Neural network driven electric vehicle temperature control strategy optimization method and system

Abstract

The invention provides a neural network driven electric vehicle temperature control strategy optimization method and system. Feature representation is carried out on a first number of strategy triggering monitoring sequences corresponding to a first number of candidate electric vehicle battery temperature control strategies, and a target representation vector corresponding to each strategy triggering monitoring sequence is obtained; determining a second number of target hybrid representation vectors according to the triggering prediction reference value, loading the target hybrid representation vectors to a hybrid triggering prediction network, and outputting a second number of undetermined hybrid triggering prediction probabilities based on the hybrid triggering prediction network; determining a third number of target mixed trigger prediction probabilities according to the mixed trigger prediction reference value, and weighting the third number of target mixed trigger prediction probabilities according to respective predetermined weighting coefficients, and taking the candidate electric vehicle battery temperature control strategy corresponding to the weighted maximum target hybrid triggering prediction probability as the electric vehicle temperature control strategy. The scientificity of electric vehicle battery temperature control can be improved, and battery health maintenance and service life prolonging are facilitated.

Description

Technical Field

[0001] The present invention relates to the fields of data processing and machine learning, and in particular to a method and system for optimizing a temperature control strategy for an electric vehicle driven by a neural network. Background Art

[0002] With the rapid expansion and widespread application of the electric vehicle market, the performance and stability of electric vehicles have received widespread attention. Among them, temperature control strategies play a vital role in the performance, life and safety of electric vehicle batteries. When operating at different ambient temperatures, battery temperature control faces huge challenges. In a high temperature environment, the battery is prone to overheating, which accelerates the internal chemical reaction of the battery, resulting in faster battery capacity decay, increased internal resistance, reduced battery charging and discharging efficiency, and even thermal runaway and other safety issues. In a low temperature environment, the activity of the battery is reduced, the conductivity of the electrolyte deteriorates, and the output power and capacity of the battery are greatly reduced, which significantly shortens the cruising range of the electric vehicle and even makes it impossible to start. The existing electric vehicle temperature control strategy lacks the ability to accurately predict and optimize the triggering probability of different types of temperature control strategies. In actual applications, there are a variety of temperature control strategies to choose from, such as active cooling, passive cooling, active heating, etc., and each strategy contains different execution types. The current technology is difficult to accurately determine which temperature control strategy should be selected under what circumstances, resulting in low temperature control efficiency and failure to give full play to the advantages of various temperature control strategies. Summary of the invention

[0003] The present invention provides a method and system for optimizing the temperature control strategy of an electric vehicle driven by a neural network.

[0004] According to one aspect of the present invention, a neural network-driven electric vehicle temperature control strategy optimization method is provided, comprising: respectively representing a first number of strategy trigger monitoring sequences corresponding to a first number of candidate electric vehicle battery temperature control strategies, and obtaining a target characterization vector corresponding to each of the strategy trigger monitoring sequences; determining a second number of target hybrid characterization vectors from the first number of target characterization vectors based on a trigger prediction reference value, wherein the second number is a positive integer not greater than the first number; respectively loading the second number of target hybrid characterization vectors into a hybrid trigger prediction network, and outputting a second number of pending hybrid trigger prediction probabilities based on the hybrid trigger prediction network; determining a third number of target hybrid trigger prediction probabilities from the second number of pending hybrid trigger prediction probabilities based on a hybrid trigger prediction reference value, wherein the third number is a positive integer not greater than the second number; and weighting the third number of target hybrid trigger prediction probabilities according to respective predetermined weighting coefficients of the candidate electric vehicle battery temperature control strategies corresponding to the third number of target hybrid trigger prediction probabilities, and taking the candidate electric vehicle battery temperature control strategy corresponding to the weighted maximum target hybrid trigger prediction probability as the electric vehicle temperature control strategy.

[0005] According to another aspect of the present invention, a temperature control strategy optimization system is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described above.

[0006] The present invention includes at least the following beneficial effects: during the network training process, the present invention performs feature representation on a first number of strategy trigger monitoring sequence samples to obtain a sample characterization vector corresponding to each strategy trigger monitoring sequence sample, and then determines a second number of mixed sample characterization vectors in the first number of sample characterization vectors based on a trigger prediction reference value, and then loads the second number of mixed sample characterization vectors into a mixed trigger prediction network, and outputs a second number of mixed trigger prediction probabilities based on the mixed trigger prediction network. In addition, the second number of mixed sample characterization vectors are loaded into a collaborative trigger prediction network, and outputs a second number of collaborative trigger prediction probabilities based on the collaborative trigger prediction network. Then, based on the uniformity supervision requirement, the second number of mixed trigger prediction probabilities and the second number of collaborative trigger prediction probabilities are cost-acquired to obtain a first supervision cost, and the network parameters of the hybrid trigger prediction network are adjusted based on the first supervision cost. Based on this, we can first select a second number of mixed sample representation vectors as inputs for network training of the hybrid trigger prediction network and the collaborative trigger prediction network, respectively, to alleviate the computational cost of the hybrid trigger prediction network and improve efficiency. In addition, based on the uniform supervision requirements, the hybrid trigger prediction network that needs to be trained predicts the hybrid trigger probability of each strategy, and we can model the collaborative trigger prediction probability of each strategy type that balances the probability values ​​of each strategy type and meets the performance requirements of electric vehicles. On the one hand, based on the first supervision cost, the network parameters of the uniform supervision can be automatically optimized during the network training process. On the other hand, based on different types of strategies, the accuracy of the predicted trigger probability of the hybrid trigger prediction network can be improved to obtain a hybrid trigger prediction network that meets the performance of electric vehicles. The temperature control of the battery is more scientific, which is beneficial to the health maintenance and life extension of the battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A flow chart of a neural network driven electric vehicle temperature control strategy optimization method according to an embodiment of the present invention is shown.

[0008] Figure 2 A schematic diagram of the composition of a temperature control strategy optimization system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0009] The neural network driven electric vehicle temperature control strategy optimization method provided by the present invention can be executed by the temperature control strategy optimization system of the electric vehicle, such as an electronic control unit (ECU), a battery management system (BMS), and the like.

[0010] Please refer to Figure 1 The neural network driven electric vehicle temperature control strategy optimization method provided by the present invention includes the following steps: Step S101: feature representing a first number of strategy trigger monitoring sequences corresponding to a first number of candidate electric vehicle battery temperature control strategies respectively, and obtaining a target representation vector corresponding to each strategy trigger monitoring sequence.

[0011] The candidate electric vehicle battery temperature control strategy is a series of solutions formulated to adjust the temperature of electric vehicle batteries, including different types of strategies such as heating and cooling, and each strategy can be subdivided into different execution types. For example, cooling strategies can be divided into active cooling and passive cooling. Active cooling may be achieved by starting refrigeration equipment, while passive cooling may rely on natural heat dissipation. The first number represents the total number of candidate electric vehicle battery temperature control strategies.

[0012] The strategy trigger monitoring sequence is a detailed record of the triggering of each candidate temperature control strategy. It contains a series of information related to the strategy triggering, such as the triggering time, battery temperature at the time of triggering, ambient temperature, battery power, etc. Taking a cooling strategy as an example, its strategy trigger monitoring sequence may record the specific time and ambient temperature when the active cooling strategy was triggered every day when the battery temperature exceeded 30 degrees Celsius in the past week.

[0013] Feature representation is the process of converting the strategy trigger monitoring sequence into a target representation vector. The temperature control strategy optimization system needs to extract key features from the strategy trigger monitoring sequence and combine these features into a vector. When performing feature representation, the temperature control strategy optimization system can use a variety of technical means.

[0014] One feasible technical means is One-Hot Encoding. For categorical features, such as the execution type of the strategy (active cooling, passive cooling), One-Hot Encoding can convert each category into a binary vector. Assume that there are two execution types, active cooling is represented by the vector [1, 0], and passive cooling is represented by the vector [0, 1]. Another technical means is normalization. For numerical features, such as battery temperature, ambient temperature, battery power, etc., since the value ranges of these features may vary greatly, in order to avoid some features from having too much impact on the model, they can be normalized, such as Min-Max Normalization and Z-score normalization.

[0015] The temperature control strategy optimization system can also use feature selection technology to select the features that have the greatest impact on the prediction of policy triggering from a large number of features to reduce the amount of calculation and improve the accuracy of the model. For example, by calculating the correlation between the features and the policy triggering, the features with higher correlation are selected. After completing feature extraction and processing, the temperature control strategy optimization system combines these features into a vector, namely the target representation vector. Each dimension of the target representation vector corresponds to a feature, and its length depends on the number of selected features. Assuming that the four features of battery temperature, ambient temperature, battery power, and policy execution type are selected, after processing, the target representation vector may be a four-dimensional vector, such as [0.5, 0.3, 0.6, [1, 0]], where the first three dimensions correspond to the normalized battery temperature, ambient temperature, and battery power, respectively, and the last dimension is the policy execution type represented by the one-hot encoding.

[0016] Step S102: determining a second number of target mixed representation vectors from a first number of target representation vectors according to a trigger prediction reference value, wherein the second number is a positive integer not greater than the first number.

[0017] The trigger prediction reference value is a pre-set threshold value used to determine whether a candidate electric vehicle battery temperature control strategy corresponding to a target characterization vector has a sufficient probability of being triggered. It plays a key filtering role in the entire screening process. For example, the trigger prediction reference value is set to 0.3, which means that only when the trigger probability corresponding to a target characterization vector is greater than or equal to 0.3, the target characterization vector may be retained for subsequent analysis. The first number of target characterization vectors is the result obtained after feature representation of the strategy trigger monitoring sequence corresponding to the first number of candidate electric vehicle battery temperature control strategies in step S101. Each target characterization vector represents the relevant feature information of a candidate temperature control strategy. Assuming the first number is 10, there are 10 target characterization vectors, corresponding to 10 different candidate electric vehicle battery temperature control strategies.

[0018] When determining the second number of target mixed representation vectors, the temperature control strategy optimization system needs to evaluate the first number of target representation vectors according to the trigger prediction reference value. Specifically, a trigger prediction probability is calculated for each target representation vector, which indicates the probability of the candidate temperature control strategy corresponding to the target representation vector being triggered.

[0019] In order to calculate the trigger prediction probability, a feasible implementation method is to use a machine learning model, such as a logistic regression model, to predict the probability of an event based on the input target representation vector. The formula is: , where P(Y=1|X) represents the given target representation vector In the case of, the probability that the candidate temperature control strategy is triggered (Y=1), It is a parameter of the model and needs to be estimated through training data. Another way is to use a decision tree model. The decision tree model makes decisions by constructing a tree structure, where each internal node represents a test on a feature, each branch represents a test output, and each leaf node represents a category or value. The temperature control strategy optimization system can determine the trigger prediction probability of the target representation vector based on its path in the decision tree. After calculating the trigger prediction probability of each target representation vector, these probabilities are compared with the trigger prediction reference value. If the trigger prediction probability of a target representation vector is not less than the trigger prediction reference value, then the target representation vector will be selected and become the target mixed representation vector; conversely, if the trigger prediction probability is less than the trigger prediction reference value, the target representation vector will be filtered out.

[0020] Step S103: loading a second number of target hybrid representation vectors into the hybrid trigger prediction network respectively, and outputting a second number of pending hybrid trigger prediction probabilities based on the hybrid trigger prediction network.

[0021] The hybrid trigger prediction network is a trained neural network model that predicts the probability of triggering the corresponding candidate electric vehicle battery temperature control strategy based on the input target hybrid representation vector. The structure and parameters of the network are determined during the training process. By learning a large amount of sample data, the network can accurately capture the relationship between the target hybrid representation vector and the strategy trigger probability.

[0022] The second number of target mixed characterization vectors are selected from the first number of target characterization vectors according to the trigger prediction reference value in step S102, and the candidate temperature control strategies corresponding to them have a relatively high possibility of being triggered. For example, after screening, 5 target mixed characterization vectors are obtained, corresponding to 5 different candidate electric vehicle battery temperature control strategies.

[0023] The temperature control strategy optimization system inputs the second number of target mixed characterization vectors into the mixed trigger prediction network in sequence. During the input process, each target mixed characterization vector is transmitted and calculated between the layers of neurons in the network. The mixed trigger prediction network is composed of an input layer, a hidden layer, and an output layer. The input layer receives the target mixed characterization vector, the hidden layer performs nonlinear transformation and feature extraction on the input information, and the output layer outputs the undetermined mixed trigger prediction probability corresponding to the target mixed characterization vector.

[0024] In the actual calculation process, each layer of neurons in the hybrid trigger prediction network has corresponding weights and biases. Taking a fully connected neural network as an example, assuming that there are n neurons in the hidden layer and the target hybrid representation vector of the input layer is , the input of the i-th hidden layer neuron is ,in is the weight from the jth neuron in the input layer to the ith neuron in the hidden layer, is the bias of the i-th neuron in the hidden layer. Then, through the activation function Processing is performed to obtain the output of the i-th neuron in the hidden layer , feasible activation functions include Sigmoid function , ReLU function f(x)=max(0,x), etc. Then, the output of the hidden layer is used as the input of the output layer, and the weighted summation and activation function processing are also performed to finally obtain the predicted probability of the pending mixed trigger.

[0025] Step S104: determining a third number of target hybrid trigger prediction probabilities from a second number of pending hybrid trigger prediction probabilities according to the hybrid trigger prediction reference value, wherein the third number is a positive integer not greater than the second number.

[0026] The hybrid trigger prediction reference value is a pre-set threshold value, which is an indicator for further filtering strategies that are unlikely to be triggered. Similar to the trigger prediction reference value, the setting of the hybrid trigger prediction reference value needs to comprehensively consider multiple factors, such as the performance requirements of electric vehicles, the characteristics of batteries, historical data, etc. For example, if the hybrid trigger prediction reference value is set to 0.6, then only strategies with a pending hybrid trigger prediction probability greater than or equal to 0.6 may be retained. The second number of pending hybrid trigger prediction probabilities is the output result obtained after the temperature control strategy optimization system loads the second number of target hybrid representation vectors into the hybrid trigger prediction network in step S103. Each pending hybrid trigger prediction probability represents the possibility of the corresponding candidate electric vehicle battery temperature control strategy being triggered. Assume that 8 pending hybrid trigger prediction probabilities are obtained in step S103, corresponding to 8 different candidate temperature control strategies. The temperature control strategy optimization system compares the second number of pending hybrid trigger prediction probabilities with the hybrid trigger prediction reference value one by one. If a pending hybrid trigger prediction probability is greater than or equal to the hybrid trigger prediction reference value, then the candidate temperature control strategy corresponding to the probability will be selected, and its corresponding probability becomes the target hybrid trigger prediction probability; conversely, if the pending hybrid trigger prediction probability is less than the hybrid trigger prediction reference value, the candidate temperature control strategy will be excluded.

[0027] The temperature control strategy optimization system can be implemented by using sorting and comparison techniques. First, the temperature control strategy optimization system can sort the second number of pending mixed trigger prediction probabilities, for example, in order from large to small. In this way, during the comparison process, the temperature control strategy optimization system can start the comparison from the highest probability. Once a probability less than the mixed trigger prediction reference value is encountered, the subsequent comparison can be stopped because the subsequent probabilities will only be smaller. Sorting can use feasible sorting algorithms, such as quick sort, merge sort, etc. The average time complexity of quick sort is O(n log n). By selecting a base element, the array is divided into two parts, so that the elements in the left part are less than or equal to the base element, and the elements in the right part are greater than or equal to the base element, and then the left and right parts are recursively sorted respectively.

[0028] Assume that the eight pending mixed trigger prediction probabilities obtained through step S103 are 0.2, 0.4, 0.6, 0.7, 0.8, 0.3, 0.5, and 0.9, respectively, and the mixed trigger prediction reference value is 0.6. The temperature control strategy optimization system first sorts these probabilities and obtains 0.9, 0.8, 0.7, 0.6, 0.5, 0.4, 0.3, and 0.2. Then, starting from the highest probability of 0.9, the comparison is found that 0.9 is greater than 0.6, 0.8 is greater than 0.6, 0.7 is greater than 0.6, 0.6 is equal to 0.6, and 0.5 is less than 0.6. At this time, the temperature control strategy optimization system stops comparing. Then the candidate temperature control strategies corresponding to the four probabilities of 0.9, 0.8, 0.7, and 0.6 are selected, and their corresponding probabilities become the target mixed trigger prediction probabilities. At this time, the third number is 4. Through the screening of step S104, the temperature control strategy optimization system can further narrow the range of candidate temperature control strategies and focus on those strategies that are more likely to be triggered. This not only reduces the complexity of subsequent calculations, but also improves the accuracy and effectiveness of the final electric vehicle temperature control strategy.

[0029] Step S105: Weight the third number of target hybrid trigger prediction probabilities according to their respective predetermined weighting coefficients, and use the candidate electric vehicle battery temperature control strategy corresponding to the weighted maximum target hybrid trigger prediction probability as the electric vehicle temperature control strategy.

[0030] The third number of target hybrid trigger prediction probabilities are selected from the second number of pending hybrid trigger prediction probabilities according to the hybrid trigger prediction reference value in step S104, and the candidate temperature control strategies corresponding to these probabilities have a higher probability of being triggered. Each target hybrid trigger prediction probability represents the probability of the corresponding candidate temperature control strategy being triggered under the current conditions.

[0031] The predetermined weighting coefficients of each candidate electric vehicle battery temperature control strategy are weight values ​​set for each strategy after considering multiple factors. These factors include, but are not limited to, the energy consumption of the strategy, the impact on battery life, and cost. For example, for an active cooling strategy, if its energy consumption is high but it can quickly and effectively reduce the battery temperature, a relatively low weighting coefficient can be set for it; and for a passive cooling strategy, although the cooling speed may be slow, but the energy consumption is low and the impact on battery life is small, a relatively high weighting coefficient can be set for it. Assume that after screening in step S104, three target hybrid trigger prediction probabilities are obtained, corresponding to three candidate temperature control strategies, and their weighting coefficients are 0.2, 0.3, and 0.5, respectively.

[0032] The temperature control strategy optimization system weights the third number of target mixed trigger prediction probabilities by multiplying each target mixed trigger prediction probability by its corresponding weighting coefficient. Assuming that the three target mixed trigger prediction probabilities are 0.7, 0.8, and 0.6, respectively, the weighted results are 0.7×0.2=0.14, 0.8×0.3=0.24, and 0.6×0.5=0.3, respectively. In this process, the weighting coefficient plays a role in adjusting the importance of each strategy, so that different strategies have different proportions in the final decision.

[0033] After completing the weighted calculation, compare these weighted results to find the largest value. The candidate temperature control strategy corresponding to this maximum value is the final electric vehicle temperature control strategy. In the above example, the maximum weighted value is 0.3, and its corresponding candidate temperature control strategy will be used as the electric vehicle temperature control strategy by the temperature control strategy optimization system.

[0034] Through this weighted comparison method, the temperature control strategy optimization system can comprehensively consider the triggering possibility of the strategy and the comprehensive performance of each strategy, so as to select the temperature control strategy that best suits the current situation. This method not only considers the probability of the strategy being triggered, but also considers various factors in the actual application of the strategy, making the final temperature control strategy more scientific and reasonable.

[0035] As an implementation method, the training process of the hybrid trigger prediction network includes the following steps: Step S10: Perform feature representation on each strategy trigger monitoring sequence sample in a first number of strategy trigger monitoring sequence samples to obtain a sample characterization vector corresponding to each strategy trigger monitoring sequence sample, wherein each sample characterization vector carries a corresponding sample strategy type label.

[0036] The first number of strategy trigger monitoring sequence samples are data sets used to train the hybrid trigger prediction network. These samples record the triggering conditions of different candidate electric vehicle battery temperature control strategies. Each strategy trigger monitoring sequence sample contains a series of information related to the strategy trigger, such as the trigger time, battery temperature at the time of triggering, ambient temperature, battery power, etc.

[0037] Feature representation is the process of converting the strategy trigger monitoring sequence samples into sample representation vectors. The temperature control strategy optimization system needs to extract key features from the strategy trigger monitoring sequence samples and combine these features into a vector. When performing feature representation, the temperature control strategy optimization system can use a variety of technical means.

[0038] For classification features, such as the execution type of the strategy (active cooling, passive cooling), the temperature control strategy optimization system can use One-Hot Encoding for processing. One-Hot Encoding converts each category into a binary vector whose length is equal to the number of categories, with only the corresponding category position being 1 and the rest being 0. Assume that there are two execution types, active cooling is represented by the vector [1, 0] and passive cooling is represented by the vector [0, 1].

[0039] For numerical features, such as battery temperature, ambient temperature, battery power, etc., since the value ranges of these features may vary greatly, in order to avoid some features having too much impact on the model, they need to be normalized. The feasible normalization methods are Min-Max Normalization and Z-score Normalization.

[0040] After feature extraction and processing, the temperature control strategy optimization system combines these features into a vector, namely the sample representation vector. Each dimension of the sample representation vector corresponds to a feature, and its length depends on the number of features selected. Assuming that the four features of battery temperature, ambient temperature, battery power, and policy execution type are selected, after processing, the sample representation vector may be a four-dimensional vector, such as [0.5, 0.3, 0.6, [1, 0]], where the first three dimensions correspond to the normalized battery temperature, ambient temperature, and battery power, respectively, and the last dimension is the policy execution type represented by one-hot encoding.

[0041] Each sample representation vector also needs to carry a corresponding sample strategy type tag, which is used to indicate the strategy type corresponding to the sample representation vector, such as active cooling strategy or passive cooling strategy. The sample strategy type tag can be a label, such as "active" or "passive", or a numerical code, such as 1 for active strategy and 0 for passive strategy.

[0042] Step S20: determining a second number of mixed sample representation vectors from the first number of sample representation vectors according to the trigger prediction reference value, wherein the second number is a positive integer not greater than the first number.

[0043] In step S20, the temperature control strategy optimization system determines a second number of mixed sample representation vectors from the first number of sample representation vectors based on the trigger prediction reference value, where the second number is a positive integer not greater than the first number. This step is a key screening link in the training process of the hybrid trigger prediction network. Its purpose is to select those vectors that are more likely to correspond to the trigger strategy from a large number of sample representation vectors, so as to reduce the amount of calculation for subsequent network training, while improving training efficiency and model accuracy. The trigger prediction reference value is a pre-set threshold value used to preliminarily filter sample representation vectors corresponding to strategies that are unlikely to be triggered. The setting of this threshold value requires comprehensive consideration of multiple factors, such as the performance requirements of electric vehicles, battery characteristics, and historical data. For example, if the trigger prediction reference value is set to 0.2, it means that only sample representation vectors with a trigger probability greater than or equal to 0.2 will be retained.

[0044] The first number of sample characterization vectors is the result obtained after the first number of policy trigger monitoring sequence samples are characterized in step S10. Each sample characterization vector represents the characteristic information of a policy trigger monitoring sequence sample and carries the corresponding sample policy type tag. Assuming that the first number is 100, there are 100 sample characterization vectors, corresponding to 100 different policy trigger monitoring sequence samples.

[0045] When determining the second number of mixed sample characterization vectors, the temperature control strategy optimization system first calculates a trigger prediction probability for each sample characterization vector. Specifically, the temperature control strategy optimization system can determine the strategy trigger prediction network based on the sample strategy type tag, and load the sample characterization vector corresponding to the sample strategy type tag into the network, and the network outputs the trigger prediction probability corresponding to the sample characterization vector. The strategy trigger prediction network includes an active strategy trigger prediction network and a passive strategy trigger prediction network, and the trigger prediction probability includes an active trigger prediction probability and a passive trigger prediction probability.

[0046] If the sample strategy type is marked as an active tag, the active strategy trigger prediction network is determined, and the sample representation vector corresponding to the active tag is loaded into the network, and the active trigger prediction probability corresponding to the sample representation vector is output based on the active strategy trigger prediction network. For example, if the sample strategy type corresponding to a sample representation vector is marked as an active tag, the temperature control strategy optimization system inputs the vector into the active strategy trigger prediction network, and the active trigger prediction probability is obtained as 0.3.

[0047] If the sample strategy type is marked as a passive tag, the passive strategy trigger prediction network is determined, and the sample representation vector corresponding to the passive tag is loaded into the network, and the passive trigger prediction probability corresponding to the sample representation vector is output based on the passive strategy trigger prediction network. For example, the sample strategy type corresponding to another sample representation vector is marked as a passive tag, and after being input into the passive strategy trigger prediction network, the passive trigger prediction probability is 0.1.

[0048] After obtaining the trigger prediction probabilities of all sample characterization vectors, select the sample characterization vectors whose trigger prediction probabilities are not less than the trigger prediction reference value from the first number of sample characterization vectors, thereby obtaining the second number of mixed sample characterization vectors. For example, among the above 100 sample characterization vectors, the calculated trigger prediction probabilities are high and low. When the trigger prediction reference value is 0.2, the sample characterization vectors whose trigger prediction probabilities are greater than or equal to 0.2 are screened out. Assuming that 30 sample characterization vectors are obtained after screening, the second number is 30 at this time, and these 30 sample characterization vectors are mixed sample characterization vectors.

[0049] The temperature control strategy optimization system can use sorting and comparison techniques. First, sort the trigger prediction probabilities of all sample representation vectors, and then start the comparison from the highest probability. Once a probability less than the trigger prediction reference value is encountered, stop the subsequent comparison. Sorting can use a feasible sorting algorithm, such as quick sort, whose average time complexity is O(n logn).

[0050] Step S30: Loading a second number of mixed sample representation vectors into the mixed trigger prediction network respectively, and outputting a second number of mixed trigger prediction probabilities based on the mixed trigger prediction network.

[0051] The second number of mixed sample characterization vectors are screened out from the first number of sample characterization vectors according to the trigger prediction reference value in step S20, and the strategies corresponding to these vectors have a relatively high possibility of being triggered. Each mixed sample characterization vector represents the characteristic information of a strategy trigger monitoring sequence sample and carries the corresponding sample strategy type tag. For example, 50 mixed sample characterization vectors are obtained after screening, corresponding to 50 different strategy triggering situations. The mixed trigger prediction network is a designed and constructed neural network model, and its structure and parameters are randomly initialized at the beginning of training. The main function of the network is to receive the mixed sample characterization vector as input, and output the mixed trigger prediction probability corresponding to each sample through calculation and transmission of neurons inside the network. The mixed trigger prediction network is composed of an input layer, a hidden layer and an output layer. The input layer is responsible for receiving the mixed sample characterization vector, the hidden layer performs nonlinear transformation and feature extraction on the input information, and the output layer outputs the final mixed trigger prediction probability.

[0052] The temperature control strategy optimization system sequentially inputs the second number of mixed sample representation vectors into the mixed trigger prediction network. During the input process, each mixed sample representation vector is transmitted and calculated between neurons in each layer of the network, such as the fully connected neural network mentioned above.

[0053] Step S40: loading a second number of mixed sample representation vectors into the collaborative trigger prediction network respectively, and outputting a second number of collaborative trigger prediction probabilities based on the collaborative trigger prediction network.

[0054] The second number of mixed sample characterization vectors are obtained through screening in step S20. The strategies corresponding to these vectors have a relatively high probability of being triggered and have been used for the calculation of the mixed trigger prediction network in step S30. Each mixed sample characterization vector contains the key feature information of the strategy trigger monitoring sequence sample and the corresponding sample strategy type tag. For example, assuming that 80 mixed sample characterization vectors are obtained after screening, they correspond to different electric vehicle battery temperature control strategy triggering situations.

[0055] The collaborative trigger prediction network is a neural network model that cooperates with the hybrid trigger prediction network. Its structure may be similar or different from that of the hybrid trigger prediction network, but both are designed to predict the trigger probability of the strategy. The role of the collaborative trigger prediction network is to evaluate the strategy trigger probability from another perspective to provide additional supervision information to help the hybrid trigger prediction network learn and optimize better. The network also consists of an input layer, a hidden layer, and an output layer. The input layer receives the mixed sample representation vector, the hidden layer performs feature extraction and nonlinear transformation, and the output layer outputs the collaborative trigger prediction probability.

[0056] The temperature control strategy optimization system inputs the second number of mixed sample representation vectors into the collaborative trigger prediction network in sequence. During the input process, each mixed sample representation vector is transmitted and calculated between the neurons of each layer of the collaborative trigger prediction network. Taking a fully connected neural network as an example, assuming that the hidden layer of the collaborative trigger prediction network has k neurons, the mixed sample representation vector received by the input layer is , the input of the pth hidden layer neuron is ,in is the weight from the qth neuron in the input layer to the pth neuron in the hidden layer, is the bias of the pth neuron in the hidden layer. Then, h is activated by the activation function p Processing is performed to obtain the output d of the pth neuron in the hidden layer p =g(h p ), feasible activation functions such as Sigmoid function , ReLU function g(x)=max(0,x), etc. Then, the output of the hidden layer is used as the input of the output layer, and the weighted summation and activation function processing are also performed to finally obtain the prediction probability of collaborative triggering.

[0057] The temperature control strategy optimization system performs such processing on the second number of mixed sample characterization vectors to obtain a second number of collaborative trigger prediction probabilities. These collaborative trigger prediction probabilities are interrelated with the hybrid trigger prediction probabilities output by the hybrid trigger prediction network in step S30. By comparing these two sets of probabilities, the differences and deficiencies of the network in predicting the probability of strategy triggering can be found. For example, if the hybrid trigger prediction probability of a certain strategy is 0.6, and the collaborative trigger prediction probability is 0.8, it means that there is a difference in the judgment of the two networks on the possibility of triggering the strategy. Subsequently, based on this difference, combined with the uniformity supervision requirements, etc., the supervision cost can be calculated, and the parameters of the hybrid trigger prediction network can be adjusted, so that the hybrid trigger prediction network can more accurately predict the trigger probability of the strategy, improve the performance and reliability of the entire temperature control strategy optimization system, and achieve more scientific and reasonable electric vehicle battery temperature control.

[0058] Step S50: According to the uniformity supervision requirement, cost acquisition is performed on the second number of mixed trigger prediction probabilities and the second number of collaborative trigger prediction probabilities to obtain a first supervision cost.

[0059] In step S50, the temperature control strategy optimization system obtains the cost of the second number of hybrid trigger prediction probabilities and the second number of collaborative trigger prediction probabilities according to the uniformity supervision requirement to obtain the first supervision cost. This step compares the outputs of the hybrid trigger prediction network and the collaborative trigger prediction network to measure the difference between the two, thereby providing a basis for adjusting the network parameters, ensuring that the trigger prediction probabilities of different strategies remain consistent or similar in certain statistics, and improving the accuracy and stability of the hybrid trigger prediction network.

[0060] The uniformity supervision requirement is a constraint that ensures that the trigger prediction probabilities of different strategies are consistent or similar in certain statistics. This requirement helps to avoid the network from making overly biased predictions for different types of strategies, making the probability output by the network more consistent with the actual situation. For example, the uniformity supervision requirement can be uniformity supervision of probability averages, that is, requiring the average values ​​of the trigger prediction probabilities of different types of strategies for the hybrid trigger prediction network and the collaborative trigger prediction network to be similar.

[0061] The second number of hybrid trigger prediction probabilities are the results output by the hybrid trigger prediction network in step S30 after processing the second number of hybrid sample characterization vectors, and each probability represents the possibility of the corresponding strategy being triggered. The second number of collaborative trigger prediction probabilities are the results output by the collaborative trigger prediction network in step S40 after processing the same second number of hybrid sample characterization vectors. These two sets of probabilities are the basic data for calculating the first supervision cost.

[0062] Assume that the second number is 50, that is, there are 50 mixed trigger prediction probabilities and 50 collaborative trigger prediction probabilities. Among them, the mixed trigger prediction probabilities are , the prediction probabilities of cooperative triggering are .

[0063] If the uniform supervision requirement is the uniform supervision of the probability average, the temperature control strategy optimization system needs to obtain the cost according to the following steps. First, according to the sample strategy type label, the mixed trigger prediction probability and the collaborative trigger prediction probability are divided into active and passive categories respectively. Assume that among the 50 samples, there are 30 active samples and 20 passive samples. The active mixed trigger prediction probability is , the prediction probability of active collaborative triggering is ; The predicted probability of passive hybrid triggering is , the predicted probability of passive collaborative triggering is .

[0064] Then, the mean of each class is calculated. Active Mixed Mean , passive mixed mean , active collaborative mean , passive synergy mean .

[0065] Finally, the cost function is used to obtain the mean value. A feasible cost function is the mean square error (MSE), which is given by ,y i is the true value, is the predicted value. In this case, the first supervision cost C can be expressed as .

[0066] By calculating the first supervision cost, the difference between the outputs of the hybrid trigger prediction network and the collaborative trigger prediction network can be quantified. This cost reflects the performance of the network in meeting the uniform supervision requirements. If the first supervision cost is large, it means that the output difference of the two networks is large, and the parameters of the hybrid trigger prediction network need to be adjusted to reduce the difference so that the network output is more in line with the uniform supervision requirements. For example, the gradient descent algorithm can be used to update the parameters of the hybrid trigger prediction network according to the first supervision cost, gradually reduce the cost, and improve the performance of the network. In practical applications, according to different uniform supervision requirements, the appropriate cost function can be flexibly selected for cost acquisition. In addition to the mean square error, other cost functions can also be used, such as the cross entropy loss function. At the same time, in order to improve the calculation efficiency and accuracy, batch processing can be used to calculate multiple samples together. In this way, the temperature control strategy optimization system can more effectively train and optimize the hybrid trigger prediction network, so that it can better predict the trigger probability of the strategy, and provide more reliable support for the optimization of the temperature control strategy of electric vehicles.

[0067] Step S60: adjusting network parameters of the hybrid trigger prediction network according to the first supervision cost.

[0068] In step S60, the temperature control strategy optimization system adjusts the network parameters of the hybrid trigger prediction network according to the first supervision cost. This step adjusts the network parameters so that the output of the hybrid trigger prediction network can better meet the uniformity supervision requirements, thereby improving the accuracy of the network's prediction of the strategy trigger probability.

[0069] The first supervision cost is the result obtained by costing the second number of hybrid trigger prediction probabilities and the second number of collaborative trigger prediction probabilities according to the uniformity supervision requirement in step S50. It reflects the degree of difference between the outputs of the hybrid trigger prediction network and the collaborative trigger prediction network, and reflects the degree of compliance of the hybrid trigger prediction network with the uniformity supervision requirement under the current network parameters. For example, if the first supervision cost is large, it means that the output difference of the two networks is obvious, and the hybrid trigger prediction network performs poorly in meeting the uniformity supervision requirement; conversely, if the first supervision cost is small, it means that the network output is more in line with the requirements.

[0070] The network parameters of the hybrid trigger prediction network refer to parameters such as weights and biases in the network, which determine the network's calculation method and output results. At the beginning of training, the network parameters are randomly initialized, and as the training progresses, they need to be continuously adjusted according to the first supervision cost so that the network gradually converges to the optimal state.

[0071] When the temperature control strategy optimization system adjusts the network parameters of the hybrid trigger prediction network, a gradient descent algorithm can be used to update the network parameters along the negative gradient direction of the first supervision cost to gradually reduce the value of the loss function.

[0072] Assume that the network parameters of the hybrid trigger prediction network are , the first supervision cost is The update formula of the gradient descent algorithm is: ,in is the current network parameter, is the updated network parameter, is the learning rate, which controls the step size of each update, The first supervision cost exist The gradient at .

[0073] In each iteration, the temperature control strategy optimization system calculates the gradient based on the current first supervision cost and updates the network parameters. As the number of iterations increases, the first supervision cost will gradually decrease, and the network parameters will gradually converge to the optimal value. When the first supervision cost is less than a preset threshold or the maximum number of iterations is reached, the training process ends.

[0074] By continuously adjusting the network parameters of the hybrid trigger prediction network, the network can better learn the rules of strategy triggering and improve the accuracy of predicting the probability of strategy triggering. At the same time, meeting the uniformity supervision requirements can ensure that the network's prediction results for different types of strategies are statistically consistent, avoiding over-prediction or underestimation of certain types of strategies, thereby providing a more reliable basis for optimizing the temperature control strategy of electric vehicles and achieving more scientific and reasonable battery temperature control.

[0075] As an implementation method, step S50, based on the uniformity supervision requirement, obtains the cost of the second number of mixed trigger prediction probabilities and the second number of collaborative trigger prediction probabilities, and after obtaining the first supervision cost, the method also includes: step S501: based on the dispersion supervision requirement, obtains the cost of the second number of mixed trigger prediction probabilities to obtain the dispersion cost.

[0076] Based on this, step S60, adjusting the network parameters of the hybrid trigger prediction network according to the first supervision cost, includes: step S61: adjusting the network parameters of the hybrid trigger prediction network according to the first supervision cost and the dispersion cost.

[0077] In step S501, the temperature control strategy optimization system obtains the cost of the second number of hybrid trigger prediction probabilities according to the dispersion supervision requirement to obtain the dispersion cost, and step S61 adjusts the network parameters of the hybrid trigger prediction network according to the first supervision cost and the dispersion cost. The distribution characteristics of the network output probability are further constrained, and the network parameters are optimized by comprehensively considering various cost information, thereby improving the performance of the network and the prediction accuracy.

[0078] The dispersion supervision requirement is a difference constraint that is used to ensure that the different probabilities of the network output statistically meet certain standards or expectations. It complements the inadequacy of the uniformity supervision requirement and avoids the network output probabilities being too concentrated or lacking in diversity. For example, in the scenario of electric vehicle temperature control strategy, if the trigger probabilities of all strategies output by the network are very close, then it is impossible to effectively distinguish the pros and cons of different strategies, which is not conducive to selecting a suitable temperature control strategy. Through the dispersion supervision requirement, the probability distribution of the network output can be made more reasonable, improving the effectiveness of strategy selection.

[0079] In step S501, a cost is obtained for a second number of mixed trigger prediction probabilities to obtain a dispersed cost. Assuming that the second number is N, the mixed trigger prediction probabilities are According to the sample strategy type label, these probabilities can be divided into two categories: active and passive. Let the number of active samples be N a , the predicted probability of active hybrid triggering is ; The number of passive samples is N p , the predicted probability of passive hybrid triggering is ,and .

[0080] In order to calculate the dispersion cost, we first need to calculate the dispersion (standard deviation) and mean of each class. The formula Calculate, where , is the active mixed mean, is the predicted probability of the i-th active mixing trigger. Passive mixing discreteness The formula can be Calculate, where , is the passive mixing mean, is the predicted probability of the i-th passive hybrid trigger.

[0081] After obtaining the active mixed discreteness, active mixed mean, passive mixed discreteness and passive mixed mean, the temperature control strategy optimization system needs to obtain the cost of these values ​​to obtain the dispersion cost. One feasible way is to construct the cost function using the idea of ​​mean square error. Let the dispersion cost be ,but ,in are the expected active mixing dispersion, active mixing mean, passive mixing dispersion and passive mixing mean, respectively. is the corresponding weight coefficient, which is used to adjust the importance of each item in the dispersion cost. For example, if you want to pay more attention to the constraint of discreteness, you can set w 1 and w 3 Set larger.

[0082] After obtaining the dispersion cost, in step S61, the temperature control strategy optimization system adjusts the network parameters of the hybrid trigger prediction network according to the first supervision cost and the dispersion cost. The first supervision cost is calculated according to the uniformity supervision requirement in step S50, which reflects the difference between the outputs of the hybrid trigger prediction network and the collaborative trigger prediction network. Let the first supervision cost be , the dispersion cost is , comprehensive cost It can be obtained by weighted summation, that is, ,in is a weight coefficient used to balance the importance of the first supervision cost and the dispersion cost. For example, if the uniformity supervision requirement is more important, Set larger.

[0083] When the temperature control strategy optimization system adjusts the network parameters of the hybrid trigger prediction network, the gradient descent algorithm is feasible. Assume that the network parameters of the hybrid trigger prediction network are , the update formula of the gradient descent algorithm is ,in is the current network parameter, is the updated network parameter, is the learning rate, which controls the step size of each update, It is the comprehensive cost exist The gradient at .

[0084] When actually calculating the gradient, due to the comprehensive cost is the first supervision cost C u and the dispersion cost C d Weighted, according to the linear properties of the derivative, That is, it is necessary to calculate the gradients of the first supervision cost and the decentralized cost with respect to the network parameters separately, and then combine them according to the weights.

[0085] In each iteration, the temperature control strategy optimization system first calculates the first supervision cost C under the current network parameters. u and the dispersion cost C d , and then calculate the comprehensive cost C according to the above formula total Next, based on the automatic derivation calculation , and update the network parameters according to the update formula of the gradient descent algorithm. As the number of iterations increases, the comprehensive cost C total will gradually decrease, and the network parameters will gradually converge to the optimal value.

[0086] When the comprehensive cost C total When it is less than a preset threshold or reaches the maximum number of iterations, the training process ends. In this way, the temperature control strategy optimization system can comprehensively consider the uniformity supervision requirements and the dispersion supervision requirements and effectively adjust the network parameters of the hybrid trigger prediction network.

[0087] Through the above comprehensive adjustment method, the hybrid trigger prediction network can not only maintain a certain consistency in the trigger prediction probability of different types of strategies (meeting the uniformity supervision requirements), but also ensure that the output probability has a reasonable dispersion (meeting the dispersion supervision requirements). In the application scenario of electric vehicle temperature control strategy, such a network can more accurately predict the trigger probability of different temperature control strategies, providing a more reliable basis for selecting appropriate temperature control strategies. For example, when the trigger probabilities of different strategies output by the network have both a reasonable distribution and statistical consistency, the system can more accurately determine which temperature control strategy should be adopted in different situations, thereby achieving more scientific and efficient battery temperature control, which is beneficial to the health maintenance and life extension of the battery.

[0088] As an implementation method, if the uniformity supervision requirement is the uniformity supervision of the probability average; then step S50, based on the uniformity supervision requirement, performs cost acquisition on the second number of mixed trigger prediction probabilities and the second number of collaborative trigger prediction probabilities to obtain a first supervision cost, including: step S51: based on the sample strategy type mark, in the second number of mixed trigger prediction probabilities, determine the active mixed trigger prediction probability corresponding to the active sample strategy type mark, the active mixing number corresponding to the active mixed trigger prediction probability, the passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and the passive mixing number corresponding to the passive mixed trigger prediction probability; step S52: based on the sample strategy type mark, determine the active sample strategy type in the second number of collaborative trigger prediction probabilities. The active collaborative trigger prediction probability corresponding to the mark, the active collaborative number corresponding to the active collaborative trigger prediction probability, the passive collaborative trigger prediction probability corresponding to the passive sample strategy type mark, and the passive collaborative number corresponding to the passive collaborative trigger prediction probability; step S53: based on the active mixed trigger prediction probability and the active mixed number, obtain the active mixed mean, and based on the passive mixed trigger prediction probability and the passive mixed number, obtain the passive mixed mean; step S54: based on the active collaborative trigger prediction probability and the active collaborative number, obtain the active collaborative mean, and based on the passive collaborative trigger prediction probability and the passive collaborative number, obtain the passive collaborative mean; step S55: perform cost acquisition on the active mixed mean, the passive mixed mean, the active collaborative mean and the passive collaborative mean to obtain the first supervision cost.

[0089] Steps S51 to S55 quantify the difference between the outputs of the two networks by comparing the average values ​​of the trigger prediction probabilities of different types of strategies (active and passive) in the hybrid trigger prediction network and the collaborative trigger prediction network, thereby providing a basis for the subsequent adjustment of the network parameters of the hybrid trigger prediction network, thereby ensuring that the network can more accurately predict the trigger probabilities of different types of strategies and meet the needs of optimizing the temperature control strategy of electric vehicles.

[0090] In step S51, based on the sample strategy type mark, in the second number of mixed trigger prediction probabilities, the active mixed trigger prediction probability corresponding to the active sample strategy type mark, the active mixed quantity corresponding to the active mixed trigger prediction probability, the passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and the passive mixed quantity corresponding to the passive mixed trigger prediction probability are determined. The sample strategy type mark is marked for each sample characterization vector in step S10, and is used to distinguish whether the strategy is an active type or a passive type, such as an active cooling strategy or a passive cooling strategy. The second number of mixed trigger prediction probabilities are the results output after the mixed trigger prediction network processes the second number of mixed sample characterization vectors in step S30. The temperature control strategy optimization system divides each mixed trigger prediction probability into two categories: active and passive by checking the sample strategy type mark corresponding to each mixed trigger prediction probability.

[0091] In step S52, based on the sample strategy type tag, the active collaborative trigger prediction probability corresponding to the active sample strategy type tag, the number of active collaborations corresponding to the active collaborative trigger prediction probability, the passive collaborative trigger prediction probability corresponding to the passive sample strategy type tag, and the number of passive collaborations corresponding to the passive collaborative trigger prediction probability are determined in the second number of collaborative trigger prediction probabilities. The second number of collaborative trigger prediction probabilities is the result output after the collaborative trigger prediction network processes the same second number of mixed sample representation vectors in step S40. Similarly, the temperature control strategy optimization system classifies the collaborative trigger prediction probabilities according to the sample strategy type tag. Continuing with the above example, among the 100 collaborative trigger prediction probabilities, 60 correspond to active sample strategy type tags, and these probabilities are active collaborative trigger prediction probabilities, and the number of active collaborations is 60; 40 correspond to passive sample strategy type tags, and these probabilities are passive collaborative trigger prediction probabilities, and the number of passive collaborations is 40.

[0092] In step S53, the active mixing mean is obtained based on the active mixing trigger prediction probability and the active mixing quantity, and the passive mixing mean is obtained based on the passive mixing trigger prediction probability and the passive mixing quantity. The mean is the average value of a group of data, which can reflect the central tendency of this group of data. The calculation formula of the active mixing mean is: ,in is the active mixing amount, is the predicted probability of the i-th active mixing trigger. The calculation formula for the passive mixing mean is ,in is the passive mixing quantity, is the predicted probability of the i-th passive hybrid trigger.

[0093] In step S54, the active collaboration mean is obtained based on the active collaboration trigger prediction probability and the active collaboration quantity, and the passive collaboration mean is obtained based on the passive collaboration trigger prediction probability and the passive collaboration quantity. The calculation formula of the active collaboration mean is: , where m a is the number of active collaborations, Q ai is the predicted probability of the i-th active coordination trigger. The calculation formula for the passive coordination mean is , where m p is the number of passive synergies, is the predicted probability of the i-th passive collaborative triggering.

[0094] In step S55, the temperature control strategy optimization system obtains the cost of the active hybrid mean, the passive hybrid mean, the active collaborative mean, and the passive collaborative mean to obtain the first supervision cost. The first supervision cost is an indicator that measures the difference between the hybrid trigger prediction network and the collaborative trigger prediction network in the average value of the trigger prediction probability of different types of strategies. A feasible cost acquisition method is to use the mean square error (MSE), whose formula is: , where y i is the true value, is the predicted value. In this case, the first supervision cost C can be expressed as This formula comprehensively considers the mean difference between active and passive type strategies in the hybrid trigger prediction network and the collaborative trigger prediction network.

[0095] As an implementation method, if the dispersion supervision requirement is the dispersion supervision of the probability distribution; then step S501, according to the dispersion supervision requirement, performs cost acquisition on the second number of mixed trigger prediction probabilities to obtain a dispersion cost, including: step S5011: obtaining the active mixing discreteness based on the active mixing trigger prediction probability and the active mixing quantity, and obtaining the passive mixing discreteness based on the passive mixing trigger prediction probability and the passive mixing quantity; step S5012: obtaining the active coordination discreteness based on the active coordination trigger prediction probability and the active coordination quantity, and obtaining the passive coordination discreteness based on the passive coordination trigger prediction probability and the passive coordination quantity; step S5013: performing cost acquisition on the active mixing discreteness, the active mixing mean, the passive mixing discreteness, the passive mixing mean, the active coordination discreteness, the active coordination mean, the passive coordination discreteness and the passive coordination mean to obtain a dispersion cost.

[0096] In step S5011, the active mixing discreteness is obtained based on the active mixing trigger prediction probability and the active mixing quantity, and the passive mixing discreteness is obtained based on the passive mixing trigger prediction probability and the passive mixing quantity. Discreteness is also called standard deviation, which is a statistic used to measure the discreteness or distribution dispersion of a set of data, and it reflects the degree of deviation of the data from the mean. The active mixing trigger prediction probability is the probability of the corresponding active sample strategy type mark selected from the second number of mixing trigger prediction probabilities in step S51, and the active mixing quantity is the number of active mixing trigger prediction probabilities. The passive mixing trigger prediction probability and the passive mixing quantity are the same. The calculation formulas of active mixing discreteness and passive mixing discreteness can refer to the above content and will not be repeated here.

[0097] In step S5012, the active collaboration discreteness is obtained based on the active collaboration trigger prediction probability and the active collaboration quantity, and the passive collaboration discreteness is obtained based on the passive collaboration trigger prediction probability and the passive collaboration quantity. The active collaboration trigger prediction probability is the probability of the corresponding active sample strategy type label selected from the second number of collaboration trigger prediction probabilities in step S52, and the active collaboration quantity is the number of active collaboration trigger prediction probabilities. The passive collaboration trigger prediction probability and the passive collaboration quantity are similar. Active collaboration discreteness The calculation formula is , where m a is the number of active collaborations, Q ai is the predicted probability of the i-th active collaborative trigger, is the active coordination mean. Passive coordination dispersion The calculation formula is , where m p is the amount of passive synergy, Q pi is the predicted probability of the ith passive coordination trigger, is the passive synergy mean.

[0098] In step S5013, the active hybrid discreteness, active hybrid mean, passive hybrid discreteness, passive hybrid mean, active collaborative discreteness, active collaborative mean, passive collaborative discreteness and passive collaborative mean are cost-acquired to obtain the dispersion cost. The dispersion cost is an indicator to measure whether the dispersion degree of the probability distribution of the hybrid trigger prediction network output on different types of strategies meets the requirements.

[0099] In order to calculate the dispersion cost, the temperature control strategy optimization system can construct a cost function, for example, in the form of weighted mean square error. Suppose the dispersion cost is , which can be expressed as ,in are the expected active mixing dispersion, active mixing mean, passive mixing dispersion, passive mixing mean, active collaborative dispersion, active collaborative mean, passive collaborative dispersion and passive collaborative mean, respectively. is the corresponding weight coefficient, which is used to adjust the importance of each item in the dispersion cost.

[0100] As an implementation mode, step S20, based on the trigger prediction reference value, determines a second number of mixed sample characterization vectors from a first number of sample characterization vectors, including: step S21: based on the sample strategy type label, determines the policy trigger prediction network, and loads the sample characterization vector corresponding to the sample strategy type label into the policy trigger prediction network, and outputs the trigger prediction probability corresponding to the sample characterization vector based on the policy trigger prediction network; step S22: from the first number of sample characterization vectors, selects the sample characterization vector whose trigger prediction probability is not less than the trigger prediction reference value to obtain the second number of mixed sample characterization vectors.

[0101] Steps S21-S22 screen the sample representation vectors and preliminarily filter out the sample representation vectors corresponding to strategies that are unlikely to be triggered, thereby reducing the amount of computation for subsequent hybrid trigger prediction network training and improving training efficiency. At the same time, the training is more focused on strategies with a greater possibility of triggering, thereby improving the accuracy of the model.

[0102] In step S21, the temperature control strategy optimization system determines the strategy trigger prediction network based on the sample strategy type tag, and loads the sample characterization vector corresponding to the sample strategy type tag into the strategy trigger prediction network, and outputs the trigger prediction probability corresponding to the sample characterization vector based on the network. The sample strategy type tag is the information marked for each sample characterization vector in step S10, which is used to distinguish whether the strategy is an active type or a passive type, such as an active cooling strategy or a passive cooling strategy. The strategy trigger prediction network includes an active strategy trigger prediction network and a passive strategy trigger prediction network, which are used to predict the trigger probabilities of active strategies and passive strategies, respectively.

[0103] Specifically, if the sample strategy type is marked as an active tag, the active strategy trigger prediction network is determined, and the sample representation vector corresponding to the active tag is loaded into the network, and the active trigger prediction probability corresponding to the sample representation vector is output based on the active strategy trigger prediction network. For example, there are 100 sample representation vectors, of which 50 sample representation vectors have the sample strategy type marked as an active tag. For one of the sample representation vectors X a1, and input it into the active strategy trigger prediction network. The active strategy trigger prediction network is a pre-trained neural network model, whose structure may include an input layer, a hidden layer, and an output layer. Assume that the input layer has n neurons, corresponding to the n feature dimensions of the sample representation vector, and the hidden layer has m neurons. Sample representation vector After being input to the input layer, the input of the jth neuron in the hidden layer is ,in is the weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, is the bias of the jth neuron in the hidden layer, yes An element in yes The ith element in . After an activation function (such as the Sigmoid function ) is processed to obtain the output of the jth neuron in the hidden layer The output layer receives the output of the hidden layer, and after weighted summation and activation function processing, it outputs the active trigger prediction probability .

[0104] If the sample strategy type is marked as passive, determine the passive strategy trigger prediction network, and load the sample representation vector corresponding to the passive mark into the network, and output the passive trigger prediction probability corresponding to the sample representation vector based on the passive strategy trigger prediction network. For example, the sample strategy type of the other 50 sample representation vectors is marked as passive. For the sample representation vector X p1 , the temperature control strategy optimization system inputs it into the passive strategy trigger prediction network, and after a similar network calculation process, outputs the passive trigger prediction probability P p1 .

[0105] In step S22, the temperature control strategy optimization system selects sample characterization vectors whose trigger prediction probability is not less than the trigger prediction reference value from the first number of sample characterization vectors to obtain a second number of mixed sample characterization vectors. The trigger prediction reference value is a pre-set threshold value used to preliminarily determine whether the strategy corresponding to a sample characterization vector has a sufficient probability of being triggered. For example, if the trigger prediction reference value is set to 0.3, then only when the trigger prediction probability (active trigger prediction probability or passive trigger prediction probability) corresponding to the sample characterization vector is greater than or equal to 0.3, the sample characterization vector will be selected as a mixed sample characterization vector.

[0106] Continuing with the above example, among the 100 sample representation vectors, 50 active sample representation vectors are passed through the active strategy trigger prediction network to obtain the active trigger prediction probability , 50 passive sample representation vectors are passed through the passive strategy trigger prediction network to obtain the passive trigger prediction probability Compare these trigger prediction probabilities with the trigger prediction reference values ​​one by one. Assume =0.2, which is less than the trigger prediction reference value of 0.3, then the sample characterization vector will not be selected; if =0.4, which is greater than the trigger prediction reference value 0.3, then the sample characterization vector The passive trigger prediction probability is also screened in the same way. After screening, assuming that 30 sample representation vectors are finally selected, the second number is 30, and these 30 sample representation vectors are mixed sample representation vectors.

[0107] As an implementation method, the policy trigger prediction network includes an active policy trigger prediction network and a passive policy trigger prediction network, and the trigger prediction probability includes an active trigger prediction probability and a passive trigger prediction probability. Based on this, step S21, according to the sample policy type tag, determines the policy trigger prediction network, and loads the sample representation vector corresponding to the sample policy type tag into the policy trigger prediction network, and outputs the trigger prediction probability corresponding to the sample representation vector based on the policy trigger prediction network, including: step S211: if the sample policy type tag is an active tag, then determines the active policy trigger prediction network, and loads the sample representation vector corresponding to the active tag into the active policy trigger prediction network, and outputs the active trigger prediction probability corresponding to the sample representation vector based on the active policy trigger prediction network; step S212: if the sample policy type tag is a passive tag, then determines the passive policy trigger prediction network, and loads the sample representation vector corresponding to the passive tag into the passive policy trigger prediction network, and outputs the passive trigger prediction probability corresponding to the sample representation vector based on the passive policy trigger prediction network.

[0108] Based on this, step S22, among the first number of sample characterization vectors, select the sample characterization vector whose trigger prediction probability is not less than the trigger prediction reference value to obtain a second number of mixed sample characterization vectors, including: step S221: among the first number of sample characterization vectors, select the sample characterization vector whose active trigger prediction probability and passive trigger prediction probability are not less than the trigger prediction reference value to obtain a second number of mixed sample characterization vectors.

[0109] In step S211, if the sample strategy type mark is an active mark, the active strategy trigger prediction network is determined, and the sample characterization vector corresponding to the active mark is loaded into the network, thereby outputting the active trigger prediction probability corresponding to the sample characterization vector. The active strategy trigger prediction network is a neural network model specially designed for active strategies. Its purpose is to predict the possibility of the active strategy being triggered based on the input sample characterization vector. The sample strategy type mark is the information marked for each sample characterization vector in step S10, which is used to distinguish whether the strategy is an active type or a passive type. The active mark indicates that the strategy corresponding to the sample characterization vector is an active strategy, such as an active cooling strategy, which, for example, requires consumption of additional energy to achieve the temperature control target.

[0110] Assume that the temperature control strategy optimization system has 100 sample representation vectors, of which 50 sample representation vectors have the sample strategy type marked as active. , the temperature control strategy optimization system inputs it into the active strategy trigger prediction network. The active strategy trigger prediction network, for example, consists of an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is the same as the feature dimension of the sample representation vector, which is assumed to be n. The hidden layer has m neurons, which are used to perform nonlinear transformation and feature extraction on the input information. The output layer has one neuron, which outputs the active trigger prediction probability.

[0111] During the network calculation process, each neuron in the input layer receives an eigenvalue of the sample representation vector. The input of the jth neuron in the hidden layer is The formula can be Calculate, where is the weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, is the bias of the jth neuron in the hidden layer, is the sample representation vector In order to introduce nonlinear factors, for example, an activation function is used For processing, feasible activation functions such as Sigmoid function , after being processed by the activation function, the output of the jth neuron in the hidden layer is obtained The output layer receives the output of the hidden layer, and after weighted summation and activation function processing, it outputs the active trigger prediction probability . Assume that the weight of the output layer is , the bias is , then the input of the output layer , the final active trigger prediction probability .

[0112] In step S212, if the sample strategy type is marked as a passive mark, the passive strategy trigger prediction network is determined, and the sample characterization vector corresponding to the passive mark is loaded into the network, and then the passive trigger prediction probability corresponding to the sample characterization vector is output. The passive strategy trigger prediction network is a neural network model specially designed for passive strategies, which is used to predict the possibility of passive strategies being triggered. The passive mark indicates that the strategy corresponding to the sample characterization vector is a passive strategy, such as a passive heat dissipation strategy, which relies on natural environmental conditions to achieve temperature control and does not require additional consumption of a large amount of energy.

[0113] Also among these 100 sample representation vectors, the sample strategy type of another 50 sample representation vectors is marked as passive. , the temperature control strategy optimization system inputs it into the passive strategy trigger prediction network. The structure and calculation process of the passive strategy trigger prediction network are similar to those of the active strategy trigger prediction network. The input layer receives the eigenvalue of the sample representation vector, the hidden layer processes the input information, and the output layer outputs the passive trigger prediction probability P p1 The input of the kth neuron in the hidden layer is , is the weight from the i-th neuron in the input layer to the k-th neuron in the hidden layer, is the bias of the i-th neuron in the hidden layer, is the sample representation vector The output of the kth neuron in the hidden layer is obtained after the activation function is processed. The input of the output layer , the final passive trigger prediction probability .

[0114] After obtaining the active trigger prediction probability and passive trigger prediction probability of all sample characterization vectors based on steps S211 and S212, in step S221, the temperature control strategy optimization system selects sample characterization vectors whose active trigger prediction probability and passive trigger prediction probability are not less than the trigger prediction reference value from the first number of sample characterization vectors to obtain a second number of mixed sample characterization vectors. The trigger prediction reference value is a pre-set threshold value used to preliminarily filter sample characterization vectors corresponding to strategies that are unlikely to be triggered.

[0115] As an implementation method, in step S22, if the sample strategy type is marked as a passive mark, the passive strategy trigger prediction network is determined, and the sample characterization vector corresponding to the passive mark is loaded into the passive strategy trigger prediction network. After the passive strategy trigger prediction network outputs the passive trigger prediction probability corresponding to the sample characterization vector, the method also includes: step S23: obtaining the active cost based on the active mark and the active trigger prediction probability, and adjusting the network parameters of the active strategy trigger prediction network based on the active cost; step S24: obtaining the passive cost based on the passive mark and the passive trigger prediction probability, and adjusting the network parameters of the passive strategy trigger prediction network based on the passive cost.

[0116] Step S23 and step S24 are further operations performed by the temperature control strategy optimization system after obtaining the active trigger prediction probability and the passive trigger prediction probability corresponding to the sample characterization vector. The core purpose is to optimize the performance of the active strategy trigger prediction network and the passive strategy trigger prediction network by calculating the cost and adjusting the network parameters, so that the two networks can more accurately predict the trigger probability of the corresponding type of strategy.

[0117] In step S23, the temperature control strategy optimization system obtains the active cost according to the active tag and the active trigger prediction probability, and adjusts the network parameters of the active strategy trigger prediction network according to the active cost. The active tag is a label carried by the sample representation vector for identifying its corresponding active strategy, and the active trigger prediction probability is the result output by the active strategy trigger prediction network after processing the sample representation vector corresponding to the active tag, which indicates the possibility of the active strategy corresponding to the sample representation vector being triggered.

[0118] Active cost is an indicator that measures the difference between the prediction result of the active strategy trigger prediction network and the actual situation. In order to obtain the active cost, feasible cost functions such as mean square error (MSE) and cross entropy loss function are available. Taking mean square error as an example, assume that there are N sample representation vectors with active labels, and their corresponding true labels (which can be understood as the labels of whether they are actually triggered, if triggered, it is 1, and if not triggered, it is 0) are , the active trigger prediction probability output by the active strategy trigger prediction network is , then the active cost C a The formula can be Calculated, among which, represents the true label corresponding to the sample representation vector with active label, It is the active trigger prediction probability output by the active strategy trigger prediction network after processing the sample representation vector with active labels. This formula calculates the average of the square of the difference between the true label and the predicted probability of each sample. The larger the difference, the higher the active cost, indicating that the network's prediction result deviates more from the actual situation.

[0119] After obtaining the active cost, the network parameters of the active strategy trigger prediction network are adjusted according to the cost. Network parameters include parameters such as weights and biases in the network, which determine the calculation method and output results of the network. A feasible method for adjusting network parameters is the gradient descent algorithm. The basic idea of ​​the gradient descent algorithm is to update the network parameters along the negative gradient direction of the cost function to gradually reduce the value of the cost function.

[0120] In step S24, the temperature control strategy optimization system obtains the passive cost according to the passive tag and the passive trigger prediction probability, and adjusts the network parameters of the passive strategy trigger prediction network according to the passive cost. The passive tag is a label carried by the sample representation vector for identifying its corresponding passive strategy, and the passive trigger prediction probability is the result output by the passive strategy trigger prediction network after processing the sample representation vector corresponding to the passive tag, which indicates the possibility of the passive strategy corresponding to the sample representation vector being triggered.

[0121] Similarly, in order to obtain the passive cost, the calculation method of the active cost mentioned above can be referred to, which will not be described in detail here. After obtaining the passive cost, the network parameters of the passive strategy trigger prediction network are adjusted according to the cost.

[0122] In actual operation, steps S23 and S24 are executed iteratively for multiple times. Each iteration calculates the active cost and the passive cost, and updates the network parameters of the corresponding network according to these costs. As the number of iterations increases, the active cost and the passive cost will gradually decrease, indicating that the deviation between the network's prediction results and the actual situation is shrinking, and the network performance is constantly improving. When the active cost and the passive cost are less than a preset threshold, or the maximum number of iterations is reached, the training process ends.

[0123] As an implementation method, in step S30, a second number of mixed sample representation vectors are loaded into the mixed trigger prediction network respectively, and after a second number of mixed trigger prediction probabilities are output based on the mixed trigger prediction network, the method also includes: step S301: based on the uniformity supervision requirement, the second number of mixed trigger prediction probabilities and the trigger prediction probabilities are obtained to obtain a second supervision cost; step S302: based on the second supervision cost, the network parameters of the mixed trigger prediction network are adjusted.

[0124] In step S301, the temperature control strategy optimization system obtains the cost of the second number of mixed trigger prediction probabilities and trigger prediction probabilities according to the uniformity supervision requirement to obtain the second supervision cost. The trigger prediction probability is the result output by the strategy trigger prediction network (including the active strategy trigger prediction network and the passive strategy trigger prediction network) in step S21, which represents the possibility of the strategy corresponding to the sample representation vector being triggered. The mixed trigger prediction probability is the result output by the mixed trigger prediction network after processing the second number of mixed sample representation vectors in step S30.

[0125] In order to obtain the second supervision cost, feasible cost functions include mean square error (MSE), cross entropy loss function, etc. Assume that the uniform supervision requirement is the uniform supervision of the probability average, and the second number is N, the mixed trigger prediction probability is , the trigger prediction probability is , taking the mean square error as an example, the second supervision cost C 2 The formula can be It is calculated that, as mentioned above, represents the predicted probability of mixed triggering, It represents the trigger prediction probability, which calculates the average of the square of the difference between the mixed trigger prediction probability and the trigger prediction probability of each sample. The larger the difference, the higher the second supervision cost, which means that the greater the difference in the probability average between the output of the mixed trigger prediction network and the output of the strategy trigger prediction network, the higher the degree of non-compliance with the uniform supervision requirements.

[0126] In actual operation, the form of the cost function can be flexibly adjusted according to different uniformity supervision requirements. If the uniformity supervision requirement is uniformity supervision of the probability average of a specific type of strategy (such as active strategy or passive strategy), the temperature control strategy optimization system first classifies the mixed trigger prediction probability and the trigger prediction probability according to the sample strategy type label, and then calculates the cost of different types of strategies respectively, and then comprehensively obtains the second supervision cost.

[0127] In step S302, the temperature control strategy optimization system adjusts the network parameters of the hybrid trigger prediction network based on the second supervision cost. Network parameters include parameters such as weights and biases in the network, which determine the calculation method and output results of the network. The purpose of adjusting the network parameters is to reduce the second supervision cost so that the output of the hybrid trigger prediction network is more in line with the uniformity supervision requirements. A feasible adjustment method is the gradient descent algorithm. The basic idea of ​​the gradient descent algorithm is to update the network parameters along the negative gradient direction of the cost function to gradually reduce the value of the cost function.

[0128] Step S301 and step S302 are executed for multiple iterations. The second supervision cost is calculated in each iteration, and the network parameters of the hybrid trigger prediction network are updated according to the cost. As the number of iterations increases, the second supervision cost will gradually decrease, indicating that the difference between the output of the hybrid trigger prediction network and the output of the policy trigger prediction network in meeting the uniformity supervision requirements is shrinking, and the performance of the network is constantly improving. When the second supervision cost is less than a preset threshold, or the maximum number of iterations is reached, the training process ends.

[0129] Through step S301 and step S302, it can be ensured that the strategy trigger probability output by the hybrid trigger prediction network is statistically consistent with the output of the strategy trigger prediction network, avoiding over-prediction or underestimation of certain types of strategies. For example, in a high temperature environment, the triggering probabilities of the active cooling strategy and the passive cooling strategy should maintain a reasonable relationship to a certain extent. If the triggering probability of the active cooling strategy output by the hybrid trigger prediction network is too high, and the triggering probability of the passive cooling strategy is too low, it may cause the system to over-rely on the active cooling strategy and increase energy consumption. By adjusting the network parameters so that the output of the hybrid trigger prediction network meets the uniformity supervision requirements, the appropriate temperature control strategy can be selected more accurately, and a more scientific and efficient battery temperature control can be achieved, which is beneficial to the health maintenance and life extension of the battery.

[0130] As an implementation method, step S301, based on the uniformity supervision requirement, obtains the cost of the second number of mixed trigger prediction probabilities and the trigger prediction probabilities to obtain the second supervision cost, including: step S3011: based on the sample strategy type mark, determine the active mixed trigger prediction probability corresponding to the active sample strategy type mark, the active mixing number corresponding to the active mixed trigger prediction probability, the passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and the passive mixing number corresponding to the passive mixed trigger prediction probability in the second number of mixed trigger prediction probabilities; step S3012: based on the sample strategy type mark, determine the active trigger prediction probability corresponding to the active sample strategy type mark, the active number corresponding to the active trigger prediction probability, and the passive number corresponding to the passive mixed trigger prediction probability in the first number of trigger prediction probabilities. Measurement probability, and the passive number corresponding to the passive trigger prediction probability; step S3013: according to the active mixed trigger prediction probability and the active mixed number, obtain the active mixed mean, and according to the passive mixed trigger prediction probability and the passive mixed number, obtain the passive mixed mean; step S3014: according to the active trigger prediction probability and the active number, obtain the active mean, and according to the passive trigger prediction probability and the passive number, obtain the passive mean; step S3015: if the uniformity supervision requirement is the uniformity supervision of the active probability average value, then the active mixed mean, the passive mixed mean and the active mean are cost-acquired to obtain the second supervision cost; step S3016: if the uniformity supervision requirement is the uniformity supervision of the passive probability average value, then the active mixed mean, the passive mixed mean and the passive mean are cost-acquired to obtain the second supervision cost.

[0131] In step S3011, the temperature control strategy optimization system determines the active mixed trigger prediction probability corresponding to the active sample strategy type mark, the active mixing quantity corresponding to the active mixed trigger prediction probability, the passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and the passive mixing quantity corresponding to the passive mixed trigger prediction probability in the second number of mixed trigger prediction probabilities based on the sample strategy type mark. The sample strategy type mark is an identifier of the strategy type corresponding to the strategy trigger monitoring sequence sample, which is divided into active marks and passive marks. The active mark corresponds to active strategies, such as active cooling, active heating, etc., and the passive mark corresponds to passive strategies, such as natural heat dissipation, natural heating, etc. The mixed trigger prediction probability is the probability value output after the second number of mixed sample representation vectors are loaded into the mixed trigger prediction network. The second number of mixed trigger prediction probabilities are traversed and classified according to the sample strategy type mark.

[0132] In step S3012, the temperature control strategy optimization system determines the active trigger prediction probability corresponding to the active sample strategy type tag, the active number corresponding to the active trigger prediction probability, the passive trigger prediction probability corresponding to the passive sample strategy type tag, and the passive number corresponding to the passive trigger prediction probability in the first number of trigger prediction probabilities based on the sample strategy type tag. The trigger prediction probability is the probability value output after the sample representation vector is loaded into the strategy trigger prediction network. The temperature control strategy optimization system will also classify the first number of trigger prediction probabilities based on the sample strategy type tag.

[0133] In step S3013, the temperature control strategy optimization system obtains the active mixing mean based on the active mixing trigger prediction probability and the active mixing number, and obtains the passive mixing mean based on the passive mixing trigger prediction probability and the passive mixing number. The mean is the average value of a group of data, which is used to reflect the central tendency of the data. The calculation formula of the active mixing mean is: active mixing mean = the sum of the active mixing trigger prediction probability / active mixing number. For example, the active mixing number is 30, and the active mixing trigger prediction probability is 0.2, 0.3, ..., 0.4, etc., respectively. Add these probability values ​​and divide them by 30 to get the active mixing mean. Similarly, the calculation formula of the passive mixing mean is: passive mixing mean = the sum of the passive mixing trigger prediction probability / passive mixing number. The temperature control strategy optimization system can loop through the active mixing trigger prediction probability and the passive mixing trigger prediction probability, sum them up respectively, and then divide them by the corresponding number to obtain the corresponding mean.

[0134] In step S3014, the temperature control strategy optimization system obtains the active mean based on the active trigger prediction probability and the active number, and obtains the passive mean based on the passive trigger prediction probability and the passive number. The calculation formula for the active mean is: active mean = the sum of the active trigger prediction probability / active number. For example, the active number is 80, and the active trigger prediction probabilities are 0.1, 0.2, ..., 0.3, etc., respectively. Add these probability values ​​and divide them by 80 to get the active mean. The calculation formula for the passive mean is: passive mean = the sum of the passive trigger prediction probabilities / passive number. The temperature control strategy optimization system adopts a calculation method similar to step S3013, by looping through the active trigger prediction probability and the passive trigger prediction probability, summing them up and then dividing them by the corresponding number to obtain the corresponding mean.

[0135] In step S3015, if the uniformity supervision requirement is uniformity supervision of the active probability average, the temperature control strategy optimization system obtains the cost of the active mixed mean, the passive mixed mean and the active mean to obtain the second supervision cost. The uniformity supervision of the active probability average is intended to ensure that the probability average output by the active strategy in different networks (hybrid trigger prediction network and strategy trigger prediction network) is consistent. Cost acquisition can use a variety of loss functions, such as the mean square error loss function (MSE). In this case, the active mixed mean, the passive mixed mean and the active mean are substituted into the loss function for calculation. For example, the active mixed mean is 0.3, the passive mixed mean is 0.2, and the active mean is 0.35. Assuming that the active mean is the true value, the active mixed mean and the passive mixed mean are the predicted values, the second supervision cost is calculated by the mean square error loss function. The temperature control strategy optimization system can adjust the network parameters of the hybrid trigger prediction network according to the calculation results of the loss function to reduce the cost and improve the accuracy of the prediction.

[0136] In step S3016, if the uniformity supervision requirement is uniformity supervision of the passive probability average, the temperature control strategy optimization system obtains the cost of the active mixed mean, the passive mixed mean and the passive mean to obtain the second supervision cost. The uniformity supervision of the passive probability average is intended to ensure that the probability average output of the passive strategy in different networks is consistent. The mean square error loss function is also used to obtain the cost, and the active mixed mean, the passive mixed mean and the passive mean are substituted into the loss function. For example, the active mixed mean is 0.25, the passive mixed mean is 0.3, and the passive mean is 0.32. The passive mean is the true value, the active mixed mean and the passive mixed mean are the predicted values, and the second supervision cost is calculated by the mean square error loss function. The temperature control strategy optimization system adjusts the network parameters of the hybrid trigger prediction network based on the calculated second supervision cost, so that the probability predictions of different networks for passive strategies are more consistent.

[0137] As another implementation method, in step S30, a second number of hybrid sample representation vectors are loaded into the hybrid trigger prediction network respectively, and after a second number of hybrid trigger prediction probabilities are output based on the hybrid trigger prediction network, the method also includes: step S30A: obtaining the actual trigger rate corresponding to each strategy trigger monitoring sequence sample; step S30B: based on the uniformity supervision requirement, performing cost acquisition on the second number of hybrid trigger prediction probabilities and the actual trigger rate to obtain a third supervision cost; step S30C: based on the third supervision cost, adjusting the network parameters of the hybrid trigger prediction network.

[0138] In step S30A, the temperature control strategy optimization system obtains the actual trigger rate corresponding to each strategy trigger monitoring sequence sample. The actual trigger rate refers to the frequency at which the strategy corresponding to a strategy trigger monitoring sequence sample is triggered in actual operation. For example, for a sample of the temperature control strategy of an electric vehicle battery, 100 monitorings were performed over a period of time, of which the corresponding cooling strategy was actually triggered 20 times, then the actual trigger rate of the sample is 20%. The temperature control strategy optimization system can collect actual operation data, count the number of strategy triggers corresponding to each strategy trigger monitoring sequence sample, and divide it by the total number of monitoring times to obtain the actual trigger rate.

[0139] In step S30B, according to the uniformity supervision requirement, the second number of mixed trigger prediction probabilities and the actual trigger rate are obtained by cost acquisition to obtain the third supervision cost. This step is intended to ensure that the mixed trigger prediction probability output by the mixed trigger prediction network is consistent or similar to the actual trigger rate in certain statistics. The specific steps are as follows: First, the temperature control strategy optimization system determines the active mixed trigger prediction probability corresponding to the active sample strategy type mark, the active mixing number corresponding to the active mixed trigger prediction probability, the passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and the passive mixing number corresponding to the passive mixed trigger prediction probability in the second number of mixed trigger prediction probabilities based on the sample strategy type mark. For example, in the second number of 50 mixed trigger prediction probabilities, the sample strategy type mark shows that 20 belong to the active sample strategy type mark, then the probability values ​​corresponding to these 20 are the active mixed trigger prediction probabilities, and the active mixing number is 20; the remaining 30 belong to the passive sample strategy type mark, and the corresponding probability value is the passive mixed trigger prediction probability, and the passive mixing number is 30. Next, the temperature control strategy optimization system determines the active actual trigger rate corresponding to the active sample strategy type tag, the number of active samples corresponding to the active actual trigger rate, the passive actual trigger rate corresponding to the passive sample strategy type tag, and the number of passive samples corresponding to the passive actual trigger rate in the first number of actual trigger rates based on the sample strategy type tag. For example, the first number is 100 actual trigger rates, of which 40 are corresponding to the active sample strategy type tag, then the actual trigger rate corresponding to these 40 is the active actual trigger rate, and the number of active samples is 40; there are 60 corresponding to the passive sample strategy type tag, and the actual trigger rate corresponding to them is the passive actual trigger rate, and the number of passive samples is 60. Then, the temperature control strategy optimization system obtains the active mixing mean based on the active mixed trigger prediction probability and the active mixing number, and the calculation formula is: active mixing mean = the sum of active mixed trigger prediction probability / active mixing number. Similarly, based on the passive mixed trigger prediction probability and the passive mixing number, the passive mixing mean is obtained, and the formula is: passive mixing mean = the sum of passive mixed trigger prediction probability / passive mixing number. Then, based on the active actual trigger rate and the number of active samples, the active trigger mean is obtained, and the formula is: active trigger mean = sum of active actual trigger rates / number of active samples. Based on the passive actual trigger rate and the number of passive samples, the passive trigger mean is obtained, and the formula is: passive trigger mean = sum of passive actual trigger rates / number of passive samples. If the uniform supervision requirement is the uniform supervision of the active probability average, the temperature control strategy optimization system obtains the cost of the active mixed mean, the passive mixed mean, and the active trigger mean to obtain the third supervision cost. For example, the mean square error loss function (MSE) can be used for cost acquisition.If the uniform supervision requirement is uniform supervision of the passive probability average, the temperature control strategy optimization system obtains the cost of the active mixed mean, the passive mixed mean, and the passive trigger mean to obtain the third supervision cost. Through these steps and calculations, the temperature control strategy optimization system can accurately obtain the cost of the mixed trigger prediction probability and the actual trigger rate according to the uniform supervision requirement.

[0140] In step S30C, the temperature control strategy optimization system adjusts the network parameters of the hybrid trigger prediction network based on the third supervision cost. The third supervision cost reflects the degree of difference between the hybrid trigger prediction probability output by the hybrid trigger prediction network and the actual trigger rate. The goal of the temperature control strategy optimization system is to reduce this difference and improve the accuracy of network prediction by adjusting the network parameters. The temperature control strategy optimization system can use a gradient descent algorithm and its variants, such as stochastic gradient descent (SGD), Adagrad, Adadelta, Adam, etc., to calculate the gradient of the network parameters according to the third supervision cost, and update the network parameters in the opposite direction of the gradient. In each training iteration, the gradient is calculated and the network parameters are updated until the third supervision cost converges to a smaller value or the preset number of training times is reached.

[0141] As an implementation method, step S30B, based on the uniformity supervision requirement, obtains the cost of the second number of mixed trigger prediction probabilities and actual trigger rates to obtain a third supervision cost, including: step S30B1: based on the sample strategy type mark, determine the active mixed trigger prediction probability corresponding to the active sample strategy type mark, the active mixing number corresponding to the active mixed trigger prediction probability, the passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and the passive mixing number corresponding to the passive mixed trigger prediction probability in the second number of mixed trigger prediction probabilities; step S30B2: based on the sample strategy type mark, determine the active actual trigger rate corresponding to the active sample strategy type mark, the active sample number corresponding to the active actual trigger rate, the passive actual trigger rate corresponding to the passive sample strategy type mark, and the passive mixing number corresponding to the passive mixed trigger prediction probability in the first number of actual trigger rates. The number of passive samples corresponding to the actual trigger rate of the active mixture; step S30B3: according to the active mixture trigger prediction probability and the active mixture number, obtain the active mixture mean, and according to the passive mixture trigger prediction probability and the passive mixture number, obtain the passive mixture mean; step S30B4: according to the active actual trigger rate and the number of active samples, obtain the active trigger mean, and according to the passive actual trigger rate and the number of passive samples, obtain the passive trigger mean; step S30B5: if the uniformity supervision requirement is the uniformity supervision of the active probability average value, then the active mixture mean, the passive mixture mean and the active trigger mean are cost-acquired to obtain the third supervision cost; step S30B6: if the uniformity supervision requirement is the uniformity supervision of the passive probability average value, then the active mixture mean, the passive mixture mean and the passive trigger mean are cost-acquired to obtain the third supervision cost.

[0142] In step S30B1, the temperature control strategy optimization system determines the active mixed trigger prediction probability corresponding to the active sample strategy type mark, the active mixing number corresponding to the active mixed trigger prediction probability, the passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and the passive mixing number corresponding to the passive mixed trigger prediction probability in the second number of mixed trigger prediction probabilities according to the sample strategy type mark. The sample strategy type mark is used to distinguish different types of strategies, such as active strategies (such as active cooling, active heating, etc.) and passive strategies (such as natural heat dissipation, natural heating, etc.). The mixed trigger prediction probability is the output result obtained after loading the second number of mixed sample representation vectors into the mixed trigger prediction network. Traverse the second number of mixed trigger prediction probabilities and classify them according to the sample strategy type mark. For example, if the second number is 80, and the sample strategy type mark shows that 35 belong to the active sample strategy type mark, then the probability values ​​corresponding to these 35 are the active mixed trigger prediction probabilities, and the active mixing number is 35; the remaining 45 belong to the passive sample strategy type mark, and the corresponding probability value is the passive mixed trigger prediction probability, and the passive mixing number is 45. This step can be implemented by using data screening and statistical techniques, such as using the query function of the database to screen and count the hybrid trigger prediction probabilities according to the sample strategy type tags.

[0143] In step S30B2, the temperature control strategy optimization system determines the active actual trigger rate corresponding to the active sample strategy type tag, the number of active samples corresponding to the active actual trigger rate, the passive actual trigger rate corresponding to the passive sample strategy type tag, and the number of passive samples corresponding to the passive actual trigger rate in the first number of actual trigger rates based on the sample strategy type tag. The actual trigger rate is the frequency at which a certain strategy is triggered in the actual operation scenario. The temperature control strategy optimization system also classifies the first number of actual trigger rates based on the sample strategy type tag. Assuming that the first number is 150, of which 60 correspond to the active sample strategy type tag, then the actual trigger rate corresponding to these 60 is the active actual trigger rate, and the number of active samples is 60; there are 90 corresponding to the passive sample strategy type tag, and the corresponding actual trigger rate is the passive actual trigger rate, and the number of passive samples is 90. To implement this step, data classification and statistical tools can be used to classify the actual trigger rate data according to the sample strategy type tag, and count the number of each category.

[0144] In step S30B3, the active mixing mean is obtained based on the active mixing trigger prediction probability and the active mixing quantity, and the passive mixing mean is obtained based on the passive mixing trigger prediction probability and the passive mixing quantity. The mean is the average value of a group of data, which can reflect the concentration trend of the data. The calculation formula of the active mixing mean is: active mixing mean = the sum of the active mixing trigger prediction probability / active mixing quantity. For example, the active mixing quantity is 35, and the active mixing trigger prediction probabilities are 0.22, 0.25, ..., 0.3, etc. The temperature control strategy optimization system adds these probability values ​​and divides them by 35 to obtain the active mixing mean. Similarly, the calculation formula of the passive mixing mean is: passive mixing mean = the sum of the passive mixing trigger prediction probability / passive mixing quantity.

[0145] In step S30B4, the active trigger mean is obtained based on the active actual trigger rate and the number of active samples, and the passive trigger mean is obtained based on the passive actual trigger rate and the number of passive samples. The calculation formula for the active trigger mean is: active trigger mean = sum of active actual trigger rates / number of active samples. For example, the number of active samples is 60, and the active actual trigger rates are 0.18, 0.2, ..., 0.23, etc. The temperature control strategy optimization system adds these actual trigger rates and divides them by 60 to obtain the active trigger mean. The calculation formula for the passive trigger mean is: passive trigger mean = sum of passive actual trigger rates / number of passive samples. The temperature control strategy optimization system adopts a method similar to that for calculating the active mixed mean and the passive mixed mean, by looping through the active actual trigger rate and the passive actual trigger rate, summing them up and then dividing them by the corresponding number to obtain the corresponding mean.

[0146] In step S30B5, if the uniformity supervision requirement is uniformity supervision of the active probability average, the temperature control strategy optimization system performs cost acquisition on the active mixed mean, the passive mixed mean and the active trigger mean to obtain the third supervision cost. The uniformity supervision of the active probability average is intended to ensure that the probability average output of the mixed trigger prediction network of the active strategy is consistent or similar to the actual active trigger rate average. Cost acquisition can use a variety of loss functions, such as the mean square error loss function (MSE).

[0147] In step S30B6, if the uniformity supervision requirement is uniformity supervision of the passive probability average, the temperature control strategy optimization system obtains the cost of the active mixed mean, the passive mixed mean and the passive trigger mean to obtain the third supervision cost. The purpose of uniformity supervision of the passive probability average is to ensure that the probability average output of the mixed trigger prediction network by the passive strategy matches the actual passive trigger rate average. The mean square error loss function is also used for cost acquisition, and the active mixed mean, passive mixed mean and passive trigger mean are substituted into the loss function.

[0148] Please refer to Figure 2 , is a structural block diagram of the temperature control strategy optimization system 1000 of the present invention, and the temperature control strategy optimization system 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a ROM 1002 (i.e., a read-only memory) or a computer program loaded from a storage unit 1008 to a RAM 1003 (i.e., a random access memory). In the RAM 1003, various programs and data required for the operation of the temperature control strategy optimization system 1000 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. The I / O interface 1005 (i.e., an input / output interface) is also connected to the bus 1004.

[0149] Multiple components in the temperature control strategy optimization system 1000 are connected to the I / O interface 1005, including: an input unit 1006, an output unit 1007, a storage unit 1008, and a communication unit 1009. The input unit 1006 can be a device that can input information to the temperature control strategy optimization system 1000. The input unit 1006 can receive input digital or character information, and generate key signal input related to user settings and / or function control of the electronic device. The output unit 1007 can be any type of device that can present information, and can include but is not limited to a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 1008 can include but is not limited to a disk, an optical disk. The communication unit 1009 allows the temperature control strategy optimization system 1000 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include but is not limited to a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset.

[0150] The computing unit 1001 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1001 performs the various methods and processes described above.

[0151] That is, the present invention provides a temperature control strategy optimization system 1000, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned neural network-driven electric vehicle temperature control strategy optimization method.

Claims

1. A neural network driven electric vehicle temperature control strategy optimization method, characterized in that: include: Performing feature representation on a first number of strategy trigger monitoring sequences corresponding to a first number of candidate electric vehicle battery temperature control strategies, respectively, to obtain a target representation vector corresponding to each of the strategy trigger monitoring sequences; Determine, according to the trigger prediction reference value, a second number of target mixed representation vectors from the first number of target representation vectors, wherein the second number is a positive integer not greater than the first number; Loading a second number of the target hybrid representation vectors into a hybrid trigger prediction network respectively, and outputting a second number of to-be-determined hybrid trigger prediction probabilities based on the hybrid trigger prediction network; Determining a third number of target hybrid trigger prediction probabilities from the second number of pending hybrid trigger prediction probabilities according to the hybrid trigger prediction reference value, wherein the third number is a positive integer not greater than the second number; According to the predetermined weighting coefficients of the candidate electric vehicle battery temperature control strategies corresponding to the third number of target hybrid trigger prediction probabilities, the third number of target hybrid trigger prediction probabilities are weighted respectively, and the candidate electric vehicle battery temperature control strategy corresponding to the weighted maximum target hybrid trigger prediction probability is used as the electric vehicle temperature control strategy.

2. The method according to claim 1, characterized in that The training process of the hybrid trigger prediction network includes: Performing feature representation on each of the first number of policy trigger monitoring sequence samples to obtain a sample characterization vector corresponding to each of the policy trigger monitoring sequence samples, wherein each of the sample characterization vectors carries a corresponding sample policy type tag; Determine, according to the trigger prediction reference value, a second number of mixed sample representation vectors from the first number of sample representation vectors, wherein the second number is a positive integer not greater than the first number; Loading a second number of the mixed sample representation vectors into a mixed trigger prediction network respectively, and outputting a second number of mixed trigger prediction probabilities based on the mixed trigger prediction network; Loading a second number of the mixed sample representation vectors into the collaborative trigger prediction network respectively, and outputting a second number of collaborative trigger prediction probabilities based on the collaborative trigger prediction network; According to the uniformity supervision requirement, cost acquisition is performed on the second number of the mixed trigger prediction probabilities and the second number of the collaborative trigger prediction probabilities to obtain a first supervision cost; According to the first supervision cost, the network parameters of the hybrid trigger prediction network are adjusted.

3. The method according to claim 2, characterized in that According to the uniformity supervision requirement, after obtaining the first supervision cost, the method further includes: According to the dispersion supervision requirement, cost acquisition is performed on the second number of the mixed trigger prediction probabilities to obtain a dispersion cost; The adjusting of network parameters of the hybrid trigger prediction network according to the first supervision cost includes: According to the first supervision cost and the dispersion cost, adjusting the network parameters of the hybrid trigger prediction network; The determining, based on the trigger prediction reference value, a second number of mixed sample representation vectors from the first number of sample representation vectors comprises: Determine a policy trigger prediction network according to the sample policy type tag, load a sample representation vector corresponding to the sample policy type tag into the policy trigger prediction network, and output a trigger prediction probability corresponding to the sample representation vector based on the policy trigger prediction network; Selecting, from the first number of the sample representation vectors, the sample representation vectors whose trigger prediction probability is not less than the trigger prediction reference value, to obtain a second number of the mixed sample representation vectors; After the second number of mixed sample representation vectors are loaded into the mixed trigger prediction network respectively, and the second number of mixed trigger prediction probabilities are output based on the mixed trigger prediction network, the method further includes: According to the uniformity supervision requirement, performing cost acquisition on the second number of the mixed trigger prediction probabilities and the trigger prediction probability to obtain a second supervision cost; According to the second supervision cost, the network parameters of the hybrid trigger prediction network are adjusted.

4. The method according to claim 2, characterized in that: After the second number of mixed sample representation vectors are loaded into the mixed trigger prediction network respectively, and the second number of mixed trigger prediction probabilities are output based on the mixed trigger prediction network, the method further includes: Obtaining the actual trigger rate corresponding to each of the strategy trigger monitoring sequence samples; According to the uniformity supervision requirement, cost acquisition is performed on the second number of the mixed trigger prediction probabilities and the actual trigger rate to obtain a third supervision cost; According to the third supervision cost, the network parameters of the hybrid trigger prediction network are adjusted.

5. The method according to claim 3, characterized in that: The strategy trigger prediction network includes an active strategy trigger prediction network and a passive strategy trigger prediction network, and the trigger prediction probability includes the active trigger prediction probability and the passive trigger prediction probability; The step of determining a policy trigger prediction network according to the sample policy type tag, loading a sample characterization vector corresponding to the sample policy type tag into the policy trigger prediction network, and outputting a trigger prediction probability corresponding to the sample characterization vector based on the policy trigger prediction network includes: If the sample strategy type mark is an active mark, determining the active strategy trigger prediction network, and loading the sample representation vector corresponding to the active mark into the active strategy trigger prediction network, and outputting the active trigger prediction probability corresponding to the sample representation vector based on the active strategy trigger prediction network; If the sample strategy type mark is a passive mark, the passive strategy trigger prediction network is determined, and the sample representation vector corresponding to the passive mark is loaded into the passive strategy trigger prediction network, and the passive trigger prediction probability corresponding to the sample representation vector is output based on the passive strategy trigger prediction network; Acquire active cost according to the active mark and the active trigger prediction probability, and adjust network parameters of the active strategy trigger prediction network according to the active cost; Obtaining a passive cost according to the passive mark and the passive trigger prediction probability, and adjusting a network parameter of the passive strategy trigger prediction network according to the passive cost; The selecting, from the first number of the sample representation vectors, the sample representation vector whose trigger prediction probability is not less than the trigger prediction reference value to obtain the second number of the mixed sample representation vectors includes: Among the first number of sample characterization vectors, sample characterization vectors whose active trigger prediction probability and passive trigger prediction probability are not less than the trigger prediction reference value are selected to obtain a second number of the mixed sample characterization vectors.

6. The method according to claim 3, characterized in that If the uniformity supervision requirement is uniformity supervision of the probability average, then according to the uniformity supervision requirement, the second number of the mixed trigger prediction probabilities and the second number of the collaborative trigger prediction probabilities are obtained to obtain the first supervision cost, including: According to the sample strategy type mark, determine, among the second number of the mixed trigger prediction probabilities, an active mixed trigger prediction probability corresponding to the active sample strategy type mark, an active mixed quantity corresponding to the active mixed trigger prediction probability, a passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and a passive mixed quantity corresponding to the passive mixed trigger prediction probability; According to the sample strategy type mark, determine, among the second number of the collaboration trigger prediction probabilities, the active collaboration trigger prediction probability corresponding to the active sample strategy type mark, the active collaboration quantity corresponding to the active collaboration trigger prediction probability, the passive collaboration trigger prediction probability corresponding to the passive sample strategy type mark, and the passive collaboration quantity corresponding to the passive collaboration trigger prediction probability; According to the active mixing trigger prediction probability and the active mixing quantity, an active mixing mean is obtained, and according to the passive mixing trigger prediction probability and the passive mixing quantity, a passive mixing mean is obtained; According to the active cooperation trigger prediction probability and the active cooperation quantity, an active cooperation mean is obtained, and according to the passive cooperation trigger prediction probability and the passive cooperation quantity, a passive cooperation mean is obtained; Cost acquisition is performed on the active mixed mean, the passive mixed mean, the active collaborative mean and the passive collaborative mean to obtain the first supervision cost.

7. The method according to claim 6, characterized in that If the dispersion supervision requirement is the dispersion supervision of the probability distribution, then according to the dispersion supervision requirement, the cost acquisition is performed on the second number of the mixed trigger prediction probabilities to obtain the dispersion cost, including: Acquire active mixing discreteness according to the active mixing trigger prediction probability and the active mixing quantity, and acquire passive mixing discreteness according to the passive mixing trigger prediction probability and the passive mixing quantity; Acquire active coordination dispersion according to the active coordination trigger prediction probability and the active coordination quantity, and acquire passive coordination dispersion according to the passive coordination trigger prediction probability and the passive coordination quantity; Cost acquisition is performed on the active mixing discreteness, the active mixing mean, the passive mixing discreteness, the passive mixing mean, the active collaborative discreteness, the active collaborative mean, the passive collaborative discreteness and the passive collaborative mean to obtain the dispersion cost.

8. The method according to claim 3, characterized in that The step of acquiring the cost of the second number of the mixed trigger prediction probabilities and the trigger prediction probability according to the uniformity supervision requirement to obtain the second supervision cost includes: According to the sample strategy type mark, determine, among the second number of the mixed trigger prediction probabilities, an active mixed trigger prediction probability corresponding to the active sample strategy type mark, an active mixed quantity corresponding to the active mixed trigger prediction probability, a passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and a passive mixed quantity corresponding to the passive mixed trigger prediction probability; According to the sample strategy type mark, in the first number of the trigger prediction probabilities, determine the active trigger prediction probability corresponding to the active sample strategy type mark, the active quantity corresponding to the active trigger prediction probability, the passive trigger prediction probability corresponding to the passive sample strategy type mark, and the passive quantity corresponding to the passive trigger prediction probability; According to the active mixing trigger prediction probability and the active mixing quantity, an active mixing mean is obtained, and according to the passive mixing trigger prediction probability and the passive mixing quantity, a passive mixing mean is obtained; According to the active trigger prediction probability and the active number, an active mean is obtained, and according to the passive trigger prediction probability and the passive number, a passive mean is obtained; If the uniformity supervision requirement is uniformity supervision of the active probability average, then cost acquisition is performed on the active mixed mean, the passive mixed mean and the active mean to obtain the second supervision cost; If the uniformity supervision requirement is uniformity supervision of the passive probability average, the cost is acquired for the active mixed mean, the passive mixed mean and the passive mean to obtain the second supervision cost.

9. The method according to claim 4, characterized in that The step of obtaining the cost of the second number of the mixed trigger prediction probabilities and the actual trigger rate according to the uniformity supervision requirement to obtain the third supervision cost includes: According to the sample strategy type mark, determine, among the second number of the mixed trigger prediction probabilities, an active mixed trigger prediction probability corresponding to the active sample strategy type mark, an active mixed quantity corresponding to the active mixed trigger prediction probability, a passive mixed trigger prediction probability corresponding to the passive sample strategy type mark, and a passive mixed quantity corresponding to the passive mixed trigger prediction probability; According to the sample strategy type mark, in the first number of the actual trigger rates, determine the active actual trigger rate corresponding to the active sample strategy type mark, the number of active samples corresponding to the active actual trigger rate, the passive actual trigger rate corresponding to the passive sample strategy type mark, and the number of passive samples corresponding to the passive actual trigger rate; According to the active mixing trigger prediction probability and the active mixing quantity, an active mixing mean is obtained, and according to the passive mixing trigger prediction probability and the passive mixing quantity, a passive mixing mean is obtained; According to the active actual trigger rate and the active sample quantity, an active trigger mean is obtained, and according to the passive actual trigger rate and the passive sample quantity, a passive trigger mean is obtained; If the uniformity supervision requirement is uniformity supervision of the active probability average, then cost acquisition is performed on the active mixed mean, the passive mixed mean and the active triggered mean to obtain the third supervision cost; If the uniformity supervision requirement is uniformity supervision of a passive probability average, then cost acquisition is performed on the active mixed mean, the passive mixed mean, and the passive triggered mean to obtain the third supervision cost.

10. A temperature control strategy optimization system, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described in any one of claims 1 to 9.

Images