131 results about "Backpropagation neural nets" patented technology

The invention provides a gear peeling time-varying meshing stiffness prediction method and system based on a back propagation neural network, and the method comprises the steps: considering a tooth surface peeling fault based on a gear tooth bearing contact analysis method, and constructing a helical gear pair time-varying meshing stiffness calculation model; a real irregular tooth surface peeling area is fitted by adopting a least square ellipse fitting method to obtain an ellipse appearance representation, any peeling position is completely described through six key geometric parameters, the geometric parameters of the ellipse appearance are systematically traversed and fitted, and diversified peeling appearance samples are generated. Introducing a tooth profile deviation matrix corresponding to the peeling morphology sample into a tooth surface bearing contact analysis model, and constructing a training data set; and constructing a back propagation neural network model of multiple hidden layers, and performing end-to-end training by using the training data set, so that the back propagation neural network model learns a nonlinear mapping relationship from geometric parameters to a time-varying meshing stiffness curve, thereby predicting the time-varying meshing stiffness under any peeling morphology.

The invention provides an intelligent risk screening system and method based on distribution network load prediction, and relates to the technical field of intelligent power grids, and the method comprises the steps: carrying out the multi-scale decomposition processing and bidirectional feature coding of historical load data of a distribution network, and obtaining a time sequence coding feature; graph structure features are extracted from the multi-node load graph of the distribution network, feature fusion is carried out on the graph structure features and the time sequence coding features, and a multi-dimensional prediction matrix is obtained; performing a back propagation neural network model optimization operation based on genetic algorithm learning on the multi-dimensional prediction matrix through the environmental influence data to obtain a load prediction curve and a prediction confidence coefficient, and determining overload risk early warning of the distribution network load according to the load prediction curve and a preset risk threshold value; according to the method and the device, accurate screening of the overload risk of the distribution network based on space-time modeling and a prediction confidence mechanism can be realized, so that the accuracy of load early warning is improved.

The invention provides a gearbox variable working condition mechanical loss torque efficiency correction test method based on a neural network. The method comprises the following steps: S1, determining a to-be-tested gearbox and basic parameters; s2, building a gearbox variable working condition test system; s3, acquiring basic operation data of the gearbox; s4, calculating theoretical mechanical loss torque and efficiency of the gear box; s5, the actually-measured mechanical loss torque and efficiency of the gearbox are collected, and an error data set is generated; s6, constructing and training an error correction model fusing the back propagation neural network and support vector regression; s7, performing mechanical loss torque and efficiency correction based on the fusion model; and S8, verification and feedback optimization of a correction result. According to the method, on the basis of all-working-condition data, the defect of a single model is overcome through double-algorithm fusion, precise correction of the mechanical loss torque and efficiency of the gearbox under the variable working conditions is achieved, and reliable data support is provided for gearbox design optimization, working condition matching and energy efficiency evaluation.

The invention provides an airport noise prediction method based on a back propagation neural network and a support vector machine, and belongs to the technical field of air traffic transportation noise, and the method comprises the steps: collecting airport noise historical data, obtaining flight historical data of airplanes with different powers, and obtaining a prediction input variable and a prediction output variable; according to the input and output variables and an elastic back propagation algorithm, in combination with a gradient time sequence dynamic rarefaction method and an adaptive structure adjustment method, establishing a back propagation neural network prediction model; constructing an airport noise kernel function, and establishing a support vector machine prediction model in combination with a cross validation method; predicting by using a back propagation neural network prediction model and a support vector machine prediction model, comparing and analyzing prediction results in combination with evaluation indexes, and selecting an optimal prediction result as an airport noise prediction result; the problems that an existing noise prediction algorithm depends on a large amount of historical data, the calculated amount is huge, and the noise prediction efficiency is low are solved.

The invention discloses a time sequence reconstruction and machine learning coupled chlorophyll monitoring data breakpoint restoration method, and belongs to the technical field of water quality monitoring. The invention discloses a chlorophyll monitoring data breakpoint restoration method based on coupling of time sequence reconstruction and machine learning, and the method comprises the following steps: S1, collecting water quality monitoring data, and cleaning the monitoring data to obtain preprocessed data; s2, performing time sequence reconstruction on the preprocessed data to obtain a weekly average 1 data set; s3, respectively constructing a radial basis function neural network model and a back propagation neural network model by taking the chlorophyll concentration as a response variable and the conventional water quality parameter as a predictive variable; s4, performing performance evaluation on each model by taking a root mean square error, an average absolute percentage error, goodness of fit and relative error distribution statistics as evaluation indexes, and screening out an optimal model; and S5, applying the conventional water quality parameters in the breakpoint interval of the chlorophyll monitoring data in the water body to the optimal model, and outputting the restored chlorophyll concentration value to complete the dynamic restoration of the breakpoint.

A variable curvature bending arc control method for a roll bending machine includes specific steps as follows: collecting and pre-processing data; initializing the network structure, and determining the parameters such as a number of input layer nodes, hidden layer nodes and output layer nodes and other parameters of the back propagation neural network according to the learning sample data; encoding a weight value and a threshold value of a back propagation neural network into individuals in a population according to set coding rules, and initializing the population according to a set population size and a random initialization method; GA-PSO iterative operation; initializing the back propagation neural network parameter; training the back propagation neural network; and an application of the control model. The control method accurately controls the variable curvature bending arc torque for the roll bending machine by combining genetic algorithm, back propagation neural network and particle swarm optimization.

The invention discloses an agricultural machinery braking performance detector based on a Beidou satellite single-point positioning mode, and belongs to the technical field of braking performance detection. The agricultural machinery braking performance detector comprises a signal collector and a host, and the signal collector is internally provided with a satellite signal receiving module compatible with a BDS and a GPS; a sequence parameter extraction module, an integral distance measurement module, an error correction module, a slope calculation module and a performance evaluation module are arranged in the host; the method comprises the following steps: collecting three-dimensional motion data of a detected agricultural machine, calculating an uncorrected braking distance by using a complex Simpson integral algorithm, obtaining an error correction distance caused by single-point positioning signal drift based on a back propagation neural network, calculating the braking distance of the agricultural machine in combination with a slope correction value, and finally judging that the braking performance of the detected agricultural machine is qualified. According to the method, a ground differential base station and an external supporting assembly do not need to be erected, the problem that field detection deployment is tedious is solved, signal drifting and road slope interference are eliminated, and detection convenience and accuracy are improved.

The invention provides a material constitutive parameter inversion method which can be applied to the technical field of artificial intelligence. The material constitutive parameter inversion method comprises the following steps: acquiring an elastic modulus and an actually measured load depth curve of a to-be-tested material in a single nanoindentation test; performing extrapolation translation preprocessing on the actually measured load depth curve to obtain an equivalent ideal curve; carrying out characterization processing on the equivalent ideal curve to extract input characteristic parameters which comprise press-in rigidity, unloading rigidity and plastic work proportion; the indentation rigidity, the unloading rigidity, the plastic work ratio and the elastic modulus are input into a pre-trained indentation constitutive mapping model to predict the yield strength and the strain hardening index of the to-be-tested material, and the indentation constitutive mapping model is a full-connection back-propagation neural network pre-trained through a finite element virtual test sample. The invention further provides a material constitutive parameter inversion system and device, a storage medium and a program product.

The invention provides a turbine blade fatigue performance influence factor analysis method and system and a medium, and belongs to the technical field of turbine blade performance prediction.The turbine blade fatigue performance influence factor analysis method includes the steps that a small number of training samples are generated, and a back propagation neural network (BP neural network) describing the relation between the fatigue life of a turbine blade and input variables is constructed; a training sample is added to the sequence, and the BP neural network is updated until the calculated failure probability is converged; based on a failure sample obtained in the process of solving the failure probability, estimating conditional failure probability estimation values at different sample points by using a Bayesian inference theory; and obtaining a fatigue reliability sensitivity estimation value of each input variable according to an average difference between the fatigue failure probability estimation value of the turbine blade and the condition failure probability estimation value of each failure sample point. The method solves the problems that when an existing agent model method is used for solving the reliability and sensitivity of the turbine blade, the sample requirement is large, efficiency is low, and the method cannot be suitable for a high-dimensional problem.

The invention provides a multi-source adaptive fusion positioning method and system and a medium, and belongs to the technical field of automatic driving and navigation positioning, and the method comprises the steps: receiving original data synchronously collected by each positioning module, extracting an input feature vector, and carrying out the standardization; inputting the standardized input feature vector into a pre-trained K-nearest neighbor model, and calculating to obtain an initial confidence coefficient corresponding to each positioning module; jointly inputting the initial confidence coefficient and the input feature vector into a back propagation neural network model to obtain a corrected sensor confidence coefficient vector, and generating a positioning mode recommendation classification; according to the positioning mode recommendation classification, adaptively selecting a working mode of the positioning system; based on an extended Kalman filter algorithm, the sensor confidence vector is used to perform state update, and parameters for vehicle navigation control are output. According to the method, GNSS false fixation is effectively detected, multi-source data are fused, the confidence coefficient is dynamically adjusted, and the positioning reliability is improved.

The invention discloses a borrowing type successive approximation analog-to-digital converter adopting a back propagation neural network for calibration. The borrowing type successive approximation analog-to-digital converter comprises a system composed of a positive and negative capacitance digital-to-analog converter adopting a three-stage split bridge type capacitor array, a bootstrap sampling switch, an automatic return-to-zero comparator, an SAR logic control unit, an original-to-binary module and a BPNN calibration engine. Bootstrap sampling switches are respectively arranged at the tail ends of the positive and negative capacitance digital-to-analog converters, a comparator is connected with the output ends of the two, and the output is connected with an SAR logic control unit; the original-to-binary module converts the code generated by the SAR logic control unit into a binary code; and a BPNN calibration engine receives the code and performs calibration calculation by using a trained back propagation neural network model. By implementing the converter disclosed by the invention, hardware implementation can be simplified while high performance is ensured, so that the ADC realizes efficient and accurate signal conversion under a 180-nanometer BCD process, and power consumption and cost are reduced.

This application relates to the field of computer science and provides a fault location method and apparatus. The method includes: inputting a target fault description statement into a label recommendation model to obtain at least one fault label output by the label recommendation model; inputting the at least one fault label into a fault location model to obtain the fault root cause corresponding to the target fault description statement output by the fault location model; wherein, the label recommendation model is obtained by training a Long Short-Term Memory (LSTM) neural network model based on fault description statement samples, fault label samples corresponding to the fault description statement samples, and a weighted binary cross-entropy loss function; the fault location model is obtained by training a backpropagation backpropagation (BP) neural network model based on fault label samples and fault root cause samples corresponding to the fault label samples. The fault location method and apparatus provided in this application can quickly find the root cause of a fault, saving fault location time.

The invention discloses a radar track target classification method based on statistical knowledge and a neural network, and the method comprises the steps: collecting original track data, and carrying out the marking and cleaning processing, and obtaining a standard track data set; extracting kinematics, bistatic radar cross section and track morphological features based on the standard track data set, and constructing an original feature set; performing feature weight learning through a neighborhood component analysis method, and screening according to a preset weight threshold to obtain an effective feature subset; performing statistical distribution analysis on the effective feature subsets according to target categories, determining significant features of each target category, and setting a threshold interval; performing preliminary classification based on a statistical knowledge classifier, and determining whether the to-be-classified track is an easily-classified sample or a difficultly-classified sample through a category discrimination function; and classifying the samples difficult to classify based on a back propagation neural network classifier, outputting the probability value of each target category through a neural network model, and taking the target category with the maximum probability value as a final classification result. According to the invention, the target classification accuracy and real-time performance can be improved.

The application discloses a stiffness and damping optimization method, system, device, medium and product of a vehicle suspension based on a back propagation neural network, and relates to the technical field of vehicles. The method comprises the following steps: training a back propagation neural network based on a training set to obtain a trained back propagation neural network; extracting the trained back propagation neural network by adopting a Taylor expansion method and a finite difference method to obtain a matrix; constructing a nonlinear multivariate inter-axle stiffness-damping objective function based on the matrix; obtaining an input variable sequence and an output variable sequence of a vehicle; and optimizing the stiffness and damping of the vehicle suspension under a constraint condition by taking the function value of the nonlinear multivariate inter-axle stiffness-damping objective function as the target, so as to obtain an optimal stiffness-damping matrix and complete the stiffness and damping optimization of the vehicle suspension. The application can improve the driving smoothness and steering stability of the vehicle.

According to the multi-objective optimization method for the working parameters of the sprocket type residual film recycling machine, the sprocket type residual film recycling machine is used as a research object, and the angular velocity of a picking device, the angular velocity of a film and impurity separating device and the working angle of the film and impurity separating device are set as key variables to be optimized. In order to obtain the optimal residual film impurity rate and picking-up rate, discrete element simulation is performed through an EDEM, a back propagation neural network is trained by using simulation data, and a prediction model of the residual film impurity rate and picking-up rate is established. A prediction result of the neural network is optimized by adopting a non-dominated multi-objective genetic algorithm, and optimal working parameters are given in a Pareto frontier form. When the angular speed of the picking device is 58.80 rad / s, the angular speed of the film and impurity separating device is 40 rad / s, and the angle of the film and impurity separating device is 35.51, the impurity content of the residual film is 24.26%, and the picking rate of the residual film is 83.67%. And working parameters are selected in the Pareto front, so that the impurity rate of the residual film can be effectively reduced on the basis of keeping a relatively high residual film picking-up rate.

The application provides a building hourly indoor temperature prediction method and system based on a double driving mechanism, and belongs to the technical field of building energy consumption management and control. The method comprises the following steps: selecting a heat balance differential equation with a building indoor temperature as an interface variable as a benchmark, and representing a heat transfer process of an envelope structure in a space state, so as to construct a guide module for a physical model; a data driving module is constructed, an indoor temperature change difference caused by a non-principal component item is taken as a target variable, and outdoor meteorological parameters are taken as inputs, so that a back propagation neural network is constructed and trained; the outputs of the physical model guide module and the data driving module are combined, and indoor temperature prediction is carried out based on the back propagation neural network. The application can simultaneously consider the physical authenticity of the hourly indoor temperature prediction and the accuracy and rapidity of the building hourly indoor temperature prediction.

The invention discloses a photovoltaic inverter output power control algorithm based on a predictive control algorithm, which belongs to the predictive control algorithm, and comprises the step of establishing a BP neural network model, and the BP neural network model comprises the steps of determining a grid structure, designing an input layer, designing an output layer, designing a hidden layer, transmitting a function and calculating a predictive error. Historical power generation data and meteorological data of a photovoltaic power station are used as experimental samples, the prediction time is set to be 05: 00 to 19: 00 with good light reception in one day, and the prediction interval time is set to be 1 hour. The global power output by the photovoltaic power generation system is predicted and analyzed by using an algorithm based on a back propagation neural network, a network tracking prediction effect on the global output power is realized, and a distribution network transformer is effectively prevented from being burnt out due to overlarge generated power returned by the resident photovoltaic power generation system.

The invention provides a ternary protein textured protein prediction method based on machine learning. The method comprises the following steps: acquiring raw material characteristic parameters, process parameters and product characteristic parameters of three protein raw materials, namely vital gluten, yeast protein and soybean meal under a high-moisture extrusion condition; the method comprises the following steps: constructing a ternary protein coupling characteristic and a yeast protein threshold characteristic on the basis of an inhibition effect of yeast protein on wheat gluten cross-linking and a sensitive characteristic of yeast protein solubility on moisture; constructing a back-propagation neural network model, taking the raw material characteristic parameters, the process parameters, the ternary protein coupling features and the yeast protein threshold features as inputs of the back-propagation neural network model, taking the product characteristic parameters as outputs of the back-propagation neural network model, and performing iterative training on the back-propagation neural network model until convergence to obtain a final product. A textured protein prediction model is obtained; and performing forward quality prediction or reverse process derivation on a to-be-detected system by using the textured protein prediction model.

To provide a method for optimizing a safflower seed extraction process based on an ant colony optimization method and back propagation neural network.SOLUTION: A method for optimizing a safflower seed extraction process includes: extracting safflower seeds by ultrasonic waves, and conducting a single-factor experiment in which ultrasonic output, ethanol concentration, solid-liquid ratio, and extracted temperature are selected as affectors in the single-factor experiment; calculating the extraction ratio of CS and FS by using a HPLC method, and determining the optimal value of the affectors on the basis of the extraction ratio; and calculating the total evaluation value by using an entropy weighting method, and obtaining the optimal process parameter by using an ant colony optimization method and back propagation neural network model. Specific steps of the ant colony optimization method and back propagation neural network include a step of constructing a back propagation neural network, and a step of constructing the optimal neural network model of the ant colony optimization method.SELECTED DRAWING: Figure 4

A technique to prune weights of a neural network using an analytic threshold function h(w) provides a neural network having weights that have been optimally pruned. The neural network includes a plurality of layers in which each layer includes a set of weights w associated with the layer that enhance a speed performance of the neural network, an accuracy of the neural network, or a combination thereof. Each set of weights is based on a cost function C that has been minimized by back-propagating an output of the neural network in response to input training data. The cost function C is also minimized based on a derivative of the cost function C with respect to a first parameter of the analytic threshold function h(w) and on a derivative of the cost function C with respect to a second parameter of the analytic threshold function h(w).

The invention discloses a construction risk evaluation method for a tunnel crossing a strong outburst coal seam group. According to the evaluation method, 15 risk factors including geology, construction technologies and operators are included into a comprehensive evaluation framework; a fuzzy analytic hierarchy process is adopted to obtain global comprehensive weights of 15 risk factors, the weights serve as parameters to be input into a back propagation neural network, dynamic risk prediction is achieved, and an intelligent risk assessment model capable of being updated in real time is constructed. The problems of subjectivity and dependency relationship among indexes in expert judgment are effectively solved, and the risk of the complex system is dynamically simulated and predicted by the aid of powerful nonlinear mapping and learning capabilities of the system. The method significantly improves the tunnel construction risk identification precision and early warning capability in a complex geological environment, and can be widely applied to high-risk underground engineering safety management.

The invention discloses a signal coverage range detection method and device, equipment, a medium and a product. The method comprises the following steps: acquiring a sample data set; wherein the sample data set comprises an RSSI value of at least one detection unit in a detection area; constructing an initial neural network model based on Kriging interpolation and a back propagation neural network; training the initial neural network model by adopting the sample data set to generate a signal coverage detection model; the signal coverage detection model is used for detecting the signal coverage. By adopting the means of the invention, the signal coverage detection is realized by constructing the signal coverage detection model based on the Kriging interpolation and the back propagation neural network, and the efficiency and accuracy of the signal coverage detection are effectively improved.

The application provides a continuous casting round billet subsurface defect detection method and device, neural network parameters of a back propagation neural network are obtained through optimization based on a particle swarm optimization algorithm, and a defect type detection model is obtained by training the back propagation neural network by using training sample frequency domain signals including X frequency domain characteristic values and the true labels of the training sample frequency domain signal data, so that in the actual application process, after obtaining time domain signal data of a target continuous casting round billet, the time domain signal data is converted into frequency domain signal data; the frequency domain signal data is input into the defect type detection model, and the detection result of the subsurface defect of the target continuous casting round billet can be output. By converting the time domain signal to the frequency domain for more detailed analysis, the neural network model learns complex defect characteristics, and compared with manual experience flaw detection, different types of segregation defects can be more accurately identified, and misjudgment and missed detection are reduced.

The application relates to a dynamic response analysis method of a gas turbine cylinder elastic support structure, wherein the method comprises the following steps: generating a finite element simulation result of the gas turbine cylinder elastic support structure based on at least one of an isotropic loss factor, a structure size, an elastic modulus and an acting load of the gas turbine cylinder elastic support structure; constructing a firefly optimization algorithm-back propagation FA-BP neural network model; training the FA-BP neural network model by using a training set; evaluating the trained FA-BP neural network model, and generating a dynamic response result of the gas turbine cylinder elastic support structure after the evaluation result meets a preset condition. The embodiment of the application can realize training and prediction of a dynamic analysis model of an elastic support structure, provide a fast iteration method for design optimization, improve the calculation efficiency of elastic support dynamic simulation in the iteration design process, and realize a dynamic simulation method with high efficiency and precision.

The invention provides an FDM process energy efficiency modeling and multi-objective optimization method based on a BP neural network and NSGA-II, and relates to the technical field of fused deposition modeling energy efficiency prediction.The FDM process energy efficiency modeling and multi-objective optimization method comprises the steps that an actual measurement processing power curve graph is combined, energy consumption characteristics of the FDM processing process are analyzed, and an energy efficiency function and a printing efficiency function are constructed; designing a five-factor three-level test by using a Box-Behnken method, and researching key factors influencing energy consumption through variance analysis; energy consumption prediction models are respectively established through a back propagation neural network, support vector regression and response surface regression fitting, and an optimal prediction model is determined through analysis and comparison; and on the basis of the optimal prediction model, combining with the actual processing working condition, and comprehensively considering the specific energy consumption, the material deposition rate and the surface roughness, constructing a multi-objective optimization model, and solving the model by adopting an NSGA-II algorithm. According to the method, the energy consumption influence mechanism of the FDM process is researched, relatively accurate energy consumption prediction is realized, and meanwhile, theoretical support and an optimization strategy are provided for parameter optimization of the FDM process.

The invention discloses a distribution line fault detection and classification method based on a back propagation neural network, and the method comprises the steps: respectively obtaining a voltage measurement value and a current measurement value before and after a fault of a distribution line, carrying out the data preprocessing of the voltage measurement value and the current measurement value, and obtaining selected data for data training, and inputting the selected data into a pre-trained back propagation neural network for data training, extracting fault features according to a training result and analyzing a fault rule, and mapping the fault features to corresponding fault types according to a fault rule analysis result to classify the fault features, thereby obtaining a fault classification result. And obtaining a detection and classification result of the current distribution line fault. The method has the advantages that the fault features are accurately mapped to correct fault types or position output, and the accuracy of fault detection and classification is improved.

The application discloses a farm machine braking performance detector based on a Beidou satellite single point positioning mode, and belongs to the technical field of braking performance detection. The signal collector is internally provided with a satellite signal receiving module compatible with BDS and GPS, and the host is internally provided with a sequence parameter extraction module, an integral distance measurement module, an error correction module, a slope solution module and a performance evaluation module. Through the collection of three-dimensional motion data of the measured farm machine, the uncorrected braking distance is calculated by using the complex Simpson integral algorithm, the error correction distance caused by the single point positioning signal drift is obtained based on the back propagation neural network, and the farm machine braking distance is solved in combination with the slope correction value, so that the braking performance qualification condition of the measured farm machine is finally determined. The application does not need to erect a ground differential reference station and external supporting components, solves the problem of complicated field detection deployment, eliminates signal drift and road slope interference, and improves detection convenience and accuracy.

The invention relates to the field of metal-based composite materials, and provides a metal-based composite material optimization method based on dynamic reverse design, which comprises the following steps: S1, an early-stage preparation stage of a model: establishing different particle distribution configurations of a particle reinforced metal-based composite material at high flux by utilizing an Abaq us plug-in; s2, a data set construction stage: quantizing the long-range / short-range order degree of an enhancement body through a translation sequence parameter and a radial distribution function, and generating an enhancement body configuration and a random variant thereof; s3, a forward prediction stage: constructing a BPNN-CL model by adopting a back propagation neural network fused with continuous learning, and dynamically mapping the material / structure feature vectors to mechanical properties; and S4, a reverse optimization stage: chaos disturbance, dynamic partition monitoring and an elitism strategy are integrated into an NSGA-I algorithm to form an NSGA-I-PMCP optimizer, and multi-target optimization is carried out by taking BPNN-CL as a real-time agent model. The problem of reasonably designing a matrix, an interface product and a reinforcement of the particle reinforced metal matrix composite to synergistically improve the strength and the toughness of the metal matrix composite is solved.

This invention relates to a method for predicting the lethal effects of sediment-laden water flow on fish, comprising: data acquisition; model experiments; construction of a backpropagation neural network model; training of the backpropagation neural network model; model testing; and prediction application. This invention utilizes a backpropagation neural network model and an improved particle swarm optimization algorithm to predict the lethal effects of sediment-laden water flow on fish. By adjusting the backpropagation neural network model through multiple error calculations, the prediction accuracy is improved, keeping the prediction error within ±6%. This invention can be used to assess the impact of high-sediment-laden water flow processes on fish during reservoir sediment discharge and dam removal. Compared to existing assessment methods such as SI and SEV, this invention can comprehensively consider the influence of various environmental factors, significantly improving prediction accuracy and providing a new means and basis for assessing the aquatic ecological impact of reservoir sediment discharge.

The invention relates to the field of motor control, in particular to a motor temperature control method, device and equipment and a storage medium, and aims to accurately control the temperature of a joint motor in real time. The method comprises the following steps: acquiring a plurality of temperature signals of a plurality of target devices in a joint motor; a weighted temperature value is obtained according to the multiple temperature signals through a pre-trained neural network, and the neural network is a back propagation neural network optimized through a genetic algorithm; and adjusting the output torque of the joint motor according to the weighted temperature value through a pre-established temperature torque mapping model so as to control the temperature of the joint motor.