SOC prediction method and system for lithium ion battery of electric forklift

Abstract

The invention discloses an SOC prediction method and system for a lithium ion battery of an electric forklift. The method comprises the steps that real-time data in the discharging process of the lithium ion battery are acquired; analyzing a real-time data change rule, and constructing a characteristic parameter model; a second-order RC equivalent circuit model is constructed, the model comprises a voltage source, an ohmic resistor and a second-order RC network, and the second-order RC network is used for describing the polarization effect of the charging and discharging stage; obtaining parameter identification results of the dynamic driving working conditions at different temperatures based on an ant colony algorithm; memory historical feature extraction of the time series data is realized based on a TCN model, including expansion convolution, causal convolution and residual connection; and selecting an average absolute percentage error and a mean square error as evaluation indexes. According to the method, parameter identification results of dynamic driving working conditions at different temperatures are obtained through the ant colony algorithm, the parameter identification results are input into the TCN model through corresponding transformation, and finally SOC estimation with high precision and robustness is achieved in LiFePO4.

Description

Technical Field

[0001] The present invention belongs to the field of lithium-ion batteries for electric forklifts, and in particular relates to a SOC prediction method and system for lithium-ion batteries for electric forklifts. Background Art

[0002] With the improvement of environmental awareness, there is an increasing concern about exhaust emissions and noise pollution generated by traditional internal combustion engine forklifts. Electric forklifts have received increasing attention as a zero-emission and low-noise alternative. As we all know, lithium-ion batteries are energy storage devices with high energy density and long life, and are now widely used in the field of electric forklifts. Estimating the state of charge (SOC) of lithium batteries plays an important role in battery management. Accurate SOC estimation can improve the performance, life and safety of the battery. However, lithium batteries, characterized by their nonlinear and time-varying electrochemical properties, pose considerable challenges in observing their internal state through external instruments. Therefore, SOC can only be estimated through a series of parameters or external characteristics.

[0003] At present, a large number of SOC estimation research works have been carried out at home and abroad. Generally speaking, SOC estimation methods can be divided into four categories: methods based on representation parameter mapping, ampere-hour integration methods, model-based estimation methods and data-driven methods.

[0004] The SOC estimation method based on characterization parameters is mainly divided into two steps. The first step is to establish an offline relationship between the power battery characterization parameters and the SOC, and then calculate the power battery characterization parameter values ​​in real time to calibrate the power battery SOC value. Optional characterization parameters include impedance spectrum and open circuit voltage (OCV), etc. The relatively stable OCV-SOC relationship is usually used to calibrate the SOC of industrial power batteries, which is simple and reliable. However, the biggest disadvantage of this method is that the battery needs to be left standing for more than 2 hours until the terminal voltage is completely depolarized before the SOC can be estimated.

[0005] The ampere-hour integration method is a typical open-loop estimation method that measures the SOC by integrating the discharge current that varies with time. This method has the advantage of being simple and easy to use, but it is highly dependent on the initial value of the SOC of the lithium-ion battery. At the same time, when the accuracy of the current sensor is not high, the measurement error will accumulate over time, causing the estimated SOC to gradually deviate from the actual value

[25] . In addition, the degradation of the power battery performance leads to a decrease in static capacity, which affects the calculation accuracy of the SOC.

[0006] The model-based approach includes four steps: model selection, battery testing, model parameter identification, and algorithmic estimation of SOC. The most commonly used models for SOC estimation of lithium-ion batteries are equivalent circuit models (ECM) or electrochemical models (EM). SOC estimation under electrochemical models is based on porous electrode theory, which relies on a system of partial differential equations (PDEs) to describe the microscopic reactions inside cells. EM may be accurate, but for real-time SOC estimation, it is accompanied by too many calculations, which will lead to long calculation times and high complexity. Therefore, it is crucial to reduce the number of parameters in EM.

[0007] SOC estimation based on data-driven methods does not require a comprehensive understanding of the internal characteristics of the battery, and greatly avoids wasting time and effort to build a complex battery model to display the battery status. This method is based on a large amount of offline data to establish and train a direct mapping relationship model between the power battery current, voltage, temperature and SOC. Popular machine learning methods include support vector machine (SVM), random forest (RF) and Gaussian process regression (GPR). However, the change of lithium battery SOC is a continuous process, because each SOC value is related to the previous information, and the operating state is relatively complex. The above methods of SOC estimation performance are not as good as deep learning methods. Deep learning algorithms describe the nonlinear relationship between input and output by adjusting the number of layers and the number of neurons. Popular deep learning methods include long short-term memory (LSTM), gated recurrent unit (GRU) and time convolutional network (TCN).

[0008] In view of this, in response to the technical problems existing in the above-mentioned field, this patented technology provides a SOC prediction technology for electric forklift lithium-ion batteries based on a second-order resistor-capacitor circuit-ACO-TCN model. This method obtains parameter identification results of dynamic driving conditions at different temperatures through the Ant Colony Optimization (ACO) algorithm, and inputs them into the TCN model through corresponding transformations, and finally achieves high-precision and robust SOC estimation in LiFePO4. In the present invention, the terminal voltage under the traditional data-driven model is replaced by OCV, so that TCN directly learns the OCV-SOC mapping relationship under the dynamic driving condition data set, so that the model can better adapt to the long-term dependencies in the time series data, and provide an efficient and reliable solution for the application field of SOC estimation. Summary of the invention

[0009] The purpose of the present invention is to overcome the above problems existing in the prior art and to provide a method and system for predicting the SOC of a lithium-ion battery of an electric forklift.

[0010] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0011] A method for predicting SOC of a lithium-ion battery of an electric forklift, comprising:

[0012] Get real-time data of lithium-ion battery discharge process;

[0013] Analyze the changing rules of real-time data and build characteristic parameter models;

[0014] Construct a second-order RC equivalent circuit model, which includes a voltage source, an ohmic resistor, and a second-order RC network, where the second-order RC network is used to describe the polarization effect during the charge and discharge stages;

[0015] The parameter identification results of dynamic driving conditions under different temperatures were obtained based on the ant colony algorithm;

[0016] Memory history feature extraction of time series data based on TCN model, including dilated convolution, causal convolution and residual connection;

[0017] Mean absolute percentage error and mean square error were chosen as evaluation metrics.

[0018] Furthermore, the real-time data includes voltage, current and temperature data during the discharge process of the lithium-ion battery, and a discharge experiment is carried out using a forklift battery cell.

[0019] Furthermore, the characteristic parameter model calculation formula is as follows:

[0020] Curve slope formula:

[0021]

[0022] Area under the curve formula:

[0023]

[0024] Standard deviation formula:

[0025]

[0026] Kurtosis calculation formula:

[0027]

[0028] Skewness calculation formula:

[0029]

[0030] Furthermore, the voltage source adopts a polynomial function in the following formula:

[0031] OCV SOC =P1*SOC 7 +P2*SOC 6 +……+P6*SOC2 +P7*SOC+P8

[0032] A second-order RC network consists of

[0033] Dynamic electrical behavior equation:

[0034]

[0035] U t =U oc -U p -i0R0

[0036] Discrete-time state-space equations:

[0037]

[0038] U t,k =U oc,k -U p,k -i 0,k R 0,k .

[0039] Furthermore, the ant colony algorithm includes:

[0040] m ants are randomly placed at m locations;

[0041] Each ant independently selects the next location based on the remaining pheromones and heuristic information on the path, that is, updates the ant’s position;

[0042] When the ant completes the search for a location, the elements of that location are combined and split;

[0043] Use the sum of squared errors as the evaluation function.

[0044] Furthermore, the probability of ant k moving from i to j is:

[0045]

[0046]

[0047] When all ants have completed their search, the pheromone will be updated according to the following formula:

[0048] τ ij (t+n)=(1-ρ)τ ij (t)+Δτ ij

[0049] In the above formula, ρ represents the evaporation coefficient of pheromone on the path, 1-ρ represents the persistence coefficient of pheromone, and Δτ ij represents the increment of pheromone on path i and j in this iteration;

[0050] The formula for the sum of squared errors is as follows:

[0051]

[0052] Furthermore, the calculation formula of the receptive field RF in the dilated convolution is

[0053] RF = n*(k-1)+1;

[0054] The calculation formula of the sensory field range y is:

[0055]

[0056] Furthermore, the causal convolution calculation formula is

[0057]

[0058] Furthermore, the residual connection output equation is:

[0059] y=h(f(x)+ConvlD(x)).

[0060] The present invention also provides a SOC prediction system for an electric forklift lithium-ion battery, comprising:

[0061] A data acquisition module, used to acquire real-time data during the discharge process of the lithium-ion battery;

[0062] Characteristic parameter analysis module, used to analyze the changing rules of real-time data and build characteristic parameter models;

[0063] A model building module is used to build a second-order RC equivalent circuit model, which includes a voltage source, an ohmic resistor, and a second-order RC network, where the second-order RC network is used to describe the polarization effect during the charge and discharge stages;

[0064] Ant colony algorithm module, used to obtain parameter identification results of dynamic driving conditions under different temperatures based on ant colony algorithm;

[0065] Convolution module, which is used to extract memory history features of time series data based on the TCN model, including dilated convolution, causal convolution, and residual connection;

[0066] The indicator selection module is used to select the mean absolute percentage error and the mean square error as the evaluation indicators.

[0067] The beneficial effects of the present invention are:

[0068] 1. The present invention obtains parameter identification results of dynamic driving conditions at different temperatures through ant colony algorithm, and inputs them into the TCN model through corresponding transformation, and finally realizes high-precision and robust SOC estimation in LiFePO4 battery.

[0069] 2. The terminal voltage in the traditional data-driven model is replaced by OCV, which enables TCN to directly learn the OCV-SOC mapping relationship under the dynamic driving condition dataset, so that the model can better adapt to the long-term dependencies in time series data and provide an efficient and reliable solution for the SOC estimation application field.

[0070] 3. The model is validated on a variety of dynamic driving condition datasets and compared with traditional data-driven methods. Experimental results show that the proposed method achieves satisfactory SOC estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0072] Figure 1 is a flow chart of the method of the present invention;

[0073] Figure 2 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0075] like Figure 1 The SOC prediction method of a lithium-ion battery of an electric forklift shown in the figure specifically comprises the following steps:

[0076] Step 1: Acquire real-time data during the discharge process of the lithium-ion battery; the real-time data includes voltage, current and temperature data during the discharge process of the lithium-ion battery, and conduct a discharge experiment using a forklift battery cell.

[0077] Step 2: Analyze the changing patterns of real-time data and build a characteristic parameter model.

[0078] The real-time current, voltage, capacity and other fields of the battery cell charging and discharging process are collected for data analysis. According to the real-time change law of charging current and voltage, the current value drops by ΔI in the constant current charging stage or the voltage value drops by ΔV in the constant voltage charging stage, which is the inflection point, and the maximum slope of the real-time data curve of current, voltage, capacity, etc., the area enclosed by the curve, standard deviation, kurtosis, skewness, maximum value, mean and other related features are calculated.

[0079] The calculation formula for constructing the characteristic parameter model is as follows:

[0080] Curve slope formula:

[0081]

[0082] Area under the curve formula:

[0083]

[0084] Standard deviation formula:

[0085]

[0086] Kurtosis calculation formula:

[0087]

[0088] Skewness calculation formula:

[0089]

[0090] Step 3: Construct a second-order RC equivalent circuit model, which includes a voltage source, an ohmic resistor, and a second-order RC network, where the second-order RC network is used to describe the polarization effect during the charge and discharge stages.

[0091] Usoc represents the open circuit voltage of the battery, which is considered to be a function of SOC and current temperature. It represents the OCV of the battery under different driving conditions, that is, OCV = (SOC, T). In order to achieve higher accuracy, the polynomial function in the following formula is used and parameterized to quantitatively describe the nonlinear OCV-SOC relationship:

[0092] OCV SOC =P1*SOC 7 +P2*SOC 6 +……+P6*SOC 2 +P7*SOC+P8

[0093] (b) Ohmic resistance: R0 represents the internal resistance of the electrode, electrolyte and gasket and the contact resistance of the component.

[0094] Second-order RC network: R i Represents polarization resistance, C i Represents polarization capacitance, which is used to describe the polarization effect during the charge and discharge phases.

[0095] Dynamic electrical behavior equation:

[0096]

[0097] U t =U oc -Up -i0R0

[0098] Discrete-time state-space equations:

[0099]

[0100] U t,k =U oc,k -U p,k -i 0,k R 0,k .

[0101] Where τ represents the time constant of the RC branch, τ = R p ×C p It is worth noting that the subscript k represents the time k, Δt is the sampling time, which is equal to 1s in the present invention, and U t and U p They respectively represent the terminal voltage and polarization voltage of capacitor C. In addition, i represents the load current.

[0102] Step 4: Obtain parameter identification results of dynamic driving conditions at different temperatures based on the ant colony algorithm.

[0103] Ant colony algorithm includes:

[0104] At the beginning of the algorithm, m ants are randomly placed at m locations. Each location consists of 5N+2 elements, including 3 connection weights between hidden layer and input layer neurons, N biases of hidden layer neurons, N connection weights between output layer and hidden layer neurons, 1 bias of output layer neurons, and the remaining element N, representing the number of hidden layer neurons, is randomly generated in the range of

[1100] . At this time, the pheromone on each path is equal, let τ ij (0) = c (c is a small constant).

[0105] Each ant independently selects the next positioning point based on the remaining pheromones and heuristic information on the path, that is, updates the position of the ant.

[0106] The probability that ant k moves from i to j is:

[0107]

[0108] In the above formula, Jk represents the point that Ant K is allowed to choose next time. Taboo table taboo k records the points that Ant K has passed. In the above formula, ηij is a heuristic factor that represents the expected level of an ant moving from position I to position J, while α and β represent the importance and expected heuristic factors of pheromone, respectively. When all ants have completed their search, the pheromone will be updated according to the following formula:

[0109] τ ij(t+n)=(1-ρ)τ ij (t)+Δτ ij

[0110] In the above formula, ρ(0<ρ<1) represents the evaporation coefficient of pheromone on the path, 1-ρ represents the persistence coefficient of pheromone, and Δτij represents the increment of pheromone on path ij in this iteration.

[0111]

[0112] where Q is a constant and Lk is the path length for this iteration.

[0113] When the ant completes its search for a location, the elements of that location are combined and split. 3N elements form the N×3 connection weight matrix w2. N elements consist of the N×1 bias matrix b1 of the hidden layer neurons. N elements consist of the 1×N connection weight matrix w3 between the output layer and the hidden layer neurons. One element consists of the 1×1 bias matrix b2 of the output layer neurons. The generated matrix and the input matrix U are calculated according to expressions (1), (2), and (3), and the corresponding result values ​​are output.

[0114] To evaluate the results, the sum of squared errors (SSE) is used as the evaluation function. The following is a specific expression of SSE.

[0115]

[0116] X i is the output, X is the reference, and is the actual value of SOC.

[0117] The smaller the SSE, the better the position the ant is searching. In other words, the search weights and number of neurons are optimal.

[0118] Step 5: Extract memory history features of time series data based on the TCN model, including dilated convolution, causal convolution, and residual connection.

[0119] The TCN model belongs to the convolutional network. Through improved causal convolution, dilated convolution and residual block technologies, it realizes the memory history feature extraction suitable for time series data. Its special network structure not only has the advantages of flexible perception field size, low memory consumption, parallel computing, but also avoids the problem of gradient vanishing or gradient exploding in RNN training.

[0120] The expansion convolution is as follows:

[0121] Since the length of the time series modeled by the traditional convolutional neural network is limited by the size of the convolution kernel and the depth of the network structure, if longer and more complex time dependencies are to be processed, more layers of linear stacking are required. In the n-layer structure of the standard convolutional network, the size of the convolution kernel is k, and the size of the calculated receptive field RF is expressed as follows:

[0122] RF=n*(k-1)+1

[0123] By introducing the extended convolution structure, the relationship between the sensory field range of the neural network and the number of hidden layers changes from linear to exponential. Therefore, the model can obtain a larger sensory field range while maintaining a shallow network depth, which makes it easier to optimize and converge. At the same time, the formula for the sensory field range y becomes:

[0124]

[0125] Where n represents the number of hidden layers, k represents the size of the convolution kernel, and dn represents the expansion factor of the nth hidden layer, which is calculated as dn=2n-1.

[0126] The causal convolution is as follows:

[0127] Causal convolution means that the output of each time step depends on the previous time step. For the value of the previous layer at time t, it only depends on the value of the next layer at time t and before. Unlike traditional convolutional neural networks, causal convolution cannot see future data. It is a unidirectional structure instead of a bidirectional structure, and is a model strictly limited by time. In other words, in the time series X = (x1, x2, ..., x T ) and filter F = (f1, f2, ..., f K ), the causal convolution t at x can be expressed as:

[0128]

[0129] The residual connection is as follows:

[0130] Residual connection is a technique for transmitting information by adding cross-layer connections to the network model, where the input signal can be directly passed to the output of the subsequent layer through the cross-layer connection. This residual connection mechanism enables the network to better capture the difference between input and output, which helps to speed up the network training process and avoid problems such as gradient explosion or gradient disappearance, even if the network is too deep. By introducing residual connectivity, TCN can better convey and utilize historical information and improve the ability to model time series data. The output equation of the residual block can be expressed as follows:

[0131] y=h(f(x)+ConvlD(x))

[0132] Where y is the output of the residual block, x is the input of the residual block, f(·) is a series of transformation operations represented by the left branch, ConvlD(·) is a one-dimensional convolution operation, and h(·) represents the activation function ReLU.

[0133] Step 6: Select mean absolute percentage error (MAE) and mean square error (RMSE) as evaluation indicators.

[0134] The present invention also provides a SOC prediction system for an electric forklift lithium-ion battery, comprising:

[0135] A data acquisition module, used to acquire real-time data during the discharge process of the lithium-ion battery;

[0136] Characteristic parameter analysis module, used to analyze the changing rules of real-time data and build characteristic parameter models;

[0137] A model building module is used to build a second-order RC equivalent circuit model, which includes a voltage source, an ohmic resistor, and a second-order RC network, where the second-order RC network is used to describe the polarization effect during the charge and discharge stages;

[0138] Ant colony algorithm module, used to obtain parameter identification results of dynamic driving conditions under different temperatures based on ant colony algorithm;

[0139] Convolution module, which is used to extract memory history features of time series data based on the TCN model, including dilated convolution, causal convolution, and residual connection;

[0140] The indicator selection module is used to select the mean absolute percentage error and the mean square error as the evaluation indicators.

[0141] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0142] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for predicting the SOC of a lithium-ion battery of an electric forklift, characterized in that: include: Get real-time data of lithium-ion battery discharge process; Analyze the changing rules of real-time data and build characteristic parameter models; Construct a second-order RC equivalent circuit model, which includes a voltage source, an ohmic resistor, and a second-order RC network, where the second-order RC network is used to describe the polarization effect during the charge and discharge stages; The parameter identification results of dynamic driving conditions under different temperatures were obtained based on the ant colony algorithm; Memory history feature extraction of time series data based on TCN model, including dilated convolution, causal convolution and residual connection; Mean absolute percentage error and mean square error were chosen as evaluation metrics.

2. The SOC prediction method of a lithium-ion battery for an electric forklift according to claim 1, characterized in that: Real-time data includes voltage, current and temperature data during the discharge process of lithium-ion batteries, and discharge experiments are carried out using forklift battery cells.

3. The SOC prediction method of a lithium-ion battery for an electric forklift according to claim 1, characterized in that: The calculation formula for constructing the characteristic parameter model is as follows: Curve slope formula: Area under the curve formula: Standard deviation formula: Kurtosis calculation formula: Skewness calculation formula:

4. The SOC prediction method of a lithium-ion battery for an electric forklift according to claim 1, characterized in that: The voltage source uses the polynomial function given by: OCV SOC =P1*SOC 7 +P2*SOC 6 +……+P6*SOC 2 +P7*SOC+P8 A second-order RC network consists of Dynamic electrical behavior equation: U t =U oc -U p -i0R0 Discrete time state-space equations: IN t,k =U oc,k -IN p,k -and 0,k R 0,k 。 5. The SOC prediction method of a lithium-ion battery for an electric forklift according to claim 1, characterized in that: Ant colony algorithm includes: m ants are randomly placed at m locations; Each ant independently selects the next location based on the remaining pheromones and heuristic information on the path, that is, updates the ant’s position; When the ant completes the search for a location, the elements of that location are combined and split; Use the sum of squared errors as the evaluation function.

6. The SOC prediction method of a lithium-ion battery for an electric forklift according to claim 5, characterized in that: The probability that ant k moves from i to j is: When all ants have completed their search, the pheromone will be updated according to the following formula: t ij (t+n)=(1-ρ)τ ij (t)+Δτ ij In the above formula, ρ represents the evaporation coefficient of pheromone on the path, 1-ρ represents the persistence coefficient of pheromone, and Δτ ij represents the increment of pheromone on path i and j in this iteration; The formula for the sum of squared errors is as follows:

7. The SOC prediction method of a lithium-ion battery for an electric forklift according to claim 1, characterized in that: The calculation formula of receptive field RF in dilated convolution is RF = n*(k-1)+1; The calculation formula of the sensory field range y is:

8. The SOC prediction method for an electric forklift lithium-ion battery according to claim 7, characterized in that: The calculation formula of causal convolution is:

9. The SOC prediction method of a lithium-ion battery for an electric forklift according to claim 8, characterized in that: The residual connection output equation is: y=h(f(x)+ConvlD(x)).

10. A SOC prediction system for a lithium-ion battery of an electric forklift, characterized in that: include: A data acquisition module, used to acquire real-time data during the discharge process of the lithium-ion battery; Characteristic parameter analysis module, used to analyze the changing rules of real-time data and build characteristic parameter models; A model building module is used to build a second-order RC equivalent circuit model, which includes a voltage source, an ohmic resistor, and a second-order RC network, where the second-order RC network is used to describe the polarization effect during the charge and discharge stages; Ant colony algorithm module, used to obtain parameter identification results of dynamic driving conditions under different temperatures based on ant colony algorithm; Convolution module, which is used to extract memory history features of time series data based on the TCN model, including dilated convolution, causal convolution, and residual connection; The indicator selection module is used to select the mean absolute percentage error and the mean square error as the evaluation indicators.

Images