in the red pine nut kernels is generally complex in operation process, high in large-batch detection cost, long in consumed time and difficult to achieve detection in the whole storage and transportation process. The method is characterized by comprising the following steps: acquiring original near infrared spectrum data of a pine nut sample through a collected hyperspectrum; determining the lipid reference content truth value and the oxidation deterioration reference degree of the pine nut samples at different sampling times, and establishing a database according to the values; the method comprises the following steps: preprocessing collected original near infrared spectrum data of red pine nuts, and dividing red pine nut sample data collected in different batches into a training set and a verification set; and designing an improved one-dimensional cavity convolutional network based on a dynamic weight distribution module to construct a deep learning model, wherein the deep learning model is used for constructing a deep learning structure suitable for spectral feature analysis of the red pine nuts. And a back propagation algorithm is adopted to train the constructed model, and reverse updating of network weight parameters is realized by minimizing a loss function. When the performance of the constructed model meets the rapid detection requirement, the method is used for efficient and lossless synchronous detection of the lipid content and the oxidative rancidity degree of the red pine nuts.

The invention relates to the technical field of ship design optimization, and discloses a hull shape optimization method based on neural network modeling. In the data acquisition stage of the method, initial appearance parameters and hydrodynamic performance data of a ship body are obtained, the appearance parameters comprise geometric dimensions and shape features, and the performance data comprise resistance coefficients and wave-making resistance values. In the neural network construction stage, a neural network model with a multi-layer perceptron structure is trained by using collected data, weights are updated through a back propagation algorithm, and a nonlinear mapping relation between appearance parameters and hydrodynamic performance indexes is established. In the shape optimization stage, the trained neural network model is used for carrying out iterative adjustment on the shape of the ship body, fluid dynamic performance indexes are recalculated through the model after each adjustment until preset convergence conditions are met, and finally optimized ship body shape data are output. According to the method, partial complex calculation is replaced by the neural network, and intelligent optimization of the hull appearance is realized.

The invention provides a model training method and device, equipment and a storage medium, and relates to the technical field of computers, in particular to the technical field of neural network models and model training. The specific implementation scheme is as follows: a calculation unit executes quantization matrix multiplication based on Hadamard pre-transformation on an activation tensor and a weight tensor of a target model stored in a memory so as to generate an output tensor of a linear layer based on a low-precision tensor with smaller data bit width; using the output tensor and a subsequent network layer of the target model to complete forward propagation so as to obtain a loss value; and according to the loss value, updating model parameters of the target model stored in a memory through a back propagation algorithm. By means of the technical scheme, on the premise that the model training precision is guaranteed, memory resource occupation and the calculation amount in the calculation process can be remarkably reduced, and the training cost is reduced.

The invention discloses a neural radiation field rendering method based on dynamic hash coding, and the method comprises the steps: employing the feature sequence data as the input, calculating the density value and color value of each sampling point through the forward propagation of a neural network, carrying out the volume rendering integral operation according to the ray tracing principle in the direction of a ray, and obtaining the feature sequence data; judging a final color output result of the current pixel point; according to an error value between the color output result and a real image, updating a network parameter weight through a back propagation algorithm, and if the error value is greater than a convergence threshold, continuing to iterate the training process to adjust a feature coding strategy to obtain an optimized neural radiation field model parameter; and after the rendering performance configuration parameters are obtained, optimizing a storage allocation strategy of feature data through a memory pool management mechanism, and if the current memory occupancy rate exceeds a safety threshold, starting a data compression algorithm to reduce the storage space requirement, and obtaining a real-time rendering output result. According to the invention, high-quality real-time rendering of the dynamic scene is realized.

The invention provides an atom pair distribution function calculation method based on a back propagation algorithm, and the method comprises the steps: building a micromapping relation between a material microstructure and experimental PDF data through constructing a physical calculation layer containing lattice parameters and atom displacement; defining a loss function by taking experimental data as a supervision signal, and synchronously optimizing tens of thousands of atomic positions and lattice parameters by utilizing gradient back propagation of the loss function covering the experimental data and model calculation result deviation; and a special optimization algorithm CrystalAdam and dynamic learning rate scheduling are combined, so that global efficient search of a high-dimensional parameter space is realized. Therefore, compared with a traditional modeling method, the method provided by the invention not only can remarkably improve the optimization efficiency, but also can provide finer local adjustment in a complex structure space, thereby achieving higher precision.

The invention relates to a four-body coupling vehicle base vibration noise prediction optimization method and system, and relates to the technical field of rail transit operation and maintenance optimization, and the method comprises the steps: obtaining a vehicle base multi-physics field correlation monitoring data set; carrying out system level assembly in a multi-physics field co-simulation environment to generate a coupled system state parameter matrix; generating sound ray bending trajectory data; loading the vehicle operation condition time sequence data to a wave equation solver, and outputting vibration noise propagation characteristic field distribution data; building a network topology structure to form a prediction model, and executing a back propagation algorithm to train network parameters to a convergence state; designing a vehicle scheduling scheme chromosome coding structure to generate a candidate solution population; and performing selection, crossover and variation genetic operations on the primary solution set, and outputting a vehicle base operation and maintenance scheduling real-time optimization scheme. The method has the beneficial effects that the precision and generalization ability of vibration noise prediction in the future time period are greatly improved, and the fundamental transformation from passive vibration isolation treatment to active prediction optimization is realized.

The invention relates to the technical field of pressure control, in particular to a butterfly valve pressure self-adaptive adjusting method, device and equipment and a storage medium. According to the method, a pre-constructed neural network model is introduced, and user set pressure, actual measurement pressure, current control signals and error parameters serve as input variables; dynamic generation and real-time self-adaptive adjustment of PID control parameters are realized, the technical bottlenecks that parameter setting of a traditional PID controller is complex and adaptability to non-linear and time-varying systems is insufficient are effectively solved, and the parameter adjustment period is shortened; the weight parameter of the neural network is continuously optimized through a back propagation algorithm, so that the inhibition capability of the system on external disturbance is remarkably enhanced, the overshoot is reduced, and the long-term stability and control precision of pressure control are improved; the target step number is calculated in combination with butterfly valve hardware parameters, the accuracy and rapidity of stepping motor driving are ensured, and the problem of response delay of traditional PID control is effectively solved.

The invention relates to the technical field of image classification, in particular to an edge-cloud collaborative rice disease monitoring method and system for precision agriculture, and the method comprises the steps: constructing a rice disease recognition network; a squeezing-incentive attention module and a multi-scale convolution-space attention module are introduced into a backbone network and a neck network of the rice disease recognition network; replacing standard convolution in the backbone network and the neck network with deep separable convolution; training the improved and optimized rice disease recognition network to obtain a lightweight student model; soft label distillation loss is constructed based on prediction distribution output by a pre-trained deep teacher model and a student model; a total loss function is constructed in combination with cross entropy loss, training of student models is guided through a back propagation algorithm, and a rice disease lightweight recognition model is obtained and used for recognizing rice diseases. Through edge end deployment of a lightweight model obtained through distillation, rapid reasoning of a high-precision model is realized on low-power-consumption edge equipment.

The invention discloses a user behavior prediction method based on a causal tendency score. The method comprises the following steps: collecting historical behavior information of a to-be-detected user from a historical database; carrying out behavior feature extraction, and carrying out normalization processing on behavior features; establishing a behavior prediction model; model parameters are adjusted through a back propagation algorithm and an optimizer; analyzing a causal relationship between the behavior characteristics; analyzing a reason for generating a prediction error result by the behavior prediction model; and performing user behavior prediction by using the optimized behavior prediction model, and making an optimization strategy according to a causal relationship revealed by the tendency score. According to the method, deep mining and quantitative evaluation of the causal relationship behind the user behavior are realized by constructing the feature association network, screening the causal candidate pairs, quantifying the causal effect and calculating the causal tendency score, so that a root cause of a behavior prediction error is disclosed, and the interpretability and decision support capability of the model are also remarkably improved.

The invention relates to a fermentation process optimization regulation and control method based on reinforcement learning and comparative learning. The method comprises the following steps: acquiring historical batch data and batch data to be regulated and controlled in a penicillin fermentation process, and constructing an optimized regulation and control model by adopting a deep reinforcement learning double-actor network; optimizing an action value function by combining a loss function of comparative learning, enhancing the identification capability of similarity and difference between states, inputting historical batch data into a model for training, calculating a difference value between the penicillin yield and an expected yield, updating parameters by adopting a back propagation algorithm, and automatically adjusting hyper-parameters through Bayesian optimization; and finally, inputting data of a batch to be regulated and controlled into the trained neural network regulation and control model, and realizing penicillin yield regulation and control through iterative action output. By adopting the method, the penicillin yield can be effectively regulated and controlled, and powerful technical support and theoretical guidance are provided for optimization regulation and control of the penicillin fermentation process.

The invention relates to the technical field of laser control, in particular to an intelligent control method and system of an MOPA laser. The method comprises the following steps: constructing a target training data set and a one-dimensional convolutional neural network model; training a one-dimensional convolutional neural network model based on the target training data set in combination with a target loss function and a back propagation algorithm to obtain a target model; acquiring current operation parameters of the MOPA laser; inputting the current operation parameters into a target model, and outputting target regulation and control parameters by the target model; the target regulation and control parameters comprise the working current of a pumping module in the MOPA laser and the working temperature of a doped fiber; based on the target regulation and control parameters, the working current of a pumping module in the MOPA laser and the working temperature of a doped optical fiber are adjusted, the working temperature of the doped optical fiber comprises a plurality of different target temperatures, and each target temperature corresponds to a different position on the doped optical fiber. In this way, the problems caused by the nonlinear effect generated under high-power pumping can be reduced.

The invention discloses an image classification method and device, equipment, a medium and a product, and relates to the field of image classification, and the method comprises the steps: obtaining to-be-classified image data; according to the to-be-classified image data, performing classification by using a classification model to obtain a classification category; the classification model is a trained pulse neural network; the spiking neural network comprises a plurality of neural layers; each neural layer comprises a plurality of neurons; each neuron comprises an emission threshold value and an input resistance; the training process of the classification model specifically comprises the following steps: taking sample image data as the input of a spiking neural network, taking a sample classification category as the output of the spiking neural network, and determining a total loss function of the spiking neural network according to cross entropy loss and adaptive sparse loss; and optimizing the synaptic weight, the emission threshold and the input resistance of the spiking neural network by using a back propagation algorithm to obtain a classification model. The method can improve the classification precision.

The invention discloses a wind power generation system control method and system based on an improved recurrent neural network (RNN), and belongs to the field of wind power generation system control. The method comprises the following steps: training an improved RNN model by adopting a training data set, and carrying out forward propagation on an input vector to obtain gain data; quantizing a loss value between the gain parameter and an ideal gain parameter, constructing a loss function according to the loss value, obtaining a weight gradient of each layer by using a gradient back propagation algorithm, updating a network weight according to the weight gradient and in combination with a momentum gradient descent algorithm, applying amplitude constraint to a recursive weight, and performing iterative training to obtain an ideal gain parameter. A trained RNN network model is obtained; obtaining a gain parameter of the wind power generation system according to the trained RNN network model, and controlling the operation state of the wind power generation system according to the gain parameter; according to the method, PI controller self-adaption is achieved by introducing the recurrent neural network, the system response time, static errors and overshoot can be remarkably reduced, and therefore the system performance is improved.

The invention provides a low-temperature-resistant performance evaluation method for a composite material, and belongs to the technical field of material performance determination.The method comprises the steps that firstly, in a low-temperature environment box, a temperature sensor is used for accurately controlling the temperature of a sample, and meanwhile, a high-frequency strain sensor is used for collecting data; then, wavelet transform is adopted to carry out denoising and decomposition on the strain signals, and key low-temperature characteristic parameters are extracted; and constructing a performance evaluation model based on deep learning, inputting the temperature field, the strain characteristics and the material basic parameters into the model, and predicting the low-temperature strength. A prediction result is verified through a standard test, a prediction error is calculated, an error distribution matrix is established, and an improved back propagation algorithm is adopted to optimize the model. Finally, a fuzzy comprehensive evaluation method is applied, indexes such as strength retention rate and strain stability are comprehensively considered, and the low temperature resistance grade of the composite material is scientifically evaluated. The method solves the problem that the actual mechanical properties of the composite material in the low-temperature environment are difficult to comprehensively reflect in the prior art.