Prediction method and device for nuclear capacity discharge of storage battery

Abstract

The invention relates to the field of storage battery capacity detection, and provides a method for predicting nuclear capacity discharge of a storage battery. The method comprises the following steps: firstly, sampling by adopting a traditional static discharge test method to obtain actual discharge capacity time sequence data of a storage battery pack, and preprocessing the time sequence data; then, constructing a hyper-parameter-based long short-term memory (LSTM) network model, and exploring global optimal hyper-parameters of the LSTM algorithm model by adopting a novel particle swarm optimization (PSO) algorithm to obtain an optimized LSTM algorithm model; and finally, carrying out network training and data prediction on the optimized LSTM algorithm model by adopting the obtained training set data and test set data, and verifying the prediction effect of the storage battery nuclear capacity discharge prediction model. Through verification, complete discharge data of the storage battery can be accurately predicted through short-time discharge data of the storage battery, and the problem that traditional storage battery nuclear capacity discharge detection is time-consuming and labor-consuming is solved.

Description

Technical Field

[0001] The invention relates to the field of battery capacity detection, and in particular to a method and a device for predicting battery core capacity discharge. Background Art

[0002] As an energy storage device, batteries play an important role in power systems, renewable energy systems and other fields. The discharge capacity of a battery refers to the total amount of electrical energy that the battery can release under specific conditions, usually expressed in ampere hours (Ah) or watt hours (Wh). Discharge capacity is an important indicator for measuring battery performance, which directly affects the battery's continuous working time and power output capacity in practical applications. During the discharge process, the battery capacity gradually decreases as the current is consumed, and is affected by many factors, including the battery's chemical properties, temperature, discharge rate, and number of cycles. Understanding and predicting the discharge capacity of a battery is of great significance for optimizing the battery management system, the reliability of equipment operation, and improving user experience. Accurate discharge capacity prediction can help users better manage battery use, avoid over-discharge or over-charging, and thus extend the battery life.

[0003] The core capacity discharge test of the battery is an important means to evaluate the capacity, health and reliability of the battery, and is an important part of battery maintenance. Traditional discharge devices are often large and heavy, not only difficult to carry, but also restricted by the use environment, especially in some on-site maintenance or emergency situations, which brings inconvenience and causes great energy waste due to its own false load. In addition, the methods commonly used to obtain the discharge capacity of batteries in substations include static discharge test method, dynamic discharge test method and model estimation method. The static discharge test method is the most direct measurement method. After the battery is fully charged, it is continuously discharged at a specific discharge current, and the discharge time and voltage change are recorded until the battery voltage drops to the specified minimum value. According to the discharge current and discharge time, the discharge capacity can be calculated. This method is more accurate, but it takes a long time and requires the battery to be fully discharged, which may lead to battery loss and shortened life. In addition, frequent static discharge tests will increase the maintenance cost of the battery and affect its normal use. The dynamic test method discharges the battery under actual operating conditions, usually measured under different load conditions. This method can more truly reflect the performance of the battery in actual applications. However, due to the large changes in the test environment and load, the repeatability and accuracy of the test results may be affected. In addition, this method still requires a long test time, which may affect the continuous operation of the system. Summary of the invention

[0004] The technical problem to be solved by the present invention is to make up for the shortcomings of the above-mentioned battery core capacity discharge test method and provide a battery core capacity discharge prediction method to shorten the time required for battery core capacity discharge detection.

[0005] The present invention solves the above technical problems through the following technical means:

[0006] The present invention provides a method for predicting the discharge of a battery core capacity, comprising the following steps:

[0007] S1. Using the traditional static discharge test method, the actual discharge capacity time series data of the battery pack is obtained by sampling and the time series data is preprocessed;

[0008] S2. Constructing a long short-term memory network LSTM model based on hyperparameters, the model includes an input layer, an LSTM layer, a Relu activation layer, a Dropout layer, a fully connected layer and a regression layer, wherein the input data of the input layer is the discharge capacity, and the Dropout layer parameter is set to 0.2;

[0009] S3, using a novel particle swarm optimization PSO algorithm to explore the global optimal hyperparameters of the LSTM algorithm model described in step S2, and obtaining an optimized LSTM algorithm model;

[0010] S4. Based on the training set data and test set data obtained in step S1, network training and data prediction are performed on the optimized LSTM algorithm model obtained in step S3, and the prediction effect of the battery core capacity discharge prediction model is verified.

[0011] Furthermore, the data described in step S1 is preprocessed, specifically:

[0012] First, the first fifth of all the time series data is used as the training set data, and the remaining data is used as the test set data; secondly, the data normalization function mapminmax is used to normalize the training set and test set data to the interval [0,1]; finally, the data is analyzed, the total number of data samples is calculated, and the delay step and prediction span are set to 20 and 1 respectively.

[0013] Furthermore, step S3 includes the following steps:

[0014] S31, setting particle swarm optimization algorithm parameters;

[0015] S32. Define a function as a fitness function, which is used to train the LSTM model based on the hyperparameters using the training set data, and evaluate the model performance through the fitness value;

[0016] S33, using the optimized new particle swarm optimization algorithm, iterative calculation is carried out to obtain the global optimal hyperparameters;

[0017] S34. Use the optimal hyperparameters obtained in step S33 to update the training parameters of the long short-term memory network LSTM model based on hyperparameters described in step S2 to complete the optimization of the model.

[0018] Furthermore, the step S31 is specifically as follows:

[0019] Initialize the position and velocity information of the population, set the population size to 3, the maximum number of iterations to 8, and the optimization parameter dimension to 3, indicating that three hyperparameters need to be optimized; the hyperparameters are learning rate, batch size, and regularization coefficient, and their upper limits are 0.0051, 4097, and 0.051, and their lower limits are 0.0049, 4095, and 0.049, respectively;

[0020] Furthermore, the specific calculation formula of the fitness value described in step S32 is:

[0021]

[0022] Among them, n represents the number of samples, x i Represents the predicted value of the i-th sample, y i Represents the true value of the i-th sample. This function is used to measure the quality of the model hyperparameter combination. The maximum number of training times of the fitness function is set to 200, and the learning rate reduction factor parameter is set to 0.5.

[0023] Furthermore, the step S33 is specifically as follows:

[0024] The new particle swarm optimization PSO algorithm formula is used for calculation, and the iteration is carried out. The hyperparameter combination learning rate, batch size, and regularization coefficient obtained from each optimization are passed into the fitness function described in step S32, and the fitness value is calculated. This value is returned to the particle swarm optimization PSO algorithm as the basis for the next optimization; during the iterative calculation process, the particle swarm optimization PSO algorithm will dynamically adjust the hyperparameter combination that needs to be optimized to minimize the fitness value returned by the fitness function, and finally obtain the optimal fitness value that represents the minimum prediction error of the currently optimized LSTM model on the training set, the corresponding optimal combination of learning rate, batch size, and regularization coefficient, and the convergence speed curve of the optimal fitness value change process after each iteration.

[0025] Furthermore, the novel particle swarm optimization algorithm formula described in step S33 is specifically:

[0026]

[0027]

[0028] The value of the inertia weight is changed from a constant to an exponential function that changes with the number of iterations, and the individual and group optimal solutions are improved to a linear combination of the two; the particle velocity update term in the position update formula is discarded, and the concept of average dimensional information is introduced;

[0029] Among them, c1 and c2 are learning factors, r1 and r2 are random numbers in the range of (0,1), and pbest t i The best historical position found for the i-th particle is called the individual optimal solution, gbest t The historical optimal position searched for the entire particle swarm is called the group optimal solution, X i t is the position of the i-th particle in the t-th iteration; ω is the inertia weight, and its calculation formula is:

[0030] ω=ω max -(ω max -ω min )·e -0.1t

[0031] Among them, ω max and ω min They are the upper and lower limits of the inertia weight, which are set to 0.9 and 0.2 respectively.

[0032] Furthermore, the specific method of network training and data prediction described in step S4 is:

[0033] During the training process, the historical data of the first 20 time points of the training set are extracted as independent variables, their change characteristics are learned, and the data of the 21st time point is predicted and used as output; then the data of the 2nd to 21st time points are extracted as independent variables, and the data of the 22nd time point is predicted as output;

[0034] During the prediction process, the data of the last 20 time points in the training set are used to predict the data of the first time point in the validation set, and so on, to finally obtain all the prediction results.

[0035] The present invention also provides a device for predicting the discharge of a battery core capacity, comprising:

[0036] The data acquisition module is used to obtain the actual discharge capacity data of the battery using the traditional static discharge test method.

[0037] A main control MCU module, including a built-in memory and a chip, is used to execute the battery core capacity discharge prediction method according to claims 1-8, and predict the total discharge capacity of the battery according to the actual discharge capacity of the battery part collected by the data acquisition module;

[0038] The AD analog-to-digital conversion module is used to convert the collected battery pack discharge capacity data signal into a digital signal and input it into the main control MCU module.

[0039] The advantages of the present invention are:

[0040] The present invention makes the training of LSTM model more efficient and improves the prediction accuracy by utilizing the outstanding performance of the new particle swarm algorithm in hyperparameter adjustment and feature selection. Secondly, the present invention effectively copes with the timing and dynamic changes of data through the powerful memory capacity of LSTM, greatly shortening the time for obtaining discharge capacity data. Finally, the new particle swarm algorithm adaptively adjusts during the iteration process, avoiding the problem of local optimality and making the prediction results more stable and reliable. The application of this technology can not only improve the intelligence level of energy management system, but also has important significance for the utilization of renewable energy and the development of electric vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a method for predicting the discharge of a battery core capacity according to Embodiment 1 of the present invention;

[0042] Figure 2 This is a comparison chart of the discharge capacity prediction results based on the LSTM network and the measured results in Example 1 of the present invention;

[0043] Figure 3 This is a comparison chart of the discharge capacity prediction results based on the PSO-LSTM network and the measured results in Example 1 of the present invention;

[0044] Figure 4 This is a comparison chart between the discharge capacity prediction results based on the novel PSO-LSTM network and the actual measured results according to Example 1 of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Example 1

[0047] This embodiment provides a method for predicting the discharge capacity of a battery, which is based on an LSTM neural network model of a new particle swarm optimization (PSO) algorithm. In the LSTM neural network model, some parameters need to be set manually. These parameters directly affect the structure of the network model, and different parameter combinations will lead to significant differences in the model prediction performance. Therefore, the early network parameter settings are crucial. This embodiment uses a new particle swarm optimization (PSO) algorithm to optimize the key parameters of the LSTM model, thereby solving the problems of high computational overhead and long training time caused by large-scale parameter tuning, and helps find the optimal parameter combination of the LSTM neural network, further improving the prediction performance of the network model. The new PSO algorithm is used to determine the input parameters of the LSTM network, including the number of iterations, training batches, and learning rates, etc., in order to find suitable initialization network parameters for each mode, and adopt a direct multi-output lead-time multi-step prediction strategy to construct a multi-step LSTM network prediction model.

[0048] The present invention is based on the LSTM battery discharge capacity prediction method of the novel particle swarm optimization algorithm, such as Figure 1 As shown, the specific implementation steps are as follows:

[0049] S1. Use the traditional static discharge test method to obtain the actual discharge capacity time series data of the battery pack through sampling and preprocess the time series data. Obtain several discharge capacity data of the battery pack sampled according to the discharge time; this embodiment uses the actual nuclear capacity discharge data of the substation remote nuclear capacity system using the traditional static discharge test method, the discharge current is 20A, and the discharge type is 100% deep discharge. The sampling time is 11h, the sampling frequency is 1min, and the total discharge capacity data volume is 660. The 660 time series data are divided into training sets and test sets, where the training set data is the first 132, that is, the discharge capacity value of the first 2.2h, and the test set data is the last 528, that is, the discharge capacity value of the last 8.8h. The above 660 data use the data normalization function mapminmax to normalize the training set and test set data to the [0,1] interval.

[0050] S2. Build a long short-term memory network LSTM model based on hyperparameters, which includes an input layer, LSTM layer, Relu activation layer, Dropout layer, fully connected layer and a regression layer. The input data of the input layer is the discharge capacity, and the Dropout layer parameter is set to 0.2.

[0051] S3. Use a novel particle swarm optimization (PSO) algorithm to explore the global optimal hyperparameters of the LSTM algorithm model described in step S2 to obtain an optimized LSTM algorithm model.

[0052] In order to obtain a better LSTM network structure, it is necessary to find and determine the optimal hyperparameter combination of the network, including learning rate, batch size, and regularization coefficient. The optimal hyperparameters will be found through the new PSO optimization algorithm. The specific steps are as follows:

[0053] S31. Set the parameters of the particle swarm optimization algorithm, initialize the position and velocity information of the population, set the population size to 3, the maximum number of iterations to 8, and the optimization parameter dimension to 3, indicating that three hyperparameters need to be optimized. Set the upper limits of the learning rate, batch size, and regularization coefficient to 0.0051, 4097, and 0.051, respectively, and the lower limits to 0.0049, 4095, and 0.049, respectively.

[0054] S32. Define a function as a fitness function to train the LSTM model based on hyperparameters using the training set data, and evaluate the model performance through the fitness value. The fitness value calculation formula is:

[0055]

[0056] Among them, n represents the number of samples, x i Represents the predicted value of the i-th sample, y i Represents the true value of the i-th sample. This fitness value is used to measure the quality of the model hyperparameter combination. The relevant parameters inside the fitness function are set as follows: the maximum number of training times is set to 200, and the learning rate reduction factor is set to 0.5, that is, every 100 training times, the learning rate decays to 50% of the original.

[0057] S33, using the new particle swarm optimization PSO algorithm formula calculation, expansion iteration. The hyperparameter combination learning rate, batch size, regularization coefficient obtained by each optimization is passed into the fitness function described in step S32, and the fitness value is calculated according to the above formula. This value is returned to the particle swarm optimization PSO algorithm as the basis for the next step of optimization. During the iterative calculation process, the particle swarm optimization PSO algorithm will dynamically adjust the hyperparameter combination that needs to be optimized to minimize the fitness value returned by the fitness function, and finally obtain the optimal fitness value representing the minimum prediction error of the currently optimized LSTM model on the training set, the corresponding learning rate, batch size, the optimal combination of regularization coefficients, and the convergence speed curve of the optimal fitness value change process after each iteration.

[0058] The novel particle swarm optimization PSO algorithm formula is specifically as follows:

[0059]

[0060]

[0061] The value of the inertia weight is changed from a constant to an exponential function that changes with the number of iterations, and the individual and group optimal solutions are improved to a linear combination of the two; the particle velocity update term in the position update formula is discarded, and the concept of average dimensional information is introduced;

[0062] Among them, c1 and c2 are learning factors, r1 and r2 are random numbers in the range of (0,1), and pbest t i The best historical position found for the i-th particle is called the individual optimal solution, gbest t The historical optimal position searched for the entire particle swarm is called the group optimal solution, X i t is the position of the i-th particle in the t-th iteration; ω is the inertia weight, and its calculation formula is:

[0063] ω=ω max -(ω max -ω min )·e -0.1t

[0064] Among them, ω max and ω min They are the upper and lower limits of the inertia weight, which are set to 0.9 and 0.2 respectively.

[0065] S34. Use the optimal hyperparameters obtained in step S33 to update the training parameters of the long short-term memory network LSTM model based on hyperparameters described in step S2 to complete the optimization of the model.

[0066] S4. Based on the training set data and test set data obtained in step S1, the optimized LSTM algorithm model obtained in step S3 is subjected to network training and data prediction, and the prediction effect of the battery core capacity discharge prediction model is verified. The training set and test set data obtained in step S1 are used to train and simulate the model. The input training set data is used for network training, and the test set is predicted to obtain the predicted value. The specific methods of network training and data testing are as follows:

[0067] During the training process, the historical data of the first 20 time points of the training set are extracted as independent variables, their change characteristics are learned, and the data of the 21st time point is predicted and used as output; then the data of the 2nd to 21st time points are extracted as independent variables, and the data of the 22nd time point is predicted as output;

[0068] During the prediction process, the data of the last 20 time points in the training set are used to predict the data of the first time point in the validation set, and so on, to finally obtain all the prediction results.

[0069] Finally, the predicted values ​​of the training set are denormalized, mapped back to the original data range, and the visualization results are output. A comparison chart of the actual value and predicted value of the test set is drawn to intuitively evaluate the prediction effect.

[0070] This embodiment also provides a comparison of prediction errors between this method and the LSTM model algorithm based on particle swarm optimization and the ordinary LSTM model algorithm. The error evaluation uses the root mean square error RMSEP, the mean absolute percentage error MAPE and the prediction accuracy as evaluation indicators. Among them, the smaller the values ​​of RMSEP and MAPE, the higher the prediction accuracy; the prediction accuracy is calculated by MAPE, and the formula is as follows:

[0071] Prediction accuracy = (1-MAPE)*100%

[0072] The following table shows the prediction error indicators of different algorithms

[0073]

[0074] The above error indicators show that the LSTM model based on the novel particle swarm optimization PSO algorithm described in this embodiment has a high prediction accuracy and can be put into practical application. The obtained partial actual discharge capacity data samples can be input into the trained prediction model to more accurately predict the data of all discharges. According to the data of this embodiment, the discharge capacity data of the previous 2.2 hours is used to predict the discharge capacity data of the next 8.8 hours, and the discharge data of the battery group for all 11 hours is obtained. By using the method of this embodiment, the time of the existing static discharge test can be shortened by 80%.

[0075] The novel particle swarm algorithm used in the present invention performs well in hyperparameter adjustment and feature selection, making the training of the LSTM model more efficient, improving the prediction accuracy and significantly shortening the time for obtaining discharge capacity data. It can not only improve the intelligence level of the energy management system, but also has important significance for the utilization of renewable energy and the development of electric vehicles.

[0076] Example 2

[0077] It should be further explained that, based on the same inventive concept, the present invention also provides a device for predicting the discharge of a battery core capacity, comprising:

[0078] The data acquisition module is used to obtain the actual discharge capacity data of the battery using the traditional static discharge test method.

[0079] A main control MCU module, including a built-in memory and a chip, is used to execute the battery core capacity discharge prediction method described in Example 1, and predict the total discharge capacity of the battery according to the actual discharge capacity of the battery collected by the data collection module;

[0080] The AD analog-to-digital conversion module is used to convert the collected battery pack discharge capacity data signal into a digital signal and input it into the main control MCU module.

[0081] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the discharge of a battery, characterized in that: The following steps are involved: S1. Using the traditional static discharge test method, the actual discharge capacity time series data of the battery pack is obtained by sampling and the time series data is preprocessed; S2. Constructing a long short-term memory network LSTM model based on hyperparameters, the model includes an input layer, an LSTM layer, a Relu activation layer, a Dropout layer, a fully connected layer and a regression layer, wherein the input data of the input layer is the discharge capacity, and the Dropout layer parameter is set to 0.2; S3, using a novel particle swarm optimization PSO algorithm to explore the global optimal hyperparameters of the LSTM algorithm model described in step S2, and obtaining an optimized LSTM algorithm model; S4. Based on the training set data and test set data obtained in step S1, network training and data prediction are performed on the optimized LSTM algorithm model obtained in step S3, and the prediction effect of the battery core capacity discharge prediction model is verified.

2. A method for predicting battery core capacity discharge according to claim 1, characterized in that: The data described in step S1 is preprocessed, specifically: First, the first fifth of all the time series data is used as the training set data, and the remaining data is used as the test set data; secondly, the data normalization function mapminmax is used to normalize the training set and test set data to the interval [0,1]; finally, the data is analyzed, the total number of data samples is calculated, and the delay step and prediction span are set to 20 and 1 respectively.

3. A method for predicting battery core capacity discharge according to claim 1, characterized in that: The step S3 comprises the following steps: S31, setting particle swarm optimization algorithm parameters; S32. Define a function as a fitness function, which is used to train the LSTM model based on the hyperparameters using the training set data, and evaluate the model performance through the fitness value; S33, using the optimized new particle swarm optimization algorithm, iterative calculation is carried out to obtain the global optimal hyperparameters; S34. Use the optimal hyperparameters obtained in step S33 to update the training parameters of the long short-term memory network LSTM model based on hyperparameters described in step S2 to complete the optimization of the model.

4. A method for predicting battery core capacity discharge according to claim 1, characterized in that: The step S31 is specifically as follows: Initialize the position and velocity information of the population, set the population size to 3, the maximum number of iterations to 8, and the optimization parameter dimension to 3, indicating that three hyperparameters need to be optimized; the hyperparameters are learning rate, batch size, and regularization coefficient, and their upper limits are 0.0051, 4097, and 0.051, and their lower limits are 0.0049, 4095, and 0.049, respectively.

5. A method for predicting battery core capacity discharge according to claim 3, characterized in that: The specific calculation formula of the fitness value described in step S32 is: Among them, n represents the number of samples, x i Represents the predicted value of the i-th sample, y i Represents the true value of the i-th sample. This function is used to measure the quality of the model hyperparameter combination. The maximum number of training times of the fitness function is set to 200, and the learning rate reduction factor parameter is set to 0.

5.

6. A method for predicting battery core capacity discharge according to claim 3, characterized in that: The step S33 is specifically as follows: A novel particle swarm optimization (PSO) algorithm formula is used for calculation, and iteration is carried out. The hyperparameter combination learning rate, batch size, and regularization coefficient obtained from each optimization are passed into the fitness function described in step S32, and the fitness value is calculated. This value is returned to the particle swarm optimization (PSO) algorithm as the basis for the next step of optimization. During the iterative calculation process, the particle swarm optimization (PSO) algorithm will dynamically adjust the hyperparameter combination that needs to be optimized to minimize the fitness value returned by the fitness function, and finally obtain the optimal fitness value representing the minimum prediction error of the currently optimized LSTM model on the training set, the corresponding optimal combination of learning rate, batch size, and regularization coefficient, and the convergence speed curve of the optimal fitness value change process after each iteration.

7. A method for predicting battery core capacity discharge according to claim 6, characterized in that: The novel particle swarm optimization algorithm formula described in step S33 is specifically: The value of the inertia weight is changed from a constant to an exponential function that changes with the number of iterations, and the individual and group optimal solutions are improved to a linear combination of the two; the particle velocity update term in the position update formula is discarded, and the concept of average dimensional information is introduced; Among them, c1 and c2 are learning factors, r1 and r2 are random numbers in the range of (0,1), and pbest t i The best historical position found for the i-th particle is called the individual optimal solution, gbest t The historical optimal position searched for the entire particle swarm is called the group optimal solution, X i t is the position of the i-th particle in the t-th iteration; ω is the inertia weight, and its calculation formula is: oh = oh max -(oh max -oh min )·e -0.1t Among them, ω max and ω min They are the upper and lower limits of the inertia weight, which are set to 0.9 and 0.2 respectively.

8. A method for predicting battery core capacity discharge according to claim 1 or 2, characterized in that: The specific method of network training and data prediction described in step S4 is: During the training process, the historical data of the first 20 time points of the training set are extracted as independent variables to learn their changing characteristics. Predict the data at the 21st time point and use it as output; then extract the data from the 2nd to 21st time points as independent variables, and predict the data at the 22nd time point as output; During the prediction process, the data of the last 20 time points in the training set are used to predict the data of the first time point in the validation set, and so on, to finally obtain all the prediction results.

9. A device for predicting the discharge of a battery, characterized in that: include: A data acquisition module is used to obtain the actual discharge capacity data of the battery using a traditional static discharge test method; A main control MCU module, including a built-in memory and a chip, is used to execute the battery core capacity discharge prediction method according to claims 1-8, and predict the total discharge capacity of the battery according to the actual discharge capacity of the battery part collected by the data acquisition module; The AD analog-to-digital conversion module is used to convert the collected battery pack discharge capacity data signal into a digital signal and input it into the main control MCU module.

Images