Lithium ion battery pack liquid cooling system temperature prediction method and system based on improved PSO-ABC-BP

Abstract

The invention relates to a lithium ion battery pack liquid cooling system temperature prediction method and system based on improved PSO-ABC-BP, and the method comprises the steps: carrying out the lumped modeling of a lithium ion battery pack liquid cooling system, and obtaining a lithium ion battery pack liquid cooling system model; acquiring operation data of the liquid cooling system of the lithium ion battery pack, and performing data preprocessing to form an operation data set of the liquid cooling system of the lithium ion battery pack; designing the structure of a BP neural network according to the battery pack temperature and the temperature influence parameters of the lithium ion battery pack liquid cooling system operation data set; using an improved PSO-ABC algorithm to optimize the initial weight and the threshold value of the BP neural network to obtain a PSO-ABC-BP lithium ion battery pack liquid cooling system temperature prediction model; the method comprises the following steps: acquiring the latest acquired actual operation data of the lithium ion battery pack liquid cooling system, inputting temperature influence parameters into a PSO-BP lithium ion battery pack liquid cooling system temperature prediction model to obtain a temperature prediction value, and combining an improved PSO-ABC algorithm with a BP neural network to seek a better parameter combination, thereby improving the accuracy of a prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of battery thermal management, and in particular to a temperature prediction method and system for a lithium-ion battery pack liquid cooling system based on an improved PSO-ABC-BP. Background Art

[0002] With the continuous advancement of science and technology, electric vehicles (EVs) and large-scale energy storage systems play an increasingly important role in the energy field. They provide a practical solution for achieving green energy and reducing carbon emissions. Lithium-ion batteries, due to their high energy density, long cycle life and relatively low self-discharge rate, have become indispensable energy storage devices in these systems. However, the performance and life of lithium-ion batteries are significantly affected by operating temperature. Excessively high temperatures may cause battery performance to deteriorate and even cause safety issues such as thermal runaway, while too low temperatures will affect the battery's charging efficiency and discharge capacity.

[0003] In electric vehicles and energy storage systems, battery packs usually operate at higher currents, which generates a lot of heat. If this heat cannot be dissipated effectively, it will cause the battery temperature to rise, affecting the electrochemical performance and life of the battery. Therefore, an efficient thermal management system is essential to maintain battery performance and ensure system safety. Although existing thermal management systems use various cooling technologies such as air cooling, liquid cooling, and phase change material cooling, they usually rely on empirical models or simple temperature control strategies, which often lack accurate predictions of the thermal dynamics inside the battery.

[0004] Traditional temperature prediction methods, such as physics-based models and data-driven models, can predict battery temperature to a certain extent, but they have certain limitations. Physical models are complex and difficult to model accurately, while data-driven models often require a large amount of experimental data for training. In addition, these methods cannot respond to rapid changes in battery temperature in real time, resulting in low prediction accuracy and slow response speed, which is difficult to meet the needs of practical applications. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in order to solve the problems existing in the prior art in the above-mentioned background technology, a temperature prediction method and system for a lithium-ion battery pack liquid cooling system based on an improved PSO-ABC-BP is provided, which combines an improved PSO-ABC algorithm with a BP neural network to seek a better parameter combination, thereby improving the accuracy of the prediction results.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a temperature prediction method of a lithium-ion battery pack liquid cooling system based on an improved PSO-ABC-BP, comprising the following steps:

[0007] Step 1: Perform lumped modeling on the lithium-ion battery pack liquid cooling system to obtain a lithium-ion battery pack liquid cooling system model; obtain the operating data of the lithium-ion battery pack liquid cooling system, and perform data preprocessing to form a lithium-ion battery pack liquid cooling system operating data set; the lithium-ion battery pack liquid cooling system operating data includes battery pack temperature and temperature influencing parameters, wherein the temperature influencing parameters include coolant temperature, coolant flow rate, battery pack voltage and battery pack power density;

[0008] Step 2: Design the structure of the BP neural network according to the battery pack temperature and temperature influencing parameters of the lithium-ion battery pack liquid cooling system operation data set obtained in step 1;

[0009] Step 3: Using the temperature influencing parameters in the lithium-ion battery pack liquid cooling system operation data set obtained in step 1 as input and temperature as output, training is performed, and the initial weights and thresholds of the BP neural network are optimized using the improved PSO-ABC algorithm to obtain a PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model;

[0010] Step 4: Obtain the latest actual operation data collected by the lithium-ion battery pack liquid cooling system, input the temperature influencing parameters into the PSO-BP lithium-ion battery pack liquid cooling system temperature prediction model obtained in step 3, and obtain the temperature prediction value.

[0011] Furthermore, in step 1, the data preprocessing process includes data normalization, and the formula is as follows:

[0012]

[0013] Among them, y is the normalized operating data of the lithium-ion battery pack liquid cooling system, y max =1,y min =0; x is the original lithium-ion battery pack liquid cooling system operation data to be normalized; x max and x min They are respectively the maximum value and the minimum value of the original lithium-ion battery pack liquid cooling system operation data to be processed in the lithium-ion battery pack liquid cooling system operation data set.

[0014] Furthermore, in step 3, the specific steps of improving the PSO-ABC algorithm are:

[0015] Step a: Change the traditional particle swarm learning factor to an adaptive learning factor;

[0016] Step b: Change the traditional PSO algorithm fixed mutation strategy to an adaptive mutation strategy;

[0017] Step c: Use the Q-learning algorithm to dynamically optimize the inertia weight in the PSO algorithm and the probability of selecting a food source in the ABC algorithm.

[0018] Furthermore, the specific steps of the adaptive learning factor in step a are:

[0019] Step a1: For the jth particle in the population, perform the following operations:

[0020] If the current fitness of a particle is equal to its historical optimal fitness, it means that the particle has not improved its optimal position in this round of iteration, so the update counter is incremented by one. If the two are not equal, the update counter is set to 0. Then based on the initial adjustment factor β init And the particle update counter Count(j) to adjust the factor β:

[0021]

[0022] Where: β init is the initial adjustment factor; β is the adjustment factor; Count(j) is the update counter of the jth particle; i is the current iteration number;

[0023] Step a2: Adaptively adjust the individual learning factor c1 and the population learning factor c2:

[0024] c1=c init_1 +β;

[0025] c2=c init_2 +β;

[0026] Where: c init_1 is the initial value of the individual learning factor; c init_2 is the initial value of the population learning factor.

[0027] Furthermore, the specific steps of the adaptive mutation strategy in step b are:

[0028] Step b1: Determination of mutation position: randomly select a position x, which represents a dimension of the particle in the parameter space;

[0029] Step b2: Adaptive adjustment of mutation probability: dynamically adjust the mutation probability according to the fitness change of the particle and the number of iterations;

[0030] Step b3: Implement mutation: decide whether to mutate the selected position based on the mutation probability determined by adaptiveness; if mutation is decided, the value of the particle at the position will be updated according to the set mutation strategy;

[0031] The execution formula of the mutation is as follows:

[0032]

[0033] Where: is the position of particle j in the dth dimension in the tth iteration; x max,d and x min,d are the maximum and minimum values ​​of the dimension respectively, and λ is a value randomly drawn from a uniform distribution [0,1].

[0034] Furthermore, in step c, the core formula of using the Q-learning algorithm to dynamically optimize the PSO algorithm is as follows:

[0035] Calculate the index of the inertia weight:

[0036]

[0037] For the tth iteration, the Q value of the jth particle is updated:

[0038] Q PSO,t+1 (j,w idx,t+1 )=Q PSO,t (j,w idx,t )+α(R PSO,t +γmax n Q PSO,t (j,n)-Q PSO,t (j,w idx,t ));

[0039] Then adjust the inertia weight:

[0040]

[0041] Where: Q PSO,t (j,w idx,t ) represents the Q table value of particle j in the tth iteration; w min and w max Respectively represent the minimum and maximum values ​​of the inertia weight; w t represents the inertia weight of the particle in the tth iteration; w idx,t represents the index of the particle's inertia weight in the Q table in the tth iteration; R PSO,t represents the reward value in the tth iteration; max n Q PSO,t (j,n) means that in the tth iteration, for the jth particle, all possible inertia weight values ​​n are traversed and the maximum Q value is found; N represents the population size, that is, the number of particles; α is the learning rate, which is used to control the speed at which new information overwrites old information; γ is the discount factor, which is used to evaluate the current value of future rewards.

[0042] Furthermore, in step c, the core formula of dynamically optimizing the ABC algorithm using the Q-learning algorithm is as follows:

[0043] Calculate the index of the probability of selecting a food source:

[0044]

[0045] Q-learning update: Update the Q table using the following formula:

[0046] Q ABC,t+1 (k,w idx_ABC,t+1 )=Q ABC,t (k,w idx_ABC,t )+α(R ABC,t +γmax m Q ABC,t (k,m)-Q ABC,t (j,w idx_ABC,t ));

[0047] The probability of the bee choosing a food source is then adjusted based on the updated Q table value:

[0048]

[0049] Where: Q ABC,t (k,w idx_ABC,t ) represents the Q table value of bee k in the tth iteration; p min and p max Respectively represent the minimum and maximum probability of selecting a food source; p best,t (k) represents the probability that bee k chooses the food source in the tth iteration; w idx,t represents the index of the particle's inertia weight in the Q table in the tth iteration; R ABC,t represents the reward value in the tth iteration; max m Q ABC,t (k,m) means that in the tth iteration, for the kth bee, all possible probability values ​​m are traversed and the maximum Q value is found; M represents the number of bees; α is the learning rate; γ is the discount factor.

[0050] Furthermore, in step 3, the specific steps of using the improved PSO-ABC algorithm to optimize the initial weights and thresholds of the BP neural network are as follows:

[0051] Step 3.1: Initialize the weights and threshold length of the BP neural network, and use the real number coding mechanism to connect the network parameters. The real number coding length calculation formula is:

[0052] L=n input ×n hidden +n hidden +n hidden ×n output +n output ;

[0053] Where L is the code length; n inputis the number of input layer nodes set in the BP neural network; n hidden is the number of hidden layer nodes; n output is the number of nodes in the output layer;

[0054] Step 3.2: Parameter initialization: including population and speed initialization: randomly generate the initial population and speed, initialize the fitness value to infinity, and record the number of individual optimal position updates;

[0055] ABC population initialization: randomly generate the initial solution of the artificial bee colony, initialize the fitness value to infinity, and initialize the global optimal solution;

[0056] Individual best and global best initialization: Initialize the individual best position as the population position, and initialize the global best fitness value and position;

[0057] Q-table initialization: Initialize Q-table for PSO algorithm and ABC algorithm to store and update strategies during learning process;

[0058] PSO velocity initialization formula:

[0059] Where rand() is a random number uniformly distributed in the interval [0,1);

[0060] Step 3.3: Calculate the fitness value F of each individual in the i-th generation population according to the fitness function i , where the fitness function is calculated as:

[0061]

[0062] In the formula, F is the fitness value, N is the number of training set samples, and d i and d′ i are the expected output value and actual output value of the i-th node of the BP neural network respectively;

[0063] Step 3.4: Iterative optimization: Perform the maximum number of iterations, each of which includes updating the learning factor, speed and position, calculating the fitness value, and performing parameter adaptive adjustment based on the Q-learning algorithm;

[0064] Step 3.5: Update individual and global optimal positions: In each iteration, update the individual optimal position and global optimal position of each particle according to the fitness value;

[0065] Step 3.6: Information exchange: Feedback the better solution found by the ABC algorithm to the PSO algorithm to optimize the search process and update the optimal fitness value;

[0066] Step 3.7: Determine the iteration termination condition: If the number of iterations reaches the preset maximum number of iterations or meets other termination conditions, stop the iteration; otherwise, return to step 3.3 to continue the iteration;

[0067] Step 3.8: Extract the optimal initial weights and thresholds: Extract weights and thresholds from the global optimal solution for initialization of the BP neural network.

[0068] Furthermore, after the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model is obtained in step 3, the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model is also evaluated. The parameter indicators used in the evaluation include: absolute value error MAE, mean square error MSE, determination coefficient R 2 , root mean square error RMSE, where MAE, MSE, RMSE and R 2 They are defined as:

[0069]

[0070]

[0071]

[0072]

[0073] Where: y i Represents the actual value of the average temperature of the lithium-ion battery pack; Represents the predicted value of the average temperature of the lithium-ion battery pack; represents the average value of the average temperature of the lithium-ion battery pack; n represents the number of test set samples.

[0074] A temperature prediction system for a lithium-ion battery pack liquid cooling system based on an improved PSO-ABC-BP, comprising:

[0075] The data set acquisition module is used to acquire the operating data of the liquid cooling system of the lithium-ion battery pack and perform data preprocessing to form the operating data set of the liquid cooling system of the lithium-ion battery pack; the operating data of the liquid cooling system of the lithium-ion battery pack includes the battery pack temperature and temperature influencing parameters;

[0076] A neural network structure design module is used to design the structure of a BP neural network based on the temperature and temperature-influencing parameters of a lithium-ion battery pack liquid cooling system operation data set;

[0077] The model training and optimization module is used to use the temperature influencing parameters in the lithium-ion battery pack liquid cooling system operation data set as input and temperature as output for training. The improved PSO-ABC algorithm is used to optimize the initial weights and thresholds of the BP neural network to obtain the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model;

[0078] The temperature prediction module is used to obtain new operating data of the lithium-ion battery pack liquid cooling system, input the temperature influencing parameters into the PSO-BP lithium-ion battery pack liquid cooling system temperature prediction model, and obtain the temperature prediction value.

[0079] Beneficial effects of the present invention: The present invention collects the operating data of the lithium-ion battery pack liquid cooling system, constructs a BP neural network structure based on the data, optimizes the neural network using an improved PSO-ABC algorithm, assigns the optimal individual generated after the optimization to the BP neural network, and obtains a PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model. According to this model, the future battery pack temperature can be predicted, and then the system temperature change condition can be evaluated.

[0080] When the coolant temperature, coolant flow rate, battery pack voltage and battery pack power density of the lithium-ion battery pack liquid cooling system are used as data sets to train the model, the traditional BP neural network is prone to fall into the local optimal solution, while the improved PSO-ABC as a global optimization method can help overcome this problem.

[0081] The improved PSO-ABC-BP method combines the improved PSO-ABC with the BP neural network, enabling the model to perform better global search, thereby finding a better parameter combination and improving the accuracy of prediction.

[0082] The improved PSO-ABC-BP algorithm enhances the algorithm's search capability by adaptively adjusting the learning factor and dynamically updating the inertia weight, enabling it to quickly find the optimal solution under complex temperature change conditions. At the same time, the Q-learning algorithm is introduced to adaptively adjust the parameters of the PSO and ABC algorithms, further improving the algorithm's intelligence and prediction accuracy. This method can achieve real-time and accurate prediction of the temperature of the liquid cooling system of the lithium-ion battery pack, providing an effective solution for battery thermal management, helping to improve battery performance and life, and ensuring the safety and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0084] Figure 1 A schematic diagram of the steps of the method for predicting the temperature of a liquid cooling system of a lithium-ion battery pack according to the present invention;

[0085] Figure 2 The flowchart of the PSO-ABC optimization BP neural network model improved by the present invention;

[0086] Figure 3 shows the comparison between the prediction results of the algorithm model and the actual results: (a) BP; (b) LSTM; (c) FNN; (d) PSO-BP; (e) PSO-ABC-BP improved by the present invention;

[0087] Figure 4 To improve the fitness curve of the temperature prediction model of the PSO-ABC-BP lithium-ion battery pack liquid cooling system with the number of iterations;

[0088] Figure 5 A structural framework diagram of a temperature prediction system for a lithium-ion battery pack liquid cooling system based on an improved PSO-ABC-BP provided by the present invention. DETAILED DESCRIPTION

[0089] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0090] Embodiment 1:

[0091] A temperature prediction method for a lithium-ion battery pack liquid cooling system based on an improved PSO-ABC-BP algorithm comprises the following steps:

[0092] Step 1: Perform lumped modeling on the lithium-ion battery pack liquid cooling system to obtain a lithium-ion battery pack liquid cooling system model; obtain the operating data of the lithium-ion battery pack liquid cooling system, and perform data preprocessing to form a lithium-ion battery pack liquid cooling system operating data set; the lithium-ion battery pack liquid cooling system operating data includes battery pack temperature and temperature influencing parameters, wherein the temperature influencing parameters include coolant temperature, coolant flow rate, battery pack voltage and battery pack power density. In the specific implementation process, the lithium-ion battery pack liquid cooling system operating data sampling time interval can be set to 1 second;

[0093] Among them, the process of data preprocessing includes data normalization, and the formula is as follows:

[0094]

[0095] Among them, y is the normalized operating data of the lithium-ion battery pack liquid cooling system, y max =1,y min =0; x is the original lithium-ion battery pack liquid cooling system operation data to be normalized; x max and x minThey are respectively the maximum value and the minimum value of the original lithium-ion battery pack liquid cooling system operation data to be processed in the lithium-ion battery pack liquid cooling system operation data set.

[0096] Step 2: According to the battery pack temperature and temperature influencing parameters of the lithium-ion battery pack liquid cooling system operation data set obtained in step 1, the structure of the BP neural network is designed, with 4 input layer nodes, 5 hidden layer nodes, and 1 output layer node;

[0097] Step 3: Take the temperature influencing parameters in the lithium-ion battery pack liquid cooling system operation data set obtained in step 1 as input and temperature as output for training. Use the improved PSO-ABC algorithm to optimize the initial weights and thresholds of the BP neural network. Continuously iterate and optimize according to the PSO-ABC algorithm to obtain the optimal individual. Assign the initial weights and thresholds corresponding to this individual to the BP network to obtain the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model.

[0098] Among them, the PSO-ABC algorithm improvements include the following three aspects:

[0099] (1) Change the traditional particle swarm learning factor to an adaptive learning factor;

[0100] The diversity and search ability of the population are enhanced by adaptively adjusting the learning factor. In each iteration, the algorithm first evaluates the current fitness of each particle and the historical optimal fitness, updates the particle's update counter based on the evaluation results, and then calculates the adjustment factor and dynamically adjusts the learning factors c1 and c2.

[0101] The updated learning factors c1 and c2 are used to adjust the speed and position of the particles to optimize the search process. This adjustment strategy based on the particle's historical performance allows the particles to explore a wider solution space in the early stages of the iteration, and to focus on a more refined search in the later stages, thereby effectively avoiding the dilemma of premature convergence and local optimal solutions. In this way, the algorithm can more flexibly guide particles to search in the solution space, improving the efficiency and accuracy of the particle swarm optimization algorithm in solving complex optimization problems.

[0102] Among them, the specific steps of the adaptive learning factor are:

[0103] ① For the jth particle in the population, perform the following operations:

[0104] If the current fitness of a particle is equal to its historical optimal fitness, it means that the particle has not improved its optimal position in this round of iteration, so the update counter is incremented by one. If the two are not equal, the update counter is set to 0. Then based on the initial adjustment factor β init And the particle update counter Count(j) to adjust the factor β:

[0105]

[0106] Where: β init is the initial adjustment factor; β is the adjustment factor; Count(j) is the update counter of the jth particle; i is the current iteration number;

[0107] ②: Adaptively adjust the individual learning factor c1 and the population learning factor c2:

[0108] c1=c init_1 +β;

[0109] c2=c init_2 +β;

[0110] Where: c init_1 is the initial value of the individual learning factor; c init_2 is the initial value of the population learning factor.

[0111] (2) Changing the traditional PSO algorithm fixed mutation strategy to an adaptive mutation strategy;

[0112] In traditional PSO algorithms, the mutation strategy is usually fixed, that is, the particles are mutated according to preset probabilities and rules. This fixed mutation strategy may not be sufficient to cope with complex optimization problems in some cases, because it cannot be adjusted according to the actual situation of the optimization process. To solve this problem, an adaptive mutation strategy can be introduced to dynamically adjust the mutation behavior according to the real-time information in the optimization process. In the adaptive mutation strategy, the mutation behavior of the particles is dynamically adjusted according to the historical performance of the particles and the current search status. The following are the implementation steps of an adaptive mutation strategy:

[0113] ①: Determination of mutation position: Randomly select a position x, which represents a dimension of the particle in the parameter space;

[0114] ②: Adaptive adjustment of mutation probability: Dynamically adjust the mutation probability according to the change of particle fitness and the number of iterations. For example, if the fitness of a particle has not improved for a long time, the mutation probability can be increased to promote the particle to jump out of the local optimal solution;

[0115] ③: Implement mutation: decide whether to mutate the selected position based on the mutation probability determined by adaptation; if mutation is decided, the value of the particle at that position will be updated according to the set mutation strategy;

[0116] The execution formula of the mutation is as follows:

[0117]

[0118] Where: is the position of particle j in the dth dimension at the tth iteration; is the position of particle j in the dth dimension at the t+1th iteration; x max,d and x min,d are the maximum and minimum values ​​of the dimension, respectively, λ is a value randomly drawn from a uniform distribution [0,1], P adapt is the adaptive mutation probability, which determines whether to mutate the value of the particle at this position. adapt <λ, then perform mutation operation. adapt Take a fixed value, the experimental value in this paper is 0.85;

[0119] (3) Use the Q-learning algorithm to dynamically optimize the inertia weight in the PSO algorithm and the probability of selecting a food source in the ABC algorithm;

[0120] In the PSO algorithm, the Q-learning algorithm is used to dynamically adjust the inertia weight w to optimize the search behavior of the particles. In the ABC algorithm, the Q-learning algorithm is used to dynamically adjust the probability p of the bee choosing a food source. best , to improve the search efficiency. The following is a summary of parameter optimization in the two algorithms:

[0121] In the parameter initialization phase, the inertia weight w of the PSO algorithm and the selection probability p of the ABC algorithm best All are randomly set within a reasonable range, providing the algorithm with a variety of initial exploration points. As the iteration proceeds, the Q-learning algorithm updates the Q table according to the fitness changes of the individuals, and then dynamically adjusts the inertia weight and selection probability.

[0122] ①: The core formula of the PSO algorithm dynamically optimized using the Q-learning algorithm is as follows:

[0123] Calculate the index of the inertia weight:

[0124]

[0125] For the tth iteration, the Q value of the jth particle is updated:

[0126] Q PSO,t+1 (j,w idx,t+1 )=Q PSO,t (j,w idx,t )+α(R PSO,t +γmax n Q PSO,t (j,n)-Q PSO,t (j,w idx,t ));

[0127] Then adjust the inertia weight:

[0128]

[0129] Where: Q PSO,t+1 (j,w idx,t+1 ) represents the Q table value of particle j in the t+1th iteration; Q PSO,t (j,w idx,t ) represents the Q table value of particle j in the tth iteration; w min and w max Respectively represent the minimum and maximum values ​​of the inertia weight; w t represents the inertia weight of the particle in the tth iteration; w t+1 represents the inertia weight of the particle in the t+1th iteration; w idx,t represents the index of the particle's inertia weight in the Q table in the tth iteration; R PSO,t represents the reward value in the tth iteration; max n Q PSO,t (j,n) means that in the tth iteration, for the jth particle, all possible inertia weight values ​​n are traversed and the maximum Q value is found; N represents the population size, that is, the number of particles; α is the learning rate, which is used to control the speed at which new information overwrites old information; γ is the discount factor, which is used to evaluate the current value of future rewards.

[0130] ②The core formula of dynamically optimizing the ABC algorithm using the Q-learning algorithm is as follows:

[0131] Calculate the index of the probability of selecting a food source:

[0132]

[0133] Q-learning update: Update the Q table using the following formula:

[0134] Q ABC,t+1 (k,w idx_ABC,t+1 )=Q ABC,t (k,w idx_ABC,t )+α(R ABC,t +γmax m Q ABC,t (k,m)-Q ABC,t (k,w idx_ABC,t ));

[0135] The probability of the bee choosing a food source is then adjusted based on the updated Q table value:

[0136]

[0137] Where: Q ABC,t (k,w idx_ABC,t ) represents the Q table value of bee k in the tth iteration; QABC,t+1 (k,w idx_ABC,t+1 ) represents the Q table value of bee k in the t+1th iteration; p min and p max Respectively represent the minimum and maximum probability of selecting a food source; p best,t (k) represents the probability that bee k chooses the food source in the tth iteration; p best,t+1 (k) represents the probability that bee k chooses the food source in the t+1th iteration; w idx_ABC,t represents the index of the probability of the kth bee selecting the food source in the tth iteration (the index is used to find and update the corresponding Q value in the Q table); R ABC,t represents the reward value in the tth iteration; max m Q ABC,t (k,m) means that in the tth iteration, for the kth bee, all possible probability values ​​m are traversed and the maximum Q value is found; M represents the number of bees; α is the learning rate; γ is the discount factor.

[0138] Among them, the improved PSO-ABC algorithm is used, such as Figure 2 As shown in Figure 1, the specific steps for optimizing the initial weights and thresholds of the BP neural network are:

[0139] Step 3.1: Initialize the weights and threshold length of the BP neural network, and use the real number coding mechanism to connect the network parameters, that is, the individual is composed of the input layer weight, input layer threshold, output layer weight, and output layer threshold of the temperature prediction model of the lithium-ion battery pack liquid cooling system based on the BP neural network. The real number coding length calculation formula is:

[0140] L=n input ×n hidden +n hidden +n hidden ×n output +n output ;

[0141] Where L is the code length; n input is the number of input layer nodes set in the BP neural network; n hidden is the number of hidden layer nodes; n output is the number of nodes in the output layer; from step 2, we know that n input =4,n hidden =5,n output =1, L=31;

[0142] Step 3.2: Parameter initialization: c init_1 =c init_2 =4.494, β init =1,M=N=50,V max =xmax =1, V min =x min =-1,w max =w0=0.9,w min =0.4, α=0.1, γ=0.9, P adapt =0.85, the number of iterations is set to 100 times; where V max Represents the maximum velocity of each particle; V min Indicates the minimum speed of each particle;

[0143] Including population and speed initialization: randomly generate the initial population and speed, initialize the fitness value to infinity, and record the number of individual optimal position updates;

[0144] ABC population initialization: randomly generate the initial solution of the artificial bee colony, initialize the fitness value to infinity, and initialize the global optimal solution;

[0145] Individual best and global best initialization: Initialize the individual best position as the population position, and initialize the global best fitness value and position;

[0146] Q-table initialization: Initialize Q-table for PSO algorithm and ABC algorithm to store and update strategies during learning process;

[0147] PSO velocity initialization formula:

[0148] Where rand() is a random number uniformly distributed in the interval [0,1);

[0149] Step 3.3: Calculate the fitness value F of each individual in the i-th generation population according to the fitness function i , where the fitness function is calculated as:

[0150]

[0151] In the formula, F is the fitness value, N is the number of training set samples, and d i and d′ i are the expected output value and actual output value of the i-th node of the BP neural network respectively;

[0152] Step 3.4: Iterative optimization: Perform the maximum number of iterations, each of which includes updating the learning factor, speed and position, calculating the fitness value, and performing parameter adaptive adjustment based on the Q-learning algorithm;

[0153] Step 3.5: Update individual and global optimal positions: In each iteration, update the individual optimal position and global optimal position of each particle according to the fitness value, such as Figure 4 As shown;

[0154] Step 3.6: Information exchange: Feedback the better solution found by the ABC algorithm to the PSO algorithm to optimize the search process and update the optimal fitness value;

[0155] Step 3.7: Determine the iteration termination condition: If the number of iterations reaches the preset maximum number of iterations or meets other termination conditions, stop the iteration; otherwise, return to step 3.3 to continue the iteration;

[0156] Step 3.8: Extract the optimal initial weights and thresholds: Extract weights and thresholds from the global optimal solution for initialization of the BP neural network.

[0157] Step 4: Obtain the latest actual operation data collected by the lithium-ion battery pack liquid cooling system, input the temperature influencing parameters into the PSO-BP lithium-ion battery pack liquid cooling system temperature prediction model obtained in step 3, and obtain the temperature prediction value.

[0158] The collected lithium-ion battery pack liquid cooling system operation data includes 1,000 sets of data, including coolant temperature, coolant flow rate, battery pack body power density, and battery pack voltage data. 80% of the 1,000 sets of data are divided into training data sets and 20% are divided into test data sets.

[0159] The training data set is used to train the BP neural network in step 3 above. The input values ​​during training are the coolant temperature, coolant flow rate, battery pack body power density, and battery pack voltage of the first 800 groups of data, and the output value is the average battery pack temperature of the first 800 groups.

[0160] The test data set is used to test and evaluate the generated temperature prediction model for the liquid cooling system of lithium-ion battery packs. The input values ​​during the test are the coolant temperature, coolant flow rate, battery pack body power density, and battery pack voltage of the last 200 sets of data, and the output value is the average temperature of the battery packs of the last 200 sets. The comparison of the test results using the test set is shown in Figure 3.

[0161] After the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model is obtained in step 3, the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model is evaluated according to the test results. The parameter indicators used in the evaluation include: absolute value error MAE, mean square error MSE, determination coefficient R 2 , root mean square error RMSE, where MAE, MSE, and RMSE are all errors. The smaller the value, the more accurate the model. 2 For measuring regression models, R 2 ∈[0,1],R 2 The closer to 1, the better the model fit; MAE, MSE, RMSE and R2 They are defined as:

[0162]

[0163]

[0164]

[0165]

[0166] Where: y i Represents the actual value of the average temperature of the lithium-ion battery pack; Represents the predicted value of the average temperature of the lithium-ion battery pack; represents the average value of the average temperature of the lithium-ion battery pack; n represents the number of test set samples.

[0167] The experiment selected classic prediction models as controls: BP neural network, long short-term memory network (LSTM), feedforward neural network (FNN), PSO-BP, and compared the prediction performance with the method of the present invention (improved PSO-ABC-BP). Figure 3 is a comparison curve of the predicted value and actual value of temperature for the five models. The comparison results of the evaluation indicators of the five models are shown in Table 1.

[0168] Table 1 Comparison of evaluation indicators of BP, LSTM, FNN, PSO-BP, and improved PSO-ABC-BP See Table

[0169]

[0170] In summary, the prediction method for the liquid cooling system of a lithium-ion battery pack based on the improved PSO-ABC-BP proposed in the present invention can obtain a lower prediction error than the existing method and improve the temperature prediction accuracy. The scheme proposed in the present invention has the lowest error in the selected experimental data, which proves that the method has good applicability. Experiments show that the method of the present invention has high prediction accuracy. Compared with BP, LSTM, FNN, and PSO-BP in traditional prediction algorithms, the improved PSO-ABC-BP of the present invention has the smallest average absolute error, mean square error, and root mean square error for data prediction, and the determination coefficient R 2 The model reduces the amount of computation and shows better prediction performance.

[0171] Embodiment 2:

[0172] like Figure 5 As shown, a temperature prediction system for a lithium-ion battery pack liquid cooling system based on an improved PSO-ABC-BP comprises:

[0173] The data set acquisition module is used to acquire the operating data of the liquid cooling system of the lithium-ion battery pack and perform data preprocessing to form the operating data set of the liquid cooling system of the lithium-ion battery pack; the operating data of the liquid cooling system of the lithium-ion battery pack includes the battery pack temperature and temperature influencing parameters;

[0174] A neural network structure design module is used to design the structure of a BP neural network based on the temperature and temperature-influencing parameters of a lithium-ion battery pack liquid cooling system operation data set;

[0175] The model training and optimization module is used to use the temperature influencing parameters in the lithium-ion battery pack liquid cooling system operation data set as input and temperature as output for training. The improved PSO-ABC algorithm is used to optimize the initial weights and thresholds of the BP neural network to obtain the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model;

[0176] The temperature prediction module is used to obtain new operating data of the lithium-ion battery pack liquid cooling system, input the temperature influencing parameters into the PSO-BP lithium-ion battery pack liquid cooling system temperature prediction model, and obtain the temperature prediction value.

[0177] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A temperature prediction method for a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP algorithm, characterized in that: The following steps are involved: Step 1: Perform lumped modeling on the lithium-ion battery pack liquid cooling system to obtain a lithium-ion battery pack liquid cooling system model; obtain the operating data of the lithium-ion battery pack liquid cooling system, and perform data preprocessing to form a lithium-ion battery pack liquid cooling system operating data set; the lithium-ion battery pack liquid cooling system operating data includes battery pack temperature and temperature influencing parameters, wherein the temperature influencing parameters include coolant temperature, coolant flow rate, battery pack voltage and battery pack power density; Step 2: Design the structure of the BP neural network according to the battery pack temperature and temperature influencing parameters of the lithium-ion battery pack liquid cooling system operation data set obtained in step 1; Step 3: Using the temperature influencing parameters in the lithium-ion battery pack liquid cooling system operation data set obtained in step 1 as input and temperature as output, training is performed, and the initial weights and thresholds of the BP neural network are optimized using the improved PSO-ABC algorithm to obtain a PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model; Step 4: Obtain the latest actual operation data collected by the lithium-ion battery pack liquid cooling system, input the temperature influencing parameters into the PSO-BP lithium-ion battery pack liquid cooling system temperature prediction model obtained in step 3, and obtain the temperature prediction value.

2. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 1, characterized in that: In step 1, the data preprocessing process includes data normalization, and the formula is as follows: Among them, y is the normalized operating data of the lithium-ion battery pack liquid cooling system, y max =1,y min =0; x is the original lithium-ion battery pack liquid cooling system operation data to be normalized; x max and x min They are respectively the maximum value and the minimum value of the original lithium-ion battery pack liquid cooling system operation data to be processed in the lithium-ion battery pack liquid cooling system operation data set.

3. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 1, characterized in that: In step 3, the specific steps of improving the PSO-ABC algorithm are: Step a: Change the traditional particle swarm learning factor to an adaptive learning factor; Step b: Change the traditional PSO algorithm fixed mutation strategy to an adaptive mutation strategy; Step c: Use the Q-learning algorithm to dynamically optimize the inertia weight in the PSO algorithm and the probability of selecting a food source in the ABC algorithm.

4. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 3, characterized in that: The specific steps of the adaptive learning factor in step a are: Step a1: For the jth particle in the population, perform the following operations: If the current fitness of a particle is equal to its historical optimal fitness, it means that the particle has not improved its optimal position in this round of iteration, so the update counter is incremented by one. If the two are not equal, the update counter is set to 0. Then based on the initial adjustment factor β init And the particle update counter Count(j) to adjust the factor β: Where: β init is the initial adjustment factor; β is the adjustment factor; Count(j) is the update counter of the jth particle; i is the current iteration number; Step a2: Adaptively adjust the individual learning factor c1 and the population learning factor c2: c1=c init_1 +β; c2=c init_2 +β; Where: c init_1 is the initial value of the individual learning factor; c init_2 is the initial value of the population learning factor.

5. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 3, characterized in that: The specific steps of the adaptive mutation strategy in step b are: Step b1: Determination of mutation position: randomly select a position x, which represents a dimension of the particle in the parameter space; Step b2: Adaptive adjustment of mutation probability: dynamically adjust the mutation probability according to the fitness change of the particle and the number of iterations; Step b3: Implement mutation: decide whether to mutate the selected position based on the mutation probability determined by adaptiveness; if mutation is decided, the value of the particle at the position will be updated according to the set mutation strategy; The execution formula of the mutation is as follows: Where: is the position of particle j in the dth dimension in the tth iteration; x max,d and x min,d are the maximum and minimum values ​​of the dimension respectively, and λ is a value randomly drawn from a uniform distribution [0,1].

6. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 3, characterized in that: The core formula of using Q-learning algorithm to dynamically optimize PSO algorithm in step c is as follows: Calculate the index of the inertia weight: For the tth iteration, the Q value of the jth particle is updated: Q PSO,t+1 (j,w idx,t+1 )=Q PSO,t (j,w idx,t )+α(R PSO,t +γmax n Q PSO,t (j,n)-Q PSO,t (j,w idx,t )); Then adjust the inertia weight: Where: Q PSO,t (j,w idx,t ) represents the Q table value of particle j in the tth iteration; w min and w max Represent the minimum and maximum values ​​of the inertia weight respectively; w t represents the inertia weight of the particle in the tth iteration; w idx,t represents the index of the particle's inertia weight in the Q table in the tth iteration; R PSO,t represents the reward value in the tth iteration; max n Q PSO,t (j,n) means that in the tth iteration, for the jth particle, all possible inertia weight values ​​n are traversed and the maximum Q value is found; N represents the population size, that is, the number of particles; α is the learning rate, which is used to control the speed at which new information overwrites old information; γ is the discount factor, which is used to evaluate the current value of future rewards.

7. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 3, characterized in that: The core formula of the ABC algorithm dynamically optimized using the Q-learning algorithm in step c is as follows: Calculate the index of the probability of selecting a food source: Q-learning update: Update the Q table using the following formula: Q ABC,t+1 (k,w idx_ABC,t+1 )=Q ABC,t (k,w idx_ABC,t )+α(R ABC,t +γmax m Q ABC,t (k,m) -Q ABC,t (j,w idx_ABC,t )); The probability of the bee choosing a food source is then adjusted based on the updated Q table value: Where: Q ABC,t (k,w idx_ABC,t ) represents the Q table value of bee k in the tth iteration; p min and p max Respectively represent the minimum and maximum probability of selecting a food source; p best,t (k) represents the probability that bee k chooses the food source in the tth iteration; w idx,t represents the index of the particle's inertia weight in the Q table in the tth iteration; R ABC,t represents the reward value in the tth iteration; max m Q ABC,t (k,m) means that in the tth iteration, for the kth bee, all possible probability values ​​m are traversed and the maximum Q value is found; M represents the number of bees; α is the learning rate; γ is the discount factor.

8. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 1, characterized in that: In step 3, the specific steps of using the improved PSO-ABC algorithm to optimize the initial weights and thresholds of the BP neural network are as follows: Step 3.1: Initialize the weights and threshold length of the BP neural network, and use the real number coding mechanism to connect the network parameters. The real number coding length calculation formula is: L=n input ×n hidden +n hidden +n hidden ×n output +n output ; Where L is the code length; n input is the number of input layer nodes set in the BP neural network; n hidden is the number of hidden layer nodes; n output is the number of nodes in the output layer; Step 3.2: Parameter initialization: including population and speed initialization: randomly generate the initial population and speed, initialize the fitness value to infinity, and record the number of individual optimal position updates; ABC population initialization: randomly generate the initial solution of the artificial bee colony, initialize the fitness value to infinity, and initialize the global optimal solution; Individual best and global best initialization: Initialize the individual best position as the population position, and initialize the global best fitness value and position; Q-table initialization: Initialize Q-table for PSO algorithm and ABC algorithm to store and update strategies during learning process; PSO velocity initialization formula: Where rand() is a random number uniformly distributed in the interval [0,1); Step 3.3: Calculate the fitness value F of each individual in the i-th generation population according to the fitness function i , where the fitness function is calculated as: In the formula, F is the fitness value, N is the number of training set samples, and d i and d′ i are the expected output value and actual output value of the i-th node of the BP neural network respectively; Step 3.4: Iterative optimization: Perform the maximum number of iterations, each of which includes updating the learning factor, speed and position, calculating the fitness value, and performing parameter adaptive adjustment based on the Q-learning algorithm; Step 3.5: Update individual and global optimal positions: In each iteration, update the individual optimal position and global optimal position of each particle according to the fitness value; Step 3.6: Information exchange: Feedback the better solution found by the ABC algorithm to the PSO algorithm to optimize the search process and update the optimal fitness value; Step 3.7: Determine the iteration termination condition: If the number of iterations reaches the preset maximum number of iterations or meets other termination conditions, stop the iteration; otherwise, return to step 3.3 to continue the iteration; Step 3.8: Extract the optimal initial weights and thresholds: Extract weights and thresholds from the global optimal solution for initialization of the BP neural network.

9. The temperature prediction method of a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP according to claim 1, characterized in that: After the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model is obtained in step 3, the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model is also evaluated. The parameter indicators used in the evaluation include: absolute value error MAE, mean square error MSE, determination coefficient R 2 , root mean square error RMSE, where MAE, MSE, RMSE and R 2 They are defined as: Where: y i Represents the actual value of the average temperature of the lithium-ion battery pack; Represents the predicted value of the average temperature of the lithium-ion battery pack; represents the average value of the average temperature of the lithium-ion battery pack; n represents the number of test set samples.

10. A temperature prediction system for a lithium-ion battery liquid cooling system based on an improved PSO-ABC-BP, characterized in that: include: The data set acquisition module is used to acquire the operating data of the liquid cooling system of the lithium-ion battery pack and perform data preprocessing to form an operating data set of the liquid cooling system of the lithium-ion battery pack; The operating data of the lithium-ion battery pack liquid cooling system includes the battery pack temperature and temperature-affecting parameters; A neural network structure design module is used to design the structure of a BP neural network based on the temperature and temperature-influencing parameters of a lithium-ion battery pack liquid cooling system operation data set; The model training and optimization module is used to use the temperature influencing parameters in the lithium-ion battery pack liquid cooling system operation data set as input and temperature as output for training. The improved PSO-ABC algorithm is used to optimize the initial weights and thresholds of the BP neural network to obtain the PSO-ABC-BP lithium-ion battery pack liquid cooling system temperature prediction model; The temperature prediction module is used to obtain new operating data of the lithium-ion battery pack liquid cooling system, input the temperature influencing parameters into the PSO-BP lithium-ion battery pack liquid cooling system temperature prediction model, and obtain the temperature prediction value.

Images