Temperature prediction method and device

Abstract

The invention relates to the technical field of soft measurement, and discloses a temperature prediction method and equipment, and the method comprises the steps: obtaining a current variable value of a temperature influence variable, and the temperature influence variable comprises a first variable, a second variable and a third variable; inputting the current variable values of the first variable, the second variable and the third variable into a first model for feature extraction to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable and a third feature vector corresponding to the third variable; and inputting the first feature vector, the second feature vector and the third feature vector into a second model, weighting the first feature vector, the second feature vector and the third feature vector by the second model, and outputting a predicted target temperature based on the weighted feature vectors. According to the invention, weighting is carried out for different features, so that features with higher importance degree can be concerned and features with low importance degree can be ignored during temperature prediction, and the prediction precision is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of soft sensing, and particularly to a temperature prediction method and device. Background Art

[0002] In modern industrial processes, in order to ensure production quality and better conduct industrial production, it is necessary to predict the temperature during the production process. In the related art, currently, the bidirectional long short-term memory network model has the best effect in temperature prediction, but this method has certain limitations, resulting in insufficient prediction accuracy and the need for improvement. Summary of the Invention

[0003] In view of this, the present disclosure provides a temperature prediction method and device to solve the problem of low temperature prediction accuracy.

[0004] In a first aspect, the present disclosure provides a temperature prediction method, which includes:

[0005] Obtain the current variable values of temperature influence variables, where the temperature influence variables include a first variable that is non-redundantly related to temperature, a second variable that is redundantly related to temperature, and a third variable that is not redundantly related to temperature;

[0006] Input the current variable values of the first variable, the second variable, and the third variable into a first model for feature extraction to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable, and a third feature vector corresponding to the third variable;

[0007] Input the first feature vector, the second feature vector, and the third feature vector into a second model, and the second model weights the first feature vector, the second feature vector, and the third feature vector, and based on the weighted feature vectors, outputs the predicted target temperature.

[0008] In the embodiments of the present disclosure, by obtaining the current variable values of temperature influence variables, where the temperature influence variables include a first variable that is non-redundantly related to temperature, a second variable that is redundantly related to temperature, and a third variable that is not redundantly related to temperature; inputting the current variable values of the first variable, the second variable, and the third variable into a first model for feature extraction to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable, and a third feature vector corresponding to the third variable; inputting the first feature vector, the second feature vector, and the third feature vector into a second model, and the second model weights the first feature vector, the second feature vector, and the third feature vector, and based on the weighted feature vectors, outputs the predicted target temperature. Since the embodiments of the present disclosure weight different features, it is possible to focus on features with higher importance and ignore features with lower importance during temperature prediction, thereby improving the prediction accuracy.

[0009] In an alternative embodiment, the first variable, the second variable, and the third variable are obtained by partitioning based on the following method:

[0010] Based on the historical variable values and historical temperatures of the temperature-influencing variables, determine the mutual information values between each temperature-influencing variable and the temperature;

[0011] Based on the mutual information values, partition the first variable, the second variable, and the third variable from the temperature-influencing variables.

[0012] In the embodiments of the present disclosure, by calculating the mutual information values between the temperature-influencing variables and the temperature, and partitioning the first variable, the second variable, and the third variable from the temperature-influencing variables based on the mutual information values, variables with strong correlation with the temperature can be screened out, the quality of the input data can be improved, and the model complexity and computational amount can be reduced.

[0013] In an alternative embodiment, partitioning the first variable, the second variable, and the third variable from the temperature-influencing variables based on the mutual information values includes:

[0014] For any temperature-influencing variable, if the mutual information value between the temperature-influencing variable and the temperature is not greater than the first threshold, then classify the temperature-influencing variable as the third variable; if the mutual information value between the temperature-influencing variable and the temperature is greater than the first threshold, then classify the temperature-influencing variable as a candidate variable;

[0015] Partition the first variable and the second variable from the candidate variables.

[0016] In the embodiments of the present disclosure, by classifying the temperature-influencing variables into the third variable and candidate variables according to the relationship between the mutual information value and the first threshold, variables with strong correlation with the temperature can be screened out, the quality of the input data can be improved, and the model complexity and computational amount can be reduced.

[0017] In an alternative embodiment, partitioning the first variable and the second variable from the candidate variables includes:

[0018] For any target variable among the candidate variables, determine the average value of the mutual information values between the target variable and other candidate variables;

[0019] Based on the mutual information value between the target variable and the temperature and the average value of the mutual information values between the target variable and other candidate variables, determine the minimum redundancy maximum correlation score of the target variable;

[0020] Classify the variables with the minimum redundancy maximum correlation score higher than the second threshold as the first variable, and classify the variables with the minimum redundancy maximum correlation score not higher than the second threshold as the second variable.

[0021] In the embodiments of the present disclosure, by dividing candidate variables into first variables and second variables according to the relationship between the minimum redundancy maximum correlation score and the second threshold, variables with strong correlation with temperature can be screened out, the quality of input data can be improved, and the model complexity and computational amount can be reduced.

[0022] In an alternative embodiment, the first model includes feature extraction networks corresponding one-to-one to the first variable, the second variable, and the third variable. Inputting the current variable values of the first variable, the second variable, and the third variable into the first model for feature extraction includes:

[0023] Inputting the current variable values of the first variable, the second variable, and the third variable into their respective corresponding feature extraction networks, performing feature extraction by the feature extraction networks, and performing non-linear transformation on the extracted features to obtain a first feature vector, a second feature vector, and a third feature vector.

[0024] In the embodiments of the present disclosure, by performing feature extraction on input variables and performing non-linear transformation on the extracted features, the ability of the model to fit complex non-linear relationships can be improved, and the prediction accuracy can be enhanced.

[0025] In an alternative embodiment, the first model is trained based on the following method:

[0026] Obtain a sample data set, where the sample data set includes sample variable values of temperature influence variables, and the sample data set has a first sample label for characterizing the actual temperature;

[0027] Input the sample variable values of the first variable, the second variable, and the third variable into the first model, perform feature extraction by the first model to obtain a first sample feature vector corresponding to the first variable, a second sample feature vector corresponding to the second variable, and a third sample feature vector corresponding to the third variable, and output a first predicted temperature based on the first sample feature vector, the second sample feature vector, and the third sample feature vector;

[0028] According to the difference between the first predicted temperature and the actual temperature characterized by the first sample label, adjust the model parameters of the first model.

[0029] In the embodiments of the present disclosure, by adjusting the model parameters of the first model according to the difference between the first predicted temperature and the actual temperature characterized by the first sample label, the performance of the first model can be optimized, and the accuracy of non-linear transformation can be improved.

[0030] In an alternative embodiment, the model parameters of the first model include weights and biases. During the process of training the first model, the method further includes:

[0031] Taking the weights and biases as the position of the first particle, based on the particle swarm optimization algorithm, by adjusting the position of the first particle, when the difference between the first predicted temperature and the actual temperature is lower than the first threshold, the model parameters corresponding to the first model are determined.

[0032] In the embodiments of the present disclosure, by determining the model parameters of the first model based on the particle swarm optimization algorithm, the performance of the first model can be optimized, and the accuracy of the non-linear conversion can be improved.

[0033] In an alternative embodiment, the second model is trained based on the following method:

[0034] Inputting the first sample feature vector, the second sample feature vector, and the third sample feature vector into the second model, and outputting the second predicted temperature by the second model;

[0035] Adjusting the model parameters of the second model according to the difference between the second predicted temperature and the actual temperature represented by the first sample label.

[0036] In the embodiments of the present disclosure, by adjusting the model parameters of the second model according to the difference between the second predicted temperature and the actual temperature represented by the first sample label, the performance of the second model can be optimized, and the temperature prediction accuracy can be improved.

[0037] In an alternative embodiment, the second model is a bidirectional long short-term memory model integrating an attention mechanism. The model parameters of the second model include the number of hidden layer nodes, the initial learning rate, and the regularization coefficient. During the process of training the second model, the method further includes:

[0038] Taking the number of hidden layer nodes, the initial learning rate, and the regularization coefficient as the position of the second particle, based on the particle swarm optimization algorithm, by adjusting the position of the second particle, when the difference between the second predicted temperature and the actual temperature is lower than the second threshold, the model parameters corresponding to the second model are determined.

[0039] In the embodiments of the present disclosure, by determining the model parameters of the second model based on the particle swarm optimization algorithm, the performance of the second model can be optimized, and the temperature prediction accuracy can be improved.

[0040] In a second aspect, the present disclosure provides a temperature prediction device, and the device includes:

[0041] An acquisition module, configured to acquire the current variable value of the temperature influence variable, where the temperature influence variable includes a first variable that is non-redundantly related to the temperature, a second variable that is redundantly related to the temperature, and a third variable that is not redundantly related to the temperature;

[0042] An extraction module for inputting the current variable values of a first variable, a second variable, and a third variable into a first model for feature extraction to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable, and a third feature vector corresponding to the third variable;

[0043] An output module for inputting the first feature vector, the second feature vector, and the third feature vector into a second model, where the second model weights the first feature vector, the second feature vector, and the third feature vector, and outputs a predicted target temperature based on the weighted feature vectors.

[0044] In a third aspect, the present disclosure provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the temperature prediction method according to the first aspect or any corresponding embodiment thereof.

[0045] In a fourth aspect, the present disclosure provides a computer-readable storage medium storing computer instructions for causing a computer to perform the temperature prediction method according to the first aspect or any corresponding embodiment thereof.

[0046] In a fifth aspect, the present disclosure provides a computer program product including computer instructions for causing a computer to perform the temperature prediction method according to the first aspect or any corresponding embodiment thereof. Description of the Drawings

[0047] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 is a network model architecture diagram of the temperature prediction method according to an embodiment of the present disclosure;

[0049] Figure 2 is a flowchart of the temperature prediction method according to an embodiment of the present disclosure;

[0050] Figure 3 is a flowchart of the temperature prediction method according to another embodiment of the present disclosure;

[0051] Figure 4 is a prediction result diagram of the simulation of an actual case by the long short-term memory network model according to an embodiment of the present disclosure;

[0052] Figure 5 It is a prediction result graph of the actual case simulation by the deep extreme learning machine network model based on the particle swarm optimization algorithm according to the embodiments of the present disclosure;

[0053] Figure 6 It is a prediction result graph of the actual case simulation by the bidirectional long short-term memory network model based on the particle swarm optimization algorithm according to the embodiments of the present disclosure;

[0054] Figure 7 It is a prediction result graph of the actual case simulation by the bidirectional long short-term memory network model based on non-linear expansion and parallel input according to the embodiments of the present disclosure;

[0055] Figure 8 It is a prediction result graph of the actual case simulation by the bidirectional long short-term memory network model based on distributed non-linear expansion and parallel input according to the embodiments of the present disclosure;

[0056] Figure 9 It is a prediction result graph of the actual case simulation by the network model according to the embodiments of the present disclosure;

[0057] Figure 10 It is a structural block diagram of the temperature prediction device according to the embodiments of the present disclosure;

[0058] Figure 11 It is a schematic hardware structure diagram of the computer device according to the embodiments of the present disclosure. Detailed implementation manners

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0060] In modern industrial processes, in order to ensure production quality and better carry out industrial production, it is necessary to predict the temperature during the production process. For example, during the extraction of honeysuckle, it is necessary to predict the temperature inside the honeysuckle extraction tank. Currently, in some technologies, using a bidirectional long short-term memory network model for temperature prediction has the best effect, but this method has certain limitations. First, a single bidirectional long short-term memory network model cannot focus on data with greater influence during prediction, resulting in poor prediction effects. Second, in modern industrial processes, multiple process variables are usually involved, and these variables usually exhibit highly complex non-linear relationships. For example, the changes in parameters such as temperature, pressure, and flow rate do not show a linear response, but rather affect each other in a complex and non-intuitive way, which also leads to poor prediction effects. In addition, the difference between input variables and output variables will also affect the prediction accuracy. Related technologies have poor prediction effects because they cannot focus on data with greater influence during temperature prediction, and do not consider the non-linear relationships between variables and the differences between input variables and output variables.

[0061] To solve the above problems, according to an embodiment of the present disclosure, an embodiment of a temperature prediction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0062] Before introducing the method of the present disclosure, a network model architecture is first introduced, as Figure 1 shown, Figure 1 is a network model architecture diagram of the temperature prediction method according to an embodiment of the present disclosure.

[0063] Based on Figure 1 the network model architecture shown, the present disclosure provides a temperature prediction method, taking the prediction of the temperature inside the honeysuckle extraction tank as an example for illustration, as Figure 2 shown, Figure 2 is a flowchart of the temperature prediction method according to an embodiment of the present disclosure. This process can be applied to a server and includes the following steps:

[0064] Step S201, obtain the current variable values of temperature influencing variables. The temperature influencing variables include a first variable that is non-redundantly related to temperature, a second variable that is redundantly related to temperature, and a third variable that is not redundantly related to temperature.

[0065] Optionally, in an embodiment of the present disclosure, taking the prediction of the temperature inside the honeysuckle extraction tank as an example, the temperature influencing variables may include valve variables, flow variables, pressure variables, time variables, signal variables, and temperature variables, etc.

[0066] Specifically, valve - type variables may include the states of valves such as upper jacket steam inlet, lower jacket steam inlet, solvent inlet valve, extraction tank liquid outlet valve, extraction tank side filtration liquid outlet valve, filter inlet liquid valve, filter outlet liquid valve, extraction liquid circulation valve, extraction liquid outlet valve, compressed air back - flushing valve, drain bypass valve, vacuum inlet valve, extraction tank discharge control valve, cooling circulating water inlet valve, filter cleaning inlet valve, filter cleaning discharge valve, filter vent valve, secondary steam valve, distillate reflux valve, aromatic water isolation valve, distillate collection valve, heat exchanger drain bypass valve, opening degrees of pre - heating steam regulating valve and extraction steam regulating valve, etc.; flow - type variables may include cumulative extraction solvent flow, primary extraction water addition amount, secondary extraction water addition amount, cumulative metered liquid outlet flow, extraction solvent flow, and metered liquid outlet flow, etc.; pressure - type variables may include purified water pre - heating steam pressure, primary heat - preservation steam pressure, secondary heat - preservation steam pressure, pressure inside the extraction tank, and primary and secondary water - adding heating steam pressures, etc.; time - type variables may include primary extraction heat - preservation time in minutes and secondary extraction heat - preservation time in minutes, etc.; signal - type variables may include automatic switch signal of liquid outlet pump, feeding completion signal, and feeding start signal, etc.; temperature - type variables may include purified water pre - heating temperature, primary heating temperature, secondary heating temperature, and temperature at the upper part of the extraction tank, etc.

[0067] The current variable value of the temperature - influencing variable refers to the values of each temperature - influencing variable in the current extraction. For example, in the current extraction, the cumulative value of the extraction solvent flow, the primary extraction water addition amount, etc.

[0068] As Figure 1 shown, the server obtains the current variable value of the temperature - influencing variable, classifies the data, and obtains the first variable that is non - redundantly related to temperature, the second variable that is redundantly related to temperature, and the third variable that is not related to temperature redundancy.

[0069] Specifically, the first variable that is non - redundantly related to temperature refers to a variable whose correlation degree with temperature is higher than the first threshold and the correlation degree between variables is lower than the second threshold; the second variable that is redundantly related to temperature refers to a variable whose correlation degree with temperature is higher than the first threshold and the correlation degree between variables is higher than the second threshold; the third variable that is not related to temperature redundancy refers to a variable whose correlation degree with temperature is lower than the first threshold.

[0070] In some alternative embodiments, the first variable, the second variable, and the third variable are obtained by the following method:

[0071] Based on the historical variable values and historical temperatures of the temperature - influencing variables, determine the mutual information values between each temperature - influencing variable and temperature;

[0072] Based on the mutual information values, divide the first variable, the second variable, and the third variable from the temperature - influencing variables.

[0073] Optionally, in the embodiments of the present disclosure, the historical variable value of the temperature influence variable is X = [x1, x2, x3, …, x i , and the historical temperature is Y. The server obtains the temperature influence variable and the historical temperature, and calculates the mutual information value I(X i ; Y) between each temperature influence variable and its corresponding historical temperature. The calculation formula of the mutual information value is as follows:

[0074] I(X i ; Y) = H(Y) - H(Y|X i )

[0075] where H(Y) represents the information entropy of the historical temperature Y, and H(Y|X i ) represents the conditional entropy of the historical temperature Y given the feature X i .

[0076] Then, the server evaluates and quantifies the dependence degree between the temperature influence variable and the temperature according to the calculated mutual information value, and divides the first variable, the second variable, and the third variable from the temperature influence variables.

[0077] In some alternative embodiments, dividing the first variable, the second variable, and the third variable from the temperature influence variables according to the mutual information value includes:

[0078] For any temperature influence variable, if the mutual information value between the temperature influence variable and the temperature is not greater than the first threshold, the temperature influence variable is divided into the third variable; if the mutual information value between the temperature influence variable and the temperature is greater than the first threshold, the temperature influence variable is divided into a candidate variable;

[0079] The first variable and the second variable are divided from the candidate variables.

[0080] Optionally, in the embodiments of the present disclosure, the server sorts the temperature influence variables according to the calculated mutual information value. For any temperature influence variable, if the mutual information value between the temperature influence variable and the temperature is not greater than the first threshold, the temperature influence variable is divided into the third variable, that is, a redundant and irrelevant feature; if the mutual information value between the temperature influence variable and the temperature is greater than the first threshold, the temperature influence variable is divided into a candidate variable, that is, a relevant feature.

[0081] Then, the server divides the candidate variables into the first variable and the second variable.

[0082] In some alternative embodiments, dividing the first variable and the second variable from the candidate variables includes:

[0083] For any target variable in the candidate variables, determine the average value of the mutual information values between the target variable and other candidate variables;

[0084] Determine the minimum redundancy maximum correlation score of the target variable based on the mutual information value between the target variable and the temperature and the average value of the mutual information values between the target variable and other candidate variables.

[0085] Divide the variables with a minimum redundancy maximum correlation score higher than the second threshold into first variables, and divide the variables with a minimum redundancy maximum correlation score not higher than the second threshold into second variables.

[0086] Optionally, in the embodiments of the present disclosure, the server calculates the average value of the mutual information values between any target variable in the candidate variables and other candidate variables, and calculates the minimum redundancy maximum correlation score of the target variable according to the mutual information value between the target variable and the temperature and the average value of the mutual information values between the target variable and other candidate variables. The calculation formula of the minimum redundancy maximum correlation score is as follows:

[0087]

[0088] where |S| represents the number of elements in the feature set S, represents the feature X k and the sum of the mutual information of all other features X j in the feature set S, represents the feature X k and the average value of the mutual information of all other features X j in the feature set S.

[0089] Then, the server sorts the candidate variables according to the calculated minimum redundancy maximum correlation score. If the minimum redundancy maximum correlation score of the candidate variable is higher than the second threshold, the candidate variable is divided into a first variable, that is, a non-redundant related feature; if the minimum redundancy maximum correlation score of the candidate variable is not higher than the second threshold, the candidate variable is divided into a second variable, that is, a redundant related feature.

[0090] In the embodiments of the present disclosure, by calculating the mutual information value between the temperature influence variable and the temperature, dividing the temperature influence variable into a third variable and a candidate variable according to the relationship between the mutual information value and the first threshold, and dividing the candidate variable into a first variable and a second variable according to the relationship between the minimum redundancy maximum correlation score and the second threshold, variables with strong correlation with the temperature can be screened out, the quality of the input data can be improved, and the model complexity and calculation amount can be reduced.

[0091] Step S202, input the current variable values of the first variable, the second variable, and the third variable into the first model for feature extraction to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable, and a third feature vector corresponding to the third variable.

[0092] Optionally, in the embodiments of the present disclosure, the first model adopted by the server is a deep extreme learning machine.

[0093] Specifically, a deep extreme learning machine is a deep learning network, which is composed of multiple extreme learning machine autoencoders stacked layer by layer. Assuming that the input layer, hidden layer, and output layer of the extreme learning machine have N, L, and 1 neurons respectively, the output of the extreme learning machine model of a neural network with a single hidden layer is:

[0094]

[0095] where, β i represents the output weight between the neurons of the i-th hidden layer and the output layer, g(x) is the activation function, ω i represents the connection weight between the neurons of the hidden layer and the i-th input layer neuron, and v j represents the threshold of the hidden layer neuron.

[0096] In addition to the connection weight ω i and the threshold v j the extreme learning machine can also identify the output matrix H of the hidden layer. Therefore, the training of the extreme learning machine model can be transformed into finding the output weight β i :

[0097] β i = H + T

[0098] where, H + represents the generalized inverse matrix of the hidden layer output matrix H, and T represents the target output matrix of the network samples, that is, the result that the model is expected to output in the training data.

[0099] The deep extreme learning machine is a special version of the extreme learning machine. The weight based on the deep extreme learning machine is expressed as follows:

[0100]

[0101] where, β represents the connection weight between the nodes of the output layer and the hidden layer, C represents the regularization parameter, X represents the input data, N represents the number of nodes in the output layer, represents the number of nodes in the hidden layer.

[0102] In some alternative embodiments, the first model includes a feature extraction network corresponding one-to-one to the first variable, the second variable, and the third variable. Inputting the current variable values of the first variable, the second variable, and the third variable into the first model for feature extraction includes:

[0103] Input the current variable values of the first variable, the second variable, and the third variable into their respective feature extraction networks. The feature extraction networks perform feature extraction and non-linearly transform the extracted features to obtain the first feature vector, the second feature vector, and the third feature vector.

[0104] Optionally, in the embodiments of the present disclosure, as Figure 1 shown, the server converts the sub-process variable spaces of the three distributions of the first variable, the second variable, and the third variable into non-linear spaces respectively through the last hidden layer of the deep extreme learning machine, and regards the output of the last hidden layer as a distributed non-linear expansion to obtain three distributed sub-inputs with independent inputs.

[0105] Specifically, the server inputs the sub-process variable spaces of the three distributions into the deep extreme learning machine respectively, and the outputs of the last hidden layers of the three deep extreme learning machines are respectively:

[0106]

[0107] where g(·) represents the sigmoid function;

[0108] and respectively represent the w N ×j-dimensional input weight and bias vector of the non-redundant relevant feature X N input into the last hidden layer of the deep extreme learning machine, and k1 is the number of hidden layer nodes of the last hidden layer of the deep extreme learning machine of the non-redundant relevant feature X N ;

[0109] and respectively represent the w RR ×l-dimensional input weight and bias vector of the redundant relevant feature X RR input into the last hidden layer of the deep extreme learning machine, and k2 is the number of hidden layer nodes of the last hidden layer of the deep extreme learning machine of the redundant relevant feature X RR ;

[0110] and respectively represent the w RN ×n-dimensional input weight and bias vector of the redundant irrelevant feature X RN input into the last hidden layer of the deep extreme learning machine, and k3 is the number of hidden layer nodes of the last hidden layer of the deep extreme learning machine of the redundant irrelevant feature X RN ;

[0111] In the embodiments of the present disclosure, by extracting features from input variables and performing non-linear transformation on the extracted features, the ability of the model to fit complex non-linear relationships can be improved, and the prediction accuracy can be enhanced.

[0112] In some alternative embodiments, the first model is trained based on the following method:

[0113] Obtain a sample data set, the sample data set includes sample variable values of temperature influence variables, and the sample data set has a first sample label for characterizing the actual temperature;

[0114] Input the sample variable values of the first variable, the second variable, and the third variable into the first model, and the first model performs feature extraction to obtain a first sample feature vector corresponding to the first variable, a second sample feature vector corresponding to the second variable, and a third sample feature vector corresponding to the third variable, and based on the first sample feature vector, the second sample feature vector, and the third sample feature vector, output a first predicted temperature;

[0115] According to the difference between the first predicted temperature and the actual temperature characterized by the first sample label, adjust the model parameters of the first model.

[0116] Optionally, in the embodiments of the present disclosure, the server obtains the sample variable values and the actual temperature of the temperature influence variables during the honeysuckle extraction process to obtain a sample data set.

[0117] Specifically, the server inputs the sample variable values of the first variable, the second variable, and the third variable into the first model for feature extraction to obtain a first sample feature vector corresponding to the first variable, a second sample feature vector corresponding to the second variable, and a third sample feature vector corresponding to the third variable, and based on the first sample feature vector, the second sample feature vector, and the third sample feature vector, output a first predicted temperature.

[0118] Then, the server obtains a first training error according to the difference between the first predicted temperature and the actual temperature characterized by the first sample label, and adjusts the model parameters of the first model according to the first training error.

[0119] In the embodiments of the present disclosure, by adjusting the model parameters of the first model according to the difference between the first predicted temperature and the actual temperature characterized by the first sample label, the performance of the first model can be optimized, and the accuracy of non-linear transformation can be improved.

[0120] In some alternative embodiments, the model parameters of the first model include weights and biases. During the process of training the first model, the method further includes:

[0121] Taking the weights and biases as the position of the first particle, based on the particle swarm optimization algorithm, by adjusting the position of the first particle, when the difference between the first predicted temperature and the actual temperature is lower than the first threshold, the model parameters corresponding to the first model are determined.

[0122] Optionally, in the embodiments of the present disclosure, the server adjusts the weights and biases of the first model based on the particle swarm optimization algorithm, so that the difference between the first predicted temperature and the actual temperature is lower than the first threshold.

[0123] Specifically, the server first takes the first training error as the fitness value, and takes the weights e and biases b of the deep extreme learning machine c as the position of the first particle, then determines whether the position of the first particle reaches the optimum. If it reaches the optimum, the step size is updated and the position of the first particle is updated. If it does not reach the optimum, the position of the first particle is directly updated. Then, the number of iterations is judged. If the maximum number of iterations is not reached, the iteration continues. If the maximum number of iterations is reached, the optimal weights e and biases b of the first model are obtained. c , at this time, the difference between the first predicted temperature and the actual temperature is lower than the first threshold, and the first model is obtained.

[0124] In the embodiments of the present disclosure, by determining the model parameters of the first model based on the particle swarm optimization algorithm, the performance of the first model can be optimized and the accuracy of the non-linear conversion can be improved.

[0125] Step S203: Input the first feature vector, the second feature vector, and the third feature vector into the second model. The second model weights the first feature vector, the second feature vector, and the third feature vector, and outputs the predicted target temperature based on the weighted feature vectors.

[0126] Optionally, in the embodiments of the present disclosure, as Figure 1 shown, the second model adopted by the server is a bidirectional long short-term memory network.

[0127] The bidirectional long short-term memory network model can be regarded as a combination of bidirectional long short-term memory networks. Assuming that X t is the input sequence at the current time step and h t is the current cell state, the long short-term memory network mainly consists of a memory cell c, an input gate i, a forget gate f, and an output gate o:

[0128]

[0129] i t =σ(W i [h t-1 ,x t +b i )

[0130] ft = σ(W f [h t-1 , x t + b f )

[0131] o t = σ(W o [h t-1 , x t + b o )

[0132] h t = o t tanhC t

[0133]

[0134] where W C , W i , W f , W o are weight matrices, σ(·) represents the Sigmoid activation function of the hidden layer, b c , b i , b f , b o are bias vectors.

[0135] The bidirectional long short-term memory network is an improved sequence processing model composed of two long short-term memory networks. One long short-term memory network processes the input from front to back, and the other long short-term memory network processes the input from back to front. The structure of the bidirectional long short-term memory network is as follows:

[0136]

[0137] where f and b represent the forward and backward hidden states respectively, are the weights of the model, is the bias of the model. The hidden state is realized during the integration process. Then, the server sends H t to the output layer to obtain the final output:

[0138]

[0139] O t = ω q H t + b q

[0140] where ω q represents the weight of the model, and b q represents the bias of the output layer.

[0141] Specifically, the server inputs the first feature vector, the second feature vector, and the third feature vector into the second model for weighting, assigns different weights respectively, and obtains the output of the attention layer of the second model:

[0142]

[0143] Among them, α1, α2, and α3 represent the weight coefficients in the attention mechanism, and h1, h2, and h3 represent the hidden layer output vectors of different parts in the bidirectional long short-term memory network model.

[0144] Then, the server inputs the weighted feature vectors into the bidirectional long short-term memory network and outputs the predicted target temperature y(t):

[0145]

[0146] y(t) = ω·h t + b

[0147] Among them, ω and b are the weight matrix and bias vector of the fully connected layer respectively, h t and are the output vector and input vector of the bidirectional long short-term memory network model respectively, and f(·) represents the calculation process of the long short-term memory network.

[0148] In some optional embodiments, the second model is trained based on the following method:

[0149] Input the first sample feature vector, the second sample feature vector, and the third sample feature vector into the second model, and the second model outputs the second predicted temperature;

[0150] According to the difference between the second predicted temperature and the actual temperature represented by the first sample label, adjust the model parameters of the second model.

[0151] Optionally, in the embodiments of the present disclosure, the server inputs the first sample feature vector, the second sample feature vector, and the third sample feature vector into the second model and outputs the second predicted temperature.

[0152] Then, the server obtains the second training error according to the difference between the second predicted temperature and the actual temperature represented by the first sample label, and adjusts the model parameters of the second model according to the second training error.

[0153] In the embodiments of the present disclosure, by adjusting the model parameters of the second model according to the difference between the second predicted temperature and the actual temperature represented by the first sample label, the performance of the second model can be optimized and the temperature prediction accuracy can be improved.

[0154] In some alternative embodiments, the second model is a bidirectional long short-term memory model integrated with an attention mechanism. The model parameters of the second model include the number of hidden layer nodes, the initial learning rate, and the regularization coefficient. During the process of training the second model, the method further includes:

[0155] Taking the number of hidden layer nodes, the initial learning rate, and the regularization coefficient as the position of the second particle, and based on the particle swarm optimization algorithm, by adjusting the position of the second particle, determining the model parameters corresponding to the second model when the difference between the second predicted temperature and the actual temperature is lower than the second threshold.

[0156] Optionally, in the embodiments of the present disclosure, the server adjusts the number of hidden layer nodes, the initial learning rate, and the regularization coefficient of the second model based on the particle swarm optimization algorithm, such that the difference between the second predicted temperature and the actual temperature is lower than the second threshold.

[0157] Specifically, the server first takes the second training error as the fitness value, takes the number of hidden layer nodes, the initial learning rate, and the regularization coefficient of the bidirectional long short-term memory network as the position of the second particle, then determines whether the position of the second particle reaches the optimum. If it reaches the optimum, the step size is updated and the position of the second particle is updated. If it does not reach the optimum, the position of the second particle is directly updated. Then, the number of iterations is determined. If the maximum number of iterations is not reached, the iteration continues. If the maximum number of iterations is reached, the optimum number of hidden layer nodes, the initial learning rate, and the regularization coefficient of the second model are obtained. At this time, the difference between the second predicted temperature and the actual temperature is lower than the second threshold, and the second model is obtained.

[0158] During the iteration process, the server first updates the particle velocity v i (n) and the particle position x i (n):

[0159] v i (n) = c1r1(P ibest (n - 1) - x i (n - 1)) + c2r2(P gbest (n - 1) - x i (n - 1)) + ωv i (n - 1)

[0160] x i (n) = x i (n - 1) + v i (n)

[0161] where P ibest and P gbest respectively represent the individual optimum position and the global optimum position, c1 and c2 are acceleration constants, r1 and r2 are random numbers, and ω is the inertia weight, which can be expressed as:

[0162]

[0163] Among them, n represents the current iteration number, n max represents the total number of iterations.

[0164] Then, the server calculates the fitness function MSE:

[0165]

[0166] Among them, M represents the number of samples, p r represents the true value, represents the predicted value.

[0167] Next, the server updates the individual optimal solution MSE according to the fitness function MSE i,best(n) :

[0168] MSE i,best(n) = min[MSE i(n) , MSE i,best(n-1)

[0169] Among them, MSE i,best(n) is updated by comparing the MSE of the current iteration and the previous iteration i(n) and MSE i,best(n-1) and taking the smaller value of the two.

[0170] Then, the server updates the global optimal solution P according to the individual optimal solution gbest(n) :

[0171]

[0172] Finally, the server obtains the optimal number of hidden layer nodes, the initial learning rate, and the regularization coefficient of the second model according to the global optimal solution.

[0173] In the embodiments of the present disclosure, by determining the model parameters of the second model based on the particle swarm optimization algorithm, the performance of the second model can be optimized, and the temperature prediction accuracy can be improved.

[0174] ​In an embodiment of the present disclosure, by obtaining the current variable value of the temperature influence variable, the temperature influence variable includes a first variable that is non-redundantly related to temperature, a second variable that is redundantly related to temperature, and a third variable that is not redundantly related to temperature; inputting the current variable values of the first variable, the second variable, and the third variable into a first model for feature extraction to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable, and a third feature vector corresponding to the third variable; inputting the first feature vector, the second feature vector, and the third feature vector into a second model, and the second model weights the first feature vector, the second feature vector, and the third feature vector, and based on the weighted feature vectors, outputs the predicted target temperature. Since the embodiments of the present disclosure weight different features, it is possible to focus on features with higher importance and ignore features with lower importance during temperature prediction, thereby improving the prediction accuracy.

[0175] In some alternative embodiments, as Figure 3 shown, Figure 3 is a schematic flowchart of a temperature prediction method according to another embodiment of the present disclosure.

[0176] During the offline training process, the server first preprocesses the temperature influence variables in the training set to obtain the classified temperature influence variables, then non-linearly expands the classified temperature influence variables to obtain three distributed sub-inputs with independent inputs, and then inputs the three distributed sub-inputs into the second model to obtain the predicted target temperature, and determines the network hyperparameters of the second model based on the particle swarm optimization algorithm.

[0177] During the online testing process, the server first preprocesses the temperature influence variables in the test set to obtain the classified temperature influence variables, then non-linearly expands the classified temperature influence variables to obtain three distributed sub-inputs with independent inputs, and then inputs the three distributed sub-inputs into the second model optimized by the particle swarm optimization algorithm to obtain the predicted output, and evaluates the performance of the model according to the predicted output.

[0178] In some alternative embodiments, the server uses MATLA software to simulate the model to obtain the prediction results and analyze the model accuracy. The magnitude of the model accuracy represents the prediction accuracy of the prediction model. When the model accuracy reaches the preset value, the model can be used for actual temperature prediction of the extraction tank in the actual honeysuckle extraction process in subsequent applications. In this embodiment, the mean square error, root mean square error, and coefficient of determination are used to calculate the model accuracy. The formula for the mean square error is as follows:

[0179]

[0180] The formula for the root mean square error is as follows:

[0181]

[0182] The formula for the coefficient of determination is as follows:

[0183]

[0184] Where N test represents the total number of samples in the test dataset, y i,test represents the true value of the i-th sample in the test set, i.e., the actual temperature, represents the predicted value of the i-th sample in the test set, i.e., the predicted temperature, represents the average value of y i,test .

[0185] To verify the feasibility of the model proposed in this disclosure, the server compared the long short-term memory network model, the deep extreme learning machine network model based on the particle swarm optimization algorithm, the bidirectional long short-term memory network model based on the particle swarm optimization algorithm, the bidirectional long short-term memory network model based on non-linear extension and parallel input, the bidirectional long short-term memory network model based on distributed non-linear extension and parallel input with the model of this disclosure respectively.

[0186] The server used MATLAB software to simulate and verify each model, calculated the mean square error, root mean square error and coefficient of determination of each model respectively, so as to obtain the prediction accuracy of each model. Among them, the values of the mean square error and the root mean square error are negatively correlated with the prediction accuracy of the model, that is, the lower the values of the mean square error and the root mean square error, the higher the prediction accuracy. In addition, the coefficient of determination is an important index to evaluate the fitting degree between the actual output and the predicted output. The higher the value of the coefficient of determination, the closer it is to 1, indicating that the consistency between the predicted output value and the expected output value is better and the model accuracy is higher.

[0187] The server used the test samples to simulate each model and examined the prediction accuracy of each model after training. As shown in Table 1, the mean square error of the model of this disclosure is 0.00025, the root mean square error is 0.0159, and the coefficient of determination is 0.975. Compared with other methods, the coefficient of determination is the highest, and the mean square error and the root mean square error are the lowest, indicating that this method is significantly superior to other methods in performance and has the best prediction performance.

[0188] Table 1 Prediction performance of different models

[0189]

[0190] As Figures 4 - 9 shown, Figure 4 is the prediction result diagram of the long short-term memory network model according to the embodiment of this disclosure for the actual case simulation; Figure 5It is a prediction result graph of the actual case simulation by the deep extreme learning machine network model based on the particle swarm optimization algorithm according to the embodiments of the present disclosure; Figure 6 It is a prediction result graph of the actual case simulation by the bidirectional long short-term memory network model based on the particle swarm optimization algorithm according to the embodiments of the present disclosure; Figure 7 It is a prediction result graph of the actual case simulation by the bidirectional long short-term memory network model based on non-linear extension and parallel input according to the embodiments of the present disclosure; Figure 8 It is a prediction result graph of the actual case simulation by the bidirectional long short-term memory network model based on distributed non-linear extension and parallel input according to the embodiments of the present disclosure; Figure 9 It is a prediction result graph of the actual case simulation by the network model according to the embodiments of the present disclosure.

[0191] It can be seen from Figures 4 - 9 comparing the prediction curves of different models that the prediction curve of the model of the present disclosure has the best fitting effect with the actual curve, and the predicted value is closest to the true value, indicating that the prediction ability and robustness of this method are the best.

[0192] In this embodiment, a temperature prediction device is also provided. This device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0193] This embodiment provides a temperature prediction device, as Figure 10 shown, including:

[0194] An acquisition module 1001, configured to acquire the current variable values of temperature influence variables, where the temperature influence variables include a first variable that is non-redundantly related to temperature, a second variable that is redundantly related to temperature, and a third variable that is not redundantly related to temperature;

[0195] An extraction module 1002, configured to input the current variable values of the first variable, the second variable, and the third variable into a first model for feature extraction, to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable, and a third feature vector corresponding to the third variable;

[0196] An output module 1003, configured to input the first feature vector, the second feature vector, and the third feature vector into a second model, and the second model weights the first feature vector, the second feature vector, and the third feature vector, and based on the weighted feature vectors, outputs the predicted target temperature.

[0197] In an embodiment of the present disclosure, by obtaining the current variable value of a temperature influence variable, the temperature influence variable includes a first variable that is non-redundantly related to temperature, a second variable that is redundantly related to temperature, and a third variable that is not redundantly related to temperature; inputting the current variable values of the first variable, the second variable, and the third variable into a first model for feature extraction to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable, and a third feature vector corresponding to the third variable; inputting the first feature vector, the second feature vector, and the third feature vector into a second model, and the second model weights the first feature vector, the second feature vector, and the third feature vector, and based on the weighted feature vectors, outputs a predicted target temperature. Since the embodiment of the present disclosure weights different features, it is possible to focus on features with higher importance and ignore features with lower importance during temperature prediction, thereby improving the prediction accuracy.

[0198] In some alternative embodiments, the obtaining module 1001 includes:

[0199] A determining sub-module, configured to determine the mutual information value between each temperature influence variable and temperature according to the historical variable value and historical temperature of the temperature influence variable;

[0200] A dividing sub-module, configured to divide the first variable, the second variable, and the third variable from the temperature influence variables according to the mutual information value.

[0201] In some alternative embodiments, the dividing sub-module includes:

[0202] A first dividing unit, for any temperature influence variable, if the mutual information value between the temperature influence variable and temperature is not greater than a first threshold, dividing the temperature influence variable into a third variable, and if the mutual information value between the temperature influence variable and temperature is greater than the first threshold, dividing the temperature influence variable into a candidate variable;

[0203] A second dividing unit, configured to divide the first variable and the second variable from the candidate variables.

[0204] In some alternative embodiments, the second dividing unit includes:

[0205] A first determining sub-unit, for any target variable in the candidate variables, determining the average value of the mutual information values between the target variable and other candidate variables;

[0206] A second determining sub-unit, configured to determine the minimum redundancy maximum correlation score of the target variable based on the mutual information value between the target variable and temperature and the average value of the mutual information values between the target variable and other candidate variables;

[0207] A sub - division unit, configured to divide variables with the minimum redundancy maximum correlation score higher than a second threshold into first variables, and divide variables with the minimum redundancy maximum correlation score not higher than the second threshold into second variables.

[0208] In some alternative embodiments, the extraction module 1002 includes:

[0209] A conversion sub - module, configured to input the current variable values of the first variable, the second variable, and the third variable into their respective feature extraction networks, perform feature extraction by the feature extraction networks, and perform non - linear conversion on the extracted features to obtain a first feature vector, a second feature vector, and a third feature vector.

[0210] In some alternative embodiments, the extraction module 1002 includes:

[0211] An acquisition sub - module, configured to acquire a sample data set, where the sample data set includes sample variable values of temperature - influencing variables, and the sample data set has a first sample label for characterizing the actual temperature;

[0212] A first output sub - module, configured to input the sample variable values of the first variable, the second variable, and the third variable into a first model, perform feature extraction by the first model to obtain a first sample feature vector corresponding to the first variable, a second sample feature vector corresponding to the second variable, and a third sample feature vector corresponding to the third variable, and output a first predicted temperature based on the first sample feature vector, the second sample feature vector, and the third sample feature vector;

[0213] A first adjustment sub - module, configured to adjust the model parameters of the first model according to the difference between the first predicted temperature and the actual temperature characterized by the first sample label.

[0214] In some alternative embodiments, the apparatus further includes:

[0215] A first determination module, configured to use the weights and biases as the position of a first particle, and based on the particle swarm optimization algorithm, determine the model parameters corresponding to the first model when the difference between the first predicted temperature and the actual temperature is lower than a first threshold by adjusting the position of the first particle.

[0216] In some alternative embodiments, the output module 1003 includes:

[0217] A second output sub - module, configured to input the first sample feature vector, the second sample feature vector, and the third sample feature vector into a second model, and output a second predicted temperature by the second model;

[0218] A second adjustment sub - module, configured to adjust the model parameters of the second model according to the difference between the second predicted temperature and the actual temperature characterized by the first sample label.

[0219] In some alternative embodiments, the apparatus further includes:

[0220] A second determination module, configured to use the number of hidden layer nodes, the initial learning rate, and the regularization coefficient as the position of a second particle, and based on the particle swarm optimization algorithm, by adjusting the position of the second particle, determine the model parameters corresponding to the second model when the difference between the second predicted temperature and the actual temperature is lower than a second threshold.

[0221] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding foregoing embodiments, and will not be elaborated herein.

[0222] The temperature prediction apparatus in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0223] This disclosure embodiment also provides a computer device having the above Figure 10 shown temperature prediction apparatus.

[0224] Please refer to Figure 11 , Figure 11 which is a schematic structural diagram of a computer device provided by an alternative embodiment of this disclosure. As Figure 11 shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if needed, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as a server array, a set of blade servers, or a multi-processor system). Figure 11 In

[0225] FIG. 24, one processor 10 is taken as an example. Processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, processor 10 can further include a hardware chip. The above hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The above programmable logic device can be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0226] Among them, the memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiments.

[0227] The memory 20 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely provided with respect to the processor 10, and these remote memories may be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0228] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 may also include a combination of the above types of memories.

[0229] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0230] The embodiments of the present disclosure also provide a computer-readable storage medium. The method according to the embodiments of the present disclosure can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be processed by such software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium may be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium may also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0231] A part of the present disclosure can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the present disclosure through the operations of the computer. Those skilled in the art should understand that the forms of existence of computer program instructions in a computer-readable medium include but are not limited to source files, executable files, installation package files, etc. Correspondingly, the ways for a computer to execute computer program instructions include but are not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible by the computer.

[0232] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A temperature prediction method, characterized in that: The method comprises: Acquire current variable values ​​of temperature-affecting variables, wherein the temperature-affecting variables include a first variable related to temperature without redundancy, a second variable related to temperature redundancy, and a third variable unrelated to temperature redundancy; Inputting the current variable values ​​of the first variable, the second variable and the third variable into a first model for feature extraction, to obtain a first feature vector corresponding to the first variable, a second feature vector corresponding to the second variable and a third feature vector corresponding to the third variable; The first eigenvector, the second eigenvector and the third eigenvector are input into a second model, the second model weights the first eigenvector, the second eigenvector and the third eigenvector, and outputs a predicted target temperature based on the weighted eigenvectors.

2. The method according to claim 1, characterized in that The first variable, the second variable and the third variable are obtained by dividing based on the following method: Determining a mutual information value between each of the temperature influencing variables and the temperature according to the historical variable value and the historical temperature of the temperature influencing variable; The first variable, the second variable and the third variable are obtained by dividing the temperature-affecting variables according to the mutual information value.

3. The method according to claim 2, characterized in that The step of dividing the temperature-affected variables according to the mutual information value to obtain the first variable, the second variable, and the third variable includes: For any of the temperature influencing variables, if the mutual information value between the temperature influencing variable and the temperature is not greater than a first threshold, the temperature influencing variable is classified as the third variable; if the mutual information value between the temperature influencing variable and the temperature is greater than the first threshold, the temperature influencing variable is classified as a candidate variable; The first variable and the second variable are divided from the candidate variables.

4. The method according to claim 3, characterized in that The step of dividing the first variable and the second variable from the candidate variables comprises: For any target variable among the candidate variables, determine an average value of mutual information values ​​between the target variable and other candidate variables; Determine the minimum redundancy maximum relevance score of the target variable based on the mutual information value between the target variable and the temperature and the average mutual information value between the target variable and other candidate variables; The variables whose minimum redundancy maximum relevance scores are higher than a second threshold are classified as the first variables, and the variables whose minimum redundancy maximum relevance scores are not higher than the second threshold are classified as the second variables.

5. The method according to claim 1, characterized in that The first model includes a feature extraction network corresponding to the first variable, the second variable and the third variable one by one, and inputting the current variable values ​​of the first variable, the second variable and the third variable into the first model for feature extraction includes: The current variable values ​​of the first variable, the second variable and the third variable are input into their respective corresponding feature extraction networks, the feature extraction network performs feature extraction, and the extracted features are nonlinearly transformed to obtain the first feature vector, the second feature vector and the third feature vector.

6. The method according to claim 1 or 5, characterized in that: The first model is trained based on the following method: Acquire a sample data set, wherein the sample data set includes sample variable values ​​of the temperature-affecting variable, and the sample data set has a first sample label for characterizing the actual temperature; Inputting sample variable values ​​of the first variable, the second variable and the third variable into the first model, performing feature extraction by the first model to obtain a first sample feature vector corresponding to the first variable, a second sample feature vector corresponding to the second variable and a third sample feature vector corresponding to the third variable, and outputting a first predicted temperature based on the first sample feature vector, the second sample feature vector and the third sample feature vector; The model parameters of the first model are adjusted according to the difference between the first predicted temperature and the actual temperature represented by the first sample label.

7. The method according to claim 6, characterized in that The model parameters of the first model include weights and biases. In the process of training the first model, the method further includes: The weight and the bias are used as the position of the first particle, and based on the particle swarm optimization algorithm, the position of the first particle is adjusted to determine the model parameters corresponding to the first model when the difference between the first predicted temperature and the actual temperature is lower than a first threshold.

8. The method according to claim 6, characterized in that The second model is trained based on the following method: Inputting the first sample feature vector, the second sample feature vector and the third sample feature vector into the second model, and outputting a second predicted temperature by the second model; The model parameters of the second model are adjusted according to the difference between the second predicted temperature and the actual temperature represented by the first sample label.

9. The method according to claim 8, characterized in that The second model is a bidirectional long short-term memory model integrating an attention mechanism, and the model parameters of the second model include the number of hidden layer nodes, the initial learning rate and the regularization coefficient. In the process of training the second model, the method further includes: The number of hidden layer nodes, the initial learning rate and the regularization coefficient are used as the position of the second particle, and based on the particle swarm optimization algorithm, the position of the second particle is adjusted to determine the model parameters corresponding to the second model when the difference between the second predicted temperature and the actual temperature is lower than a second threshold.

10. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the temperature prediction method according to any one of claims 1 to 9 by executing the computer instructions.

Images