Enhanced vehicle battery state of charge estimation method driven by model and data fusion

Abstract

The present invention relates to an enhanced vehicle battery state of charge estimation method driven by model and data fusion, and belongs to the field of battery technology. The method is composed of a state of charge SOC estimator based on a model and with real-time parameter updates and a SOC mapping network based on a machine learning algorithm in parallel, and can automatically match the best SOC estimation method according to the vehicle operating conditions. First, the vehicle operating condition is adaptively identified. If the vehicle is in a random dynamic condition, a SOC estimator based on a physical model and with real-time parameter updates is selected. The initial tuning parameters of the estimator are optimized by a reinforcement learning algorithm to achieve joint estimation of vehicle battery parameters and SOC; if the vehicle is in a continuous constant current condition, a mapping network based on a machine learning algorithm is selected to achieve accurate estimation of the vehicle battery SOC. Compared with a single battery SOC estimation method, the present invention has higher accuracy, robustness and generalization ability in practical applications.

Description

Technical Field

[0001] The present invention belongs to the technical field of batteries and relates to an enhanced vehicle-mounted battery state of charge estimation method driven by model and data fusion. Background Art

[0002] Lithium-ion batteries are widely used in the field of electrified transportation due to their excellent characteristics such as high power density, high energy density, low self-discharge rate, no memory effect and long service life. As of now, the number of new energy vehicles in China has reached 18.21 million, and lithium-ion batteries are mainly used as on-board energy sources. Therefore, accurate and robust on-board lithium-ion battery charge state estimation is crucial to achieve reliable and efficient operation of vehicles.

[0003] The state of charge of lithium-ion batteries is expressed as a percentage of the nominal capacity. It cannot be obtained by direct measurement and can only be estimated in a non-invasive way. Existing methods for estimating the state of charge of lithium-ion batteries can be divided into four categories: direct calculation method, model-based method, data-driven method and model-data fusion method. The two commonly used direct calculation methods are table lookup method and coulomb counting method. The direct calculation method is easy to implement, but its fidelity is limited due to the uncertainty and open-loop characteristics of the measurement of related parameters. The model-based method is generally implemented by combining electrochemical / equivalent circuit models and filtering algorithms. The electrochemical model uses partial differential equations to capture the aging process inside the battery from a microscopic perspective. The advantage of this method is that it can establish an accurate electrochemical model, but this method involves solving a large number of nonlinear partial differential equations, the model is complex and the calculation is large; the equivalent circuit model uses electronic components such as resistors and capacitors to simulate the external electrical characteristics of the battery. The advantage of this method is that the calculation cost is low, but due to the lack of sufficient physical and chemical aging mechanisms, its estimation accuracy is limited, and there is uncertainty in its modeling, and the estimation error is large. The data-driven method does not rely on the internal characteristics of the battery, but mainly relies on the battery charging and discharging data. Therefore, the effectiveness of this method depends largely on the quantity and quality of the battery data set. The model-data fusion method combines the advantages of the model-based method and the data-driven method, and has better fidelity and generalization ability. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an enhanced vehicle battery state of charge estimation method driven by model and data fusion. The method is composed of a state of charge SOC estimator based on a model and with real-time parameter updates and a SOC mapping network based on a machine learning algorithm in parallel, which can automatically match the best SOC estimation method according to the vehicle operating conditions. First, the vehicle operating condition is adaptively identified. If the vehicle is in a random dynamic condition, a SOC estimator based on a physical model and with real-time parameter updates is selected. The initial tuning parameters of the estimator are optimized by a reinforcement learning algorithm to achieve joint estimation of vehicle battery parameters and SOC; if the vehicle is in a continuous constant current condition, a mapping network based on a machine learning algorithm is selected to achieve accurate estimation of the vehicle battery SOC.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] An enhanced vehicle battery state of charge estimation method driven by model and data fusion, the method comprising the following steps:

[0007] S1: Select the lithium-ion battery equivalent circuit model, combine online parameter identification, ampere-hour integration and adaptive filtering algorithm to build a battery state of charge (SOC) estimation scheme based on the model and with real-time parameter updates, and optimize the initial tuning parameters of the model through reinforcement learning;

[0008] S2: Select appropriate machine learning algorithms and build a data-driven battery SOC estimation solution;

[0009] S3: Design an adaptive identification method for vehicle operating conditions to accurately identify the current operating conditions of the vehicle;

[0010] S4: According to the acquired vehicle operating conditions, select the corresponding battery SOC estimation method: if it is a random dynamic condition, adopt a battery SOC estimation scheme based on a model with real-time parameter updates; if it is a continuous constant current condition, adopt a battery SOC estimation scheme based on a machine learning algorithm.

[0011] Furthermore, in S1, a suitable equivalent circuit model is determined according to the positive electrode material of the lithium-ion battery: for lithium iron phosphate LiFePO4 batteries, a first-order RC equivalent circuit model with first-order hysteresis is selected; for ternary lithium NCM or NCA batteries, a first-order RC equivalent circuit model is selected, the online parameter identification algorithm is the recursive least squares algorithm RLS, and the adaptive filtering algorithm is the adaptive extended Kalman filter algorithm AEKF.

[0012] Furthermore, in S1, a deep Q network reinforcement learning optimization algorithm is used to solve the problem of mismatch between initial parameters and random operating conditions of the vehicle and uncertainty of initial SOC value of the lithium-ion battery.

[0013] Furthermore, in S2, the machine learning algorithm is a mapping network based on a gated recurrent unit (GRU).

[0014] Furthermore, in S3, the adaptive identification of the vehicle operating condition is achieved by collaboratively evaluating the current statistical index mean value MEAN and the overall standard deviation σ within the optimal sliding window.

[0015] Furthermore, the optimal sliding window is determined by selecting multiple values ​​in a certain interval and evaluating the estimation accuracy of the battery SOC under each selection;

[0016] If the current statistic σ in the sliding window is less than its set threshold, and MEAN is greater than its set threshold, the vehicle operating condition is determined to be a constant current condition;

[0017] If the current statistic σ in the sliding window is greater than its set threshold and MEAN is greater than its set threshold, the vehicle operating condition is determined to be a dynamic condition.

[0018] The beneficial effects of the present invention are:

[0019] (1) Taking full advantage of the advantages of each algorithm, two parallel vehicle battery SOC estimation schemes are adopted, and the corresponding vehicle battery SOC estimation method is jumped to according to the changes in the vehicle operating conditions. This method has stronger adaptability and can improve the accuracy, robustness and generalization ability of vehicle battery SOC estimation.

[0020] (2) The proposed deep Q-network reinforcement learning optimization algorithm can achieve accurate estimation of the vehicle battery SOC under random dynamic operating conditions by optimizing the initial tuning parameters of the model, and has high stability and convergence.

[0021] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0023] Figure 1 This is a flow chart of the vehicle battery state of charge estimation of the present invention;

[0024] Figure 2 is a first-order RC equivalent circuit model with first-order hysteresis in the present invention;

[0025] Figure 3is the first-order RC equivalent circuit model in the present invention;

[0026] Figure 4 It is a flow chart of the recursive least squares algorithm of the present invention;

[0027] Figure 5 This is a flow chart of the adaptive extended Kalman filter algorithm of the present invention;

[0028] Figure 6 This is a training flow chart of the deep Q network reinforcement learning of the present invention;

[0029] Figure 7 This is a structural diagram of the CRU model of the present invention. DETAILED DESCRIPTION

[0030] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0031] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0032] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0033] See also Figure 1 , an enhanced vehicle battery state of charge estimation method driven by model and data fusion, comprising the following steps:

[0034] S1: Select a suitable lithium-ion battery equivalent circuit model, combine online parameter identification, ampere-hour integration and adaptive filtering algorithm to build a battery state of charge (SOC) estimation scheme based on the model and with real-time parameter updates, and optimize the initial tuning parameters of the model through reinforcement learning;

[0035] S2: Select appropriate machine learning algorithms and build a data-driven battery SOC estimation solution;

[0036] S3: Design an adaptive identification method for vehicle operating conditions to accurately identify the current operating conditions of the vehicle;

[0037] S4: The enhanced vehicle battery state of charge estimation method driven by the model and data fusion according to claim 1 is characterized in that: in S1, a suitable equivalent circuit model is determined according to the positive electrode material of the lithium-ion battery: for lithium iron phosphate LiFePO4 battery, a first-order RC equivalent circuit model with first-order hysteresis is selected; for ternary lithium NCM or NCA battery, a first-order RC equivalent circuit model is selected, the online parameter identification algorithm is the recursive least squares algorithm RLS, and the adaptive filtering algorithm is the adaptive extended Kalman filter algorithm AEKF.

[0038] In step S1, the vehicle battery equivalent circuit model is determined according to the positive electrode material of the lithium-ion battery:

[0039] For lithium-ion batteries with lithium iron phosphate as the positive electrode material, a first-order RC equivalent circuit model with first-order hysteresis is selected, such as Figure 2 As shown, the control equation is:

[0040]

[0041] For lithium-ion batteries with ternary lithium NCM or NCA as the positive electrode material, select the first-order RC equivalent circuit model, such as Figure 3 As shown, the control equation is:

[0042]

[0043] Among them, SOC, η and C b are the battery state of charge, Coulomb efficiency and actual battery capacity, R p and C P are the polarization resistance and polarization capacitance of the RC circuit, U p is the RC circuit voltage, U oc is the battery open circuit voltage, h is the battery hysteresis voltage, U t is the battery terminal voltage, R sis the ohmic internal resistance, ε is the attenuation factor, k is a positive value used to adjust the attenuation rate of the hysteresis voltage, H is the maximum hysteresis voltage, I is the vehicle battery current, and the current direction is defined as positive for discharge and negative for charge.

[0044] The online parameter identification method is the recursive least squares algorithm RLS, which has the characteristics of strong adaptability and fast convergence speed. The principle of the recursive least squares algorithm is explained as follows:

[0045] According to the recursive least squares theory, the parameterized expression of the battery model is:

[0046]

[0047] Among them, y, θ and are the observation quantity, parameter vector and recursive quantity at time k respectively.

[0048] By Laplace transform, the RC model is discretized to obtain the expression:

[0049] U d (k+1)=b1(k)U d (k)-R s (k)I(k+1)+(b1(k)R s (k)-b2(k))I(k)

[0050] in:

[0051]

[0052] If the equivalent circuit model used is a first-order RC model with first-order hysteresis, then:

[0053] U d (k) = U t (k)-U oc (k)

[0054] If the equivalent circuit model used is a first-order RC model, then:

[0055] U d (k) = U t (k)-U oc (k)-h(k)

[0056] Where Δt is the sampling interval

[0057] like Figure 4 As shown in Figure 2, the specific implementation process of recursive least squares algorithm parameter identification is as follows:

[0058] (1) Initialize the parameter matrix, including the error covariance matrix, parameter vector, observation vector, optimization parameter vector, and recursive vector.

[0059] (2) Calculate the gain matrix, the calculation expression is:

[0060]

[0061] (3) Calculate the optimization parameter vector, the calculation expression is:

[0062]

[0063] (4) Calculate the system error covariance matrix, the calculation expression is:

[0064]

[0065] (5) According to the calculated optimized parameter vector at each moment, the R at the corresponding moment in the first-order RC model is calculated. s , R p and C p value, and complete the online parameter identification.

[0066] The ampere-hour integration method expression for open-loop battery SOC estimation is:

[0067]

[0068] Among them, SOC is the state of charge of the vehicle battery, η is the coulomb efficiency, which is generally taken as 0.98, and C b is the actual capacity of the vehicle battery, and I is the vehicle battery current.

[0069] The adaptive filtering algorithm is selected as the adaptive extended Kalman filter algorithm, and its principle is as follows:

[0070] The vehicle battery is a typical nonlinear system, and its state space equation is expressed as:

[0071]

[0072] Among them, x k and k represents the state vector and observation vector of the system at time k, f and h represent the state equation function and observation equation function of the system at time k, respectively. k and v k They represent the process noise and measurement noise of the system at time k respectively. k Represents the input of the system, that is, the current.

[0073] The battery system state vector is expressed as:

[0074] x k =[SOC k U p，k ] T

[0075] The state equation and measurement equation of the battery system are expressed as:

[0076]

[0077] State transition matrix A k , control input matrix B k And the observation matrix C k It is expressed as:

[0078]

[0079]

[0080]

[0081] like Figure 5 As shown, the specific implementation process of the adaptive extended Kalman filter algorithm is as follows:

[0082] (1) Initialize the battery system state vector x k , system process noise covariance P k , system measurement noise covariance Q k and the measurement error covariance R k ;

[0083] (2) The time update equation, i.e. the prior estimate, is calculated as follows:

[0084]

[0085] (3) The measurement update equation, i.e., the posterior estimate, is calculated as:

[0086]

[0087] Where K k+1 is the Kalman gain of the adaptive extended Kalman filter.

[0088] (4) Adaptive update of process noise and measurement noise, calculated as:

[0089]

[0090] Among them, e k+1 is the new information of adaptive extended Kalman filter, H k+1 is the estimated covariance of the new information in the time window, and N is the length of the sliding window.

[0091] Due to the characteristics of the adaptive extended Kalman filter algorithm itself (similar to the Markov decision process MDP), the process noise error covariance and the measurement noise error covariance can be updated in real time based on the adaptive update advantage of this filter algorithm. In this case, the future state of the system is only related to the current state of the system, but not to the past state of the system. Define the state of the system x∈χ, a∈A RL , the reward is r(x, a), according to the current strategy π(a k |x k ) Choose action a k And get the reward r(x k , a k ), then enter the next new state and get the reward r(x k+1 , a k+1 ).

[0092]

[0093] Wherein, N represents the length of the moving window and needs to be optimized.

[0094] When using a model-based approach to estimate the battery SOC, the mismatch between the model's initial parameters and the vehicle's random operating conditions and the uncertainty of the vehicle's battery SOC initial value will lead to inaccurate battery SOC estimation, so a reinforcement learning optimization algorithm is selected to obtain the model's initial parameters. Given that the deep Q-network reinforcement learning algorithm combines the two methods of deep neural network and Q-learning, it has superiority in achieving optimal action learning and decision-making in complex environments. Therefore, the system is trained offline in advance through deep Q-network reinforcement learning to solve the mismatch between the system's initial parameters and the vehicle's random operating conditions and the instability of the vehicle's battery SOC initial value, and to achieve adaptive parameter adjustment.

[0095] The training process of the deep Q network reinforcement learning optimization algorithm is as follows Figure 6 As shown in the figure, first initialize the experience pool and neural network parameters and reset the initial state of the environment observation; then select the action, observe the state and reward and store them in the experience pool, randomly select a batch of experience for neural network training, set the target network to update the policy network, and complete the training if it converges after training or reaches the maximum number of training times, otherwise reset the environment and continue to train the neural network until the model training is completed.

[0096] In step S2, the machine learning algorithm is selected as a mapping network based on a gated recurrent unit GRU. GRU can well solve the gradient explosion and gradient vanishing problems in weight back propagation, and the GRU model has the excellent characteristics of simple structure, fast training speed and low overfitting probability.

[0097] The structure of GRU is shown in the following figure. Figure 7As shown, the calculation expression of the model is:

[0098]

[0099] Among them, σ is the Sigmoid activation function, tanh is the Tanh activation function, z t is the update gate, r t To reset the gate, W r , W z , W h and W o Represent the weight matrices of reset gate, update gate, candidate hidden state and output point, respectively, b r , b z , b h and b o They represent the corresponding bias matrices, h t-1 ,h t They represent the candidate hidden state, the previous hidden state, and the current hidden state respectively. represents a linear activation function, I- is the forward propagation 1-z on the link t The update gate is used to control the extent to which the state information of the previous moment is brought into the current state; the reset gate is used to selectively discard irrelevant information from the input layer or hidden layer, x t is the input of the GRU model, including battery current, voltage and temperature, y t is the output of the GRU model, that is, the battery SOC at the current moment.

[0100] In step S3, the adaptive identification of the vehicle operating condition is achieved by collaboratively evaluating the average value MEAN of the current statistical index within the optimal sliding window and the overall standard deviation σ. The calculation expressions of MEAN and σ are:

[0101]

[0102]

[0103] Among them I k is the kth current value in the optimal sliding window, and N is the number of current values ​​in the optimal sliding window.

[0104] The optimal sliding window is determined by selecting multiple values ​​in a certain interval and evaluating the estimation accuracy of the battery SOC under each selection. If the current statistic σ in the sliding window is less than its set threshold and MEAN is greater than its set threshold, the vehicle operating condition is determined to be a constant current condition; if the current statistic σ in the sliding window is greater than its set threshold and MEAN is greater than its set threshold, the vehicle operating condition is determined to be a dynamic condition.

[0105] In step S4, a corresponding battery SOC estimation method is selected based on the acquired vehicle operating condition: if it is a random dynamic condition, a battery SOC estimation scheme based on a model and with real-time parameter updates is adopted; if it is a continuous constant current condition, a battery SOC estimation scheme based on a machine learning algorithm is adopted.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. An enhanced vehicle battery state of charge estimation method driven by model and data fusion, characterized by: The method comprises the following steps: S1: Select the lithium-ion battery equivalent circuit model, combine online parameter identification, ampere-hour integration and adaptive filtering algorithm to build a battery state of charge (SOC) estimation scheme based on the model and with real-time parameter updates, and optimize the initial tuning parameters of the model through reinforcement learning; S2: Select appropriate machine learning algorithms and build a data-driven battery SOC estimation solution; S3: Design an adaptive identification method for vehicle operating conditions to accurately identify the current operating conditions of the vehicle; S4: According to the acquired vehicle operating conditions, select the corresponding battery SOC estimation method: if it is a random dynamic condition, adopt a battery SOC estimation scheme based on a model and with real-time parameter updates; if it is a continuous constant current condition, adopt a battery SOC estimation scheme based on a machine learning algorithm; In the S1, a suitable equivalent circuit model is determined according to the positive electrode material of the lithium-ion battery: for lithium iron phosphate LiFePO4 batteries, a first-order RC equivalent circuit model with first-order hysteresis is selected; for ternary lithium NCM or NCA batteries, a first-order RC equivalent circuit model is selected, the online parameter identification algorithm is the recursive least squares algorithm RLS, and the adaptive filtering algorithm is the adaptive extended Kalman filter algorithm AEKF; In S1, a deep Q network reinforcement learning optimization algorithm is used to solve the problem of mismatch between initial parameters and random operating conditions of the vehicle and uncertainty of the initial SOC value of the lithium-ion battery; In S2, the machine learning algorithm is a mapping network based on a gated recurrent unit GRU; In S3, the adaptive identification of the vehicle operating condition is achieved by collaboratively evaluating the current statistical index mean value MEAN and the overall standard deviation σ within the optimal sliding window; The optimal sliding window is determined by selecting multiple values ​​in a certain interval and evaluating the estimation accuracy of the battery SOC under each selection; If the current statistic σ in the sliding window is less than its set threshold, and MEAN is greater than its set threshold, the vehicle operating condition is determined to be a constant current condition; If the current statistic σ in the sliding window is greater than its set threshold and MEAN is greater than its set threshold, the vehicle operating condition is determined to be a dynamic condition.

Images