Lithium ion battery remaining service life prediction method based on VMD-POA-LSTM

Abstract

The invention provides a lithium ion battery remaining service life prediction method based on VMD-POA-LSTM. According to the method, the collective predation behavior of pelicans is simulated to search an optimal solution. Comprising the following steps: S1, obtaining capacity data in a degradation process of the lithium ion battery; and S2, carrying out VMD decomposition on the data. S3, performing normalization processing on the data, and dividing a training set and a test set according to a proportion; and S4, optimizing the LSTM neural network by using POA, and optimizing hidden layer nodes, training times and learning rate. And S5, testing the optimized LSTM neural network, and if the prediction performance is not good, repeating the step S4. And S6, predicting the remaining service life of the lithium ion battery by using the optimized model. According to the method, the problem that the hyper-parameters of the LSTM neural network are difficult to determine can be improved, the calculation precision and speed are improved, the capacity regeneration problem is eliminated, errors are reduced fundamentally, and RUL prediction is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery remaining service life prediction, and specifically relates to a method for predicting the remaining service life of a lithium-ion battery based on VMD-POA-LSTM. Background Art

[0002] In order to achieve the goal of green and low-carbon global environmental protection, we pay more and more attention to the sustainability and environmental protection of lithium-ion batteries. As a secondary battery, lithium-ion batteries do not directly emit harmful chemicals during normal use, so they have been widely used in electric vehicles, aerospace, and energy storage equipment. However, as time goes by, lithium-ion batteries will undergo irreversible degradation due to calendar aging and environmental aging, resulting in a decline in battery performance and even more serious accidents. Therefore, it is particularly important to monitor the remaining useful life (RUL) of the battery in a timely manner, which is defined as the number of charge and discharge cycles remaining before the lithium-ion battery reaches the failure threshold. In practical applications, due to the complexity of the battery system and the limitations of the measurement technology, we cannot directly measure the remaining life of the lithium-ion battery, so we need to establish a prediction model to monitor the remaining life of the battery.

[0003] The existing RUL prediction methods are mainly divided into mechanism model-based methods and data-driven methods. The model-based method is to construct a mathematical model that reflects the battery aging state by studying the aging mechanism inside the battery, but the accuracy of the model is very dependent on its complex internal aging mechanism. The data-driven prediction method does not rely on the internal physical or chemical mechanism of the battery, but predicts the future degradation trend of the battery by analyzing the historical data of the battery during the degradation process, thereby realizing the RUL prediction of lithium-ion batteries. The model of the data-driven method often has multiple hyperparameters that can be adjusted, and the determination of the hyperparameters affects the accuracy of the model. Appropriate hyperparameter settings require many experiments to obtain better results, but this is time-consuming and labor-intensive, and increases the complexity of the model. In addition, the battery capacity will experience capacity hyperplasia during the degradation process, resulting in a decrease in prediction accuracy.

[0004] In order to solve the above technical problems and further improve the prediction level of battery remaining service life, a lithium-ion battery remaining service life prediction method based on VMD-POA-LSTM was proposed. Summary of the invention

[0005] In view of the shortcomings of the existing remaining service life prediction methods for lithium-ion batteries, the present invention proposes a remaining service life prediction method for lithium-ion batteries based on VMD-POA-LSTM, which solves the problem of capacity proliferation and the problem that hyperparameters are difficult to determine, and improves the accuracy and robustness of the prediction model.

[0006] The present invention is implemented through the following technical solutions:

[0007] The remaining service life prediction method of lithium-ion batteries based on VMD-POA-LSTM includes the following steps:

[0008] S1: Obtain capacity data of lithium-ion batteries during degradation.

[0009] S2: Perform VMD decomposition on the data.

[0010] S3: Normalize the data and divide the training set and test set into proportions.

[0011] S4: Use POA to optimize the LSTM neural network and optimize the hidden layer nodes, training times and learning rate.

[0012] S5: Test the optimized LSTM neural network. If the prediction performance is poor, repeat step S4.

[0013] S6: Use the optimized model to predict the remaining service life of lithium-ion batteries.

[0014] Furthermore, in step S2, the data is subjected to VMD decomposition, and the original capacity data is decomposed into a residual and several IMF components, thereby eliminating the capacity proliferation between the capacity data. The steps of the VMD decomposition algorithm are as follows:

[0015] VMD uses a completely non-recursive method for decomposition. Compared with traditional signal decomposition algorithms, the advantages of VMD lie in its non-recursive solution and autonomous selection of the number of modes.

[0016] VMD decomposes the original signal into several IMF components, which can be expressed as:

[0017]

[0018] Among them A k (t) is the signal, u k (t) is the envelope amplitude, is the instantaneous phase.

[0019] The VMD algorithm is divided into two parts: the construction and solution of the variational problem. Let the number of decomposition modes be K, and the variational problem is constructed as follows:

[0020]

[0021] where ω k is the center frequency corresponding to the kth eigenmode function, u kis the kth intrinsic mode function, δ(t) is the unit pulse, j is the imaginary unit, and f(t) is the original input signal.

[0022] For the above variational problem, a quadratic penalty factor is introduced and the Lagrangian penalty operator is used to solve it. The Lagrangian formula is constructed as follows:

[0023]

[0024] Where ɑ is the quadratic penalty factor and λ is the Lagrangian operator.

[0025] The above equation is solved by using the alternating direction method of multiplication operators, and the iterative update is performed and λ n+1 Find the "saddle point" of the above formula, the expression is:

[0026]

[0027]

[0028]

[0029] in For the remaining amount Wiener filter, τ is the noise tolerance.

[0030] Furthermore, the POA optimization algorithm steps in step S3 are as follows:

[0031] Step 1: Initialization preparation

[0032] The mathematical description of the initialization of the pelican population is:

[0033]

[0034] where x i,j is the j-th dimensional position of the i-th pelican; N is the population of pelicans; m is the dimension of the problem to be solved; rand is a random number in the range [0, 1]; u j and l j They are the upper and lower bounds of the j-th dimension of the problem to be solved.

[0035] Step 2: Approaching the prey (exploration phase)

[0036] The pelican determines the location of the prey and then moves toward the area of ​​the prey.

[0037] This behavior can be mathematically modeled as follows:

[0038]

[0039] in is the j-th dimensional position of the i-th pelican after the first stage update; rand is a random number in the range [0, 1]; I is a random integer of 1 or 2; p j is the j-th dimension position of the prey; F p is the objective function value of the prey.

[0040] Step 3, Surface Flight (Development Phase)

[0041] When the pelican reaches the water surface, it spreads its wings above the water, pushes the prey upward, and then hunts. This flight strategy enables the pelican to catch more fish within the attack range. Modeling this behavior of the pelican can make the POA converge to a better position in the hunting area, thereby enhancing its ability in local search and development, expressed as:

[0042]

[0043] in is the position of the jth dimension of the ith pelican after the second stage update; rand is a random number in the range [0, 1]; R is a random integer of 0 or 2; t is the current iteration number; T is the maximum iteration number; is the neighborhood radius of the population members.

[0044] Furthermore, the specific model of the traditional LSTM neural network in step S4 is:

[0045] The forget gate is used to choose whether to lose historical information. The network relationship is:

[0046] f t =σ(W f [h t-1 , x t ]+b f )

[0047] where f t is the state of the forget gate at time t; W f , b f are the weight and bias of the forget gate respectively; h t-1 is the information of the hidden layer at time t-1; x t is the input information at time t; σ is the activation function, which is the sigmoid function.

[0048] The input gate is used to selectively update the information state. The network relationship is as follows:

[0049] i t =σ(W i [h t-1 , x t ]+b i )

[0050] c′ t =tanh(W i [h t-1 , x t ]+b i )

[0051] c t =f t ·c t-1 +i t ·c′ t

[0052] where i t is the state of the input gate at time t; W i , b i are the weight and bias of the input gate respectively; c′ t is the candidate cell state; c t is the cell state at time t; c t-1 is the cell state information at time t-1.

[0053] The output gate determines whether to output information based on the current state. The network relationship is as follows:

[0054] o t =σ(W o ·[h t-1 , x t ]+b o )

[0055] h t =o t ·tanh(c t )

[0056] Among them t is the output gate state at time t; W o , b o are the weight and bias of the output gate respectively; h t is the information of the hidden layer at time t.

[0057] Further, in step S5, the optimized LSTM neural network is tested, and the test indicators are mean absolute error MAE, root mean square error RMSE, absolute error AE and relative error RE. The calculation formula is:

[0058]

[0059]

[0060] AE=|RUL t -RUL p |

[0061]

[0062] where y i is the predicted capacity value, is the actual capacity value, N is the number of cycles after the predicted starting point, RUL t is the actual RUL value, while RUL p is the predicted RUL value.

[0063] Beneficial Effects

[0064] (1) The present invention proposes a method for predicting the remaining useful life of lithium-ion batteries based on VMD-POA-LSTM. This method simulates the collective predation behavior of pelicans to search for the optimal solution. Compared with the traditional LSTM neural network model, the predation process is continuously iterated to find the optimal weights and thresholds, which shortens the training time and reduces the error.

[0065] (2) Before using POA optimization, preliminary training of the LSTM model can be performed to obtain the initial hyperparameter range and improve efficiency.

[0066] (3) This technical solution can verify the remaining service life of lithium-ion batteries and predict the end of battery life in advance, thus providing protection for equipment safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of the remaining service life prediction method of lithium-ion batteries based on VMD-POA-LSTM;

[0068] Figure 2 It is the result graph after VMD decomposition of capacity data;

[0069] Figure 3 It is the flow chart of POA optimization algorithm;

[0070] Figure 4 This is the LSTM neural network structure diagram;

[0071] Figure 5 This is the result diagram of the remaining service life prediction of lithium-ion batteries based on VMD-POA-LSTM;

[0072] Figure 6 It is a comparison with the prediction results of other different methods; DETAILED DESCRIPTION

[0073] The technical solution is described below in conjunction with the accompanying drawings to make the objects and advantages of the present invention clearer and easier to understand for those skilled in the art. However, the present invention is not limited to the scope of the specific implementation methods. For those of ordinary skill in the art, as long as the changes are obvious within the scope and spirit of the claims, they are all protected by the present invention.

[0074] like Figure 1 As shown, a method for predicting the remaining service life of a lithium-ion battery based on VMD-POA-LSTM is provided. The method specifically comprises the following steps:

[0075] S1: Obtain capacity data of lithium-ion batteries during degradation.

[0076] The battery dataset in this paper comes from NASA Prediction Center, and batteries labeled B0005 and B0006 are selected as research objects.

[0077] The rated capacity of the battery is 2A·h, and the charging and discharging process is carried out at room temperature: first, it is charged at a constant current of 1.5A until the cut-off voltage reaches 4.2V, and then it continues to charge in constant voltage mode until the charging current drops to 20mA; then it is discharged at a constant current of 2A until the capacity of B0005 and B0006 batteries drops to the failure threshold, respectively.

[0078] The failure threshold of battery capacity is defined as 70% of the rated capacity.

[0079] Collect and organize all capacity data.

[0080] S2: Perform VMD decomposition on the data.

[0081] Use decomposition methods to eliminate the problem of data capacity proliferation and reduce the errors caused by data, such as Figure 2 shown.

[0082] S3: Normalize the data and divide it into training set and test set in proportion, with the first 70% as the training set and the last 30% as the test set.

[0083] S4: Use POA to optimize the LSTM neural network and optimize the hidden layer nodes, training times and learning rate, such as Figure 3 shown.

[0084] Step 1: Initialization preparation, setting the initial population size and location of the pelican population.

[0085] In step 2, approaching the prey (exploration phase), the pelican locates the prey and then moves toward the prey's area.

[0086] Step 3, surface flight (development stage), when the pelican reaches the water surface, it spreads its wings above the water, pushes the prey upwards, and then hunts. This flight strategy enables the pelican to catch more fish within the attack range.

[0087] Furthermore, a traditional LSTM neural network model is established, such as Figure 4As shown, the data is input into the neural network for preliminary training. The specific model of the traditional LSTM neural network in step S4 is:

[0088] The forget gate is used to choose whether to lose historical information. The network relationship is:

[0089] f t =σ(W f [h t-1 , x t ]+b f )

[0090] where f t is the state of the forget gate at time t; W f , b f are the weight and bias of the forget gate respectively; h t-1 is the information of the hidden layer at time t-1; x t is the input information at time t; σ is the activation function, which is the sigmoid function.

[0091] The input gate is used to selectively update the information state. The network relationship is as follows:

[0092] i t =σ(W i [h t-1 , x t ]+b i )

[0093] c′ t =tanh(W i [h t-1 , x t ]+b i )

[0094] c t =f t ·c t-1 +i t ·c′ t

[0095] where i t is the state of the input gate at time t; W i , b i are the weight and bias of the input gate respectively; c′ t is the candidate cell state; c t is the cell state at time t; c t-1 is the cell state information at time t-1.

[0096] The output gate determines whether to output information based on the current state. The network relationship is as follows:

[0097] o t =σ(Wo ·[h t-1 , x t ]+b o )

[0098] h t =o t ·tanh(c t )

[0099] Among them t is the output gate state at time t; W o , b o are the weight and bias of the output gate respectively; h t is the information of the hidden layer at time t.

[0100] S5: Test the optimized LSTM neural network. If the prediction performance is poor, repeat step S4.

[0101] Furthermore, in step S5, the optimized LSTM neural network is tested, and the test indicators are mean absolute error MAE, root mean square error RMSE, absolute error AE and relative error RE. The calculation formula is:

[0102]

[0103]

[0104] AE=|RUL t -RUL p |

[0105]

[0106] where y i is the predicted capacity value, is the actual capacity value, N is the number of cycles after the predicted starting point, RUL t is the actual RUL value, while RUL p is the predicted RUL value.

[0107] S6: Use the optimized model to predict the remaining service life of lithium-ion batteries.

[0108] The prediction results are compared with those of several other different methods. The experimental results are shown in Table 1:

[0109] Table 1 Prediction results of different models

[0111] RUL in Table 1 t Indicates the true value of RUL, RUL pIt indicates the predicted value of RUL, and \ indicates that the prediction does not meet the degradation standard.

[0112] The experimental results show that the present invention has good prediction effects on different lithium-ion batteries, verifying the accuracy and robustness of the present invention.

[0113] In data processing, the present application uses the VMD method to decompose the capacity data, thereby reducing the impact of capacity proliferation and improving prediction accuracy.

[0114] In terms of algorithm model construction, the POA optimization algorithm is used to optimize the hyperparameters of the LSTM neural network, thereby improving the model's computing speed and accuracy.

Claims

1. A method for predicting the remaining service life of a lithium-ion battery based on VMD-POA-LSTM, characterized in that: The following steps are involved: S1: Obtain capacity data of lithium-ion batteries during degradation. S2: Perform VMD decomposition on the data. S3: Normalize the data and divide the training set and test set into proportions. S4: Use POA to optimize the LSTM neural network and optimize the hidden layer nodes, training times and learning rate. S5: Test the optimized LSTM neural network. If the prediction performance is poor, repeat step S4. S6: Use the optimized model to predict the remaining service life of lithium-ion batteries.

2. The method for predicting the remaining useful life of a lithium-ion battery based on VMD-POA-LSTM according to claim 1, characterized in that: Step S2 performs VMD decomposition on the data, decomposing the original capacity data into several IMF components, thereby eliminating the capacity proliferation problem between capacity data.

3. The method for predicting the remaining service life of a lithium-ion battery based on VMD-POA-LSTM according to claim 1, characterized in that: The steps of the Pelican Optimization Algorithm POA in step S3 are as follows: Step 1: Initialization preparation The mathematical description of the initialization of the pelican population is: x i,j =l j +rand(u j -l j ) i=1, 2, ..., N; j=1, 2, ..., m where x i，j is the j-th dimensional position of the i-th pelican; N is the population of pelicans; m is the dimension of the problem to be solved; rand is a random number in the range [0, 1]; u j and l j They are the upper and lower bounds of the j-th dimension of the problem to be solved. Step 2: Approaching the prey (exploration phase) The pelican determines the location of the prey and then moves towards the prey area. This behavior is mathematically modeled as: in is the j-th dimensional position of the i-th pelican after the first stage update; rand is a random number in the range [0, 1]; I is a random integer of 1 or 2; p j is the j-th dimension position of the prey; F p is the objective function value of the prey. Step 3, Surface Flight (Development Phase) When the pelican reaches the water surface, it spreads its wings on the water surface, pushes the prey upward, and then hunts. This flight strategy enables the pelican to catch more fish within the attack range. Modeling this behavior of the pelican can make the POA converge to a better position in the hunting area, thereby enhancing its ability in local search and development. The expression is: in is the position of the jth dimension of the ith pelican after the second stage update; rand is a random number in the range [0, 1]; R is a random integer of 0 or 2; t is the current iteration number; T is the maximum iteration number; is the neighborhood radius of the population members.

4. The method for predicting the remaining service life of a lithium-ion battery based on VMD-POA-LSTM according to claim 1, characterized in that: According to the actual aging of lithium-ion batteries, a failure threshold is set for the above steps, and a mathematical model between the predicted value and the failure threshold is established to obtain the life of the lithium-ion battery.

5. The method for predicting the remaining useful life of a lithium-ion battery based on VMD-POA-LSTM according to claim 1, characterized in that: Step S4 tests the optimized LSTM neural network. The test indicators are mean absolute error MAE, root mean square error RMSE, absolute error AE and relative error RE. The calculation formula is: Among them, y i is the predicted capacity value, is the actual capacity value, N is the number of cycles after the predicted starting point, RUL t is the actual RUL value, while RUL p is the predicted RUL value.

Images