Power battery health state management method and system of vehicle battery swap station

Abstract

The invention relates to the technical field of battery health management, in particular to a power battery health state management method and system of a vehicle battery swap station. The method comprises the following steps: acquiring charging data of a power battery; the charging data is used for processing to obtain feature data; in the early stage and the middle stage of the life cycle of the power battery, constructing a first feature data set and a second feature data set by using the feature data of each charging cycle and the maximum capacity label of the battery; taking the first feature data set and the second feature data set as a training set and a verification set of an SOH prediction framework respectively; at the late period of the life cycle of the power battery, inputting the characteristic data into an SOH prediction framework to estimate the maximum capacity of the battery after each charging cycle and determine an SOH value; and determining the charging cut-off current of the power battery according to the SOH value and whether the capacity diving phenomenon occurs in the battery. According to the method, the SOH value of the battery can be accurately predicted only by using the charging data, and fine management of the power battery by the battery swap station can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health management, and in particular to a method and system for managing the health status of a power battery in a vehicle battery swap station. Background Art

[0002] As the world strives to achieve carbon neutrality, the electrification of automobiles is accelerating. Governments in many countries and regions are actively promoting the transformation of public transportation to low-carbon development, and buses, taxis, and online ride-hailing vehicles are being converted to pure electric vehicles on a large scale. However, the long charging time and low energy efficiency of operating vehicles still restrict the promotion and application of pure electric vehicles.

[0003] In order to solve the problem of slow charging, power batteries that adapt to high charging rates have come into being, compressing the energy replenishment time to less than 20 minutes. However, in cold areas, the application of fast charging technology is greatly limited. Studies have shown that battery aging is closely related to the use environment. Especially in cold areas, low temperatures aggravate the internal aging side reactions of the battery, resulting in a capacity decay rate that is significantly higher than in warm areas. In addition, the diffusion capacity of internal battery materials decreases at low temperatures. During fast charging, the battery terminal voltage rises rapidly to the charging cut-off voltage, which also increases the risk of metallic lithium deposition. These factors seriously restrict the application of fast charging technology in cold areas.

[0004] However, the battery replacement technology of electric vehicles involves battery rotation and maintenance. Each battery pack has sufficient time for charging and maintenance, so the battery replacement mode has received special attention in cold regions. For shared power batteries, their safety and reliability are particularly important. If a battery fire accident occurs, the damage caused is often more serious. However, in the application scenarios of on-board power batteries, lithium-ion batteries will not consume all the power, making it difficult to obtain the battery's health state (SOH) through full charging and full discharging. And for power batteries that use battery replacement for energy replenishment, battery data is scattered between the vehicle end and the battery replacement station. A battery will rotate between different vehicles and different battery replacement stations, and its historical usage data is fragmented, making it more difficult to estimate the power battery SOH. Inaccurate prediction of the power battery SOH will make it difficult for the battery replacement station to take reasonable management and maintenance of the power battery according to the degree of battery aging, which may not only accelerate the aging of the battery, but also increase the safety risk of the power battery. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for managing the health status of power batteries in a vehicle battery swap station to solve the problems raised in the above-mentioned background technology.

[0006] In order to solve the above problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for managing the health status of a power battery in a vehicle battery swap station, comprising: Acquire charging data of the power battery in the battery swap station; the charging data is used to process and obtain characteristic data of each charging cycle, the characteristic data including battery incremental capacity characteristics and constant voltage charging duration characteristics; In the early stage of the life cycle of the power battery, a first feature data set is constructed using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label; in the middle stage of the life cycle of the power battery, a second feature data set is constructed using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label; Using the first feature data set as a training set of the SOH prediction framework, and using the second feature data set as a validation set of the SOH prediction framework; In the late stage of the life cycle of the power battery, the characteristic data of each charging cycle is input into the trained and verified SOH prediction framework to estimate the maximum capacity of the battery after each charging cycle; The SOH value is determined according to the estimated ratio of the maximum capacity of the battery to the initial maximum capacity; and the charging cut-off current of the power battery is determined according to the SOH value and whether the battery has a capacity drop phenomenon.

[0007] Preferably, the charging data includes the number of charging cycles, the charging time of each charging cycle, the charging current, the charging voltage, and the charging power.

[0008] Preferably, the battery incremental capacity characteristics include an incremental capacity peak position and an incremental capacity peak value of an incremental capacity curve; The calculation formula of the incremental capacity curve is:

[0009] Where IC represents the incremental capacity, Q represents the charge capacity, V represents the charge voltage, I represents the charge current, and t represents the sampling time.

[0010] Preferably, the constant voltage charging time characteristic includes constant voltage charging time, and the calculation formula of the constant voltage charging time is:

[0011] In the formula, CVT represents the constant voltage charging duration, CVT1 represents the time point when constant current charging is converted to constant voltage charging, and CVT2 represents the time point when constant voltage charging ends.

[0012] Preferably, the incremental capacity curve is filtered using a Savitzky-Golay filtering method.

[0013] Preferably, the SOH prediction framework includes a BiLSTM neural network and a Transformer framework connected in series.

[0014] Preferably, determining the charging cut-off current of the power battery according to the SOH value and whether the battery capacity drops comprises: If the battery capacity drops, the charge cut-off current remains at 0.5C; If the battery capacity does not drop and SOH>95%, the charge cut-off current remains at 0.05C; If the battery capacity does not drop and 95%>SOH>90%, the charge cut-off current remains at 0.1C; If the battery capacity does not drop and 90%>SOH>85%, the charge cut-off current remains at 0.2C; If the battery capacity does not drop and 85%>SOH>80%, the charge cut-off current remains at 0.3C; If the battery capacity does not drop and 80%>SOH, the charge cut-off current remains at 0.5C.

[0015] In a second aspect, the present invention provides a power battery health status management system for a vehicle battery swap station, comprising: An acquisition module is used to acquire charging data of the power battery in the battery swap station; the charging data is used to process and obtain characteristic data of each charging cycle, the characteristic data including battery incremental capacity characteristics and constant voltage charging duration characteristics; A feature data set construction module is used to construct a first feature data set using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label in the early stage of the power battery life cycle; and to construct a second feature data set using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label in the middle stage of the power battery life cycle; A model training and verification module, used to use the first feature data set as a training set of the SOH prediction framework, and use the second feature data set as a verification set of the SOH prediction framework; An estimation module, used for inputting the characteristic data of each charging cycle into the trained and verified SOH prediction framework in the late stage of the power battery life cycle to estimate the maximum capacity of the battery after each charging cycle respectively; The management module is used to determine the SOH value according to the estimated ratio of the maximum capacity of the battery to the initial maximum capacity; and determine the charging cut-off current of the power battery according to the SOH value and whether the battery has a capacity drop phenomenon.

[0016] Compared with the prior art, the present invention has the following beneficial effects: Under the premise of using only charging data, the present invention constructs capacity increment features and constant voltage charging time features as feature data to characterize the degree of aging of the power battery, and uses the first feature data set and the second feature data set obtained in the early and middle stages of the power battery life cycle as training sets and validation sets, respectively, so that the SOH prediction framework can capture the relationship between the feature data (IC features and CVT features) of the power battery and the maximum capacity of the battery or the degree of battery aging. Therefore, in the late stage of the power battery life cycle, the SOH prediction framework can accurately predict the maximum capacity of the power battery after each charging cycle, and based on the predicted maximum capacity of the battery, the power battery can be further managed in a targeted and refined manner to improve the battery life. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of a method for managing the health status of a power battery in a vehicle battery swap station in an embodiment; Figure 2 1 is a graph of IC curves before and after filtering in the embodiment; Figure 3 is a graph of multiple IC curves corresponding to different numbers of charging cycles (or cycle numbers) in the embodiment; Figure 4 is a curve diagram of a voltage curve when the power battery in the embodiment is fast charged at a 3C rate current (15A); Figure 5 is a graph of voltage curves of the power battery in the embodiment when it is fast charged at a 5C rate (25A) in different charging cycles; Figure 6 is an analysis diagram of the embodiment using the Pearson correlation coefficient to quantify the battery capacity, CVT characteristics, IC peak position and the correlation coefficient between the IC peak values; Figure 7 is a graph of the capacity attenuation curve of the power battery sample predicted by the SOH prediction framework in the embodiment; Figure 8 This is an error comparison chart of SOH prediction using four models: SVR, LSTM, Transformer, and BiLSTM-Transformer in the embodiment. DETAILED DESCRIPTION

[0018] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are 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 operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more. In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0019] Among the prior art known to the inventors of the present application, the prior art usually relies on battery bench testing and structural disassembly to estimate the health status of power batteries, which is complicated and costly. During the aging process of power batteries (usually lithium-ion batteries), the degree of battery capacity decay is easier to predict in the early stage of the life cycle of the power battery due to high battery consistency, high capacity retention and relatively linear capacity decay. However, in the late stage of the life cycle of the power battery, the capacity decay curve will show certain nonlinear characteristics. If the algorithm in the prior art that only uses battery capacity and battery service time data for training is used, the accuracy of the final prediction of the battery capacity decay curve will be insufficient.

[0020] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0021] Combination Figure 1 This embodiment provides a method for managing the health status of a power battery in a vehicle battery swap station, which includes: Step S1: Acquire charging data of the power battery in the battery swap station, and the charging data is used to process and obtain characteristic data of each charging cycle, wherein the characteristic data includes battery incremental capacity characteristics and constant voltage charging duration characteristics.

[0022] Among them, the charging data of the power battery in the battery swap station is provided by the data center of the battery swap station. The acquired charging data includes the number of charging cycles of the power battery in the battery swap station (used to characterize the timing sequence of each charging cycle) and the charging time, charging current, charging voltage and charging power of each charging cycle or each charging. This embodiment uses the battery health status of the power battery to characterize the attenuation degree of the battery capacity. The battery health status, i.e., SOH, is calculated by the ratio of the current maximum battery capacity of the battery to the initial maximum capacity. This is common knowledge in the art and will not be elaborated here.

[0023] In order to more accurately predict the health status of the power battery, this embodiment uses charging data to construct two artificial features related to the internal aging mechanism of the battery, namely the battery incremental capacity feature (IC) and the constant voltage charging time feature (CVT). The battery incremental capacity feature (synonymous with IC feature in the following text) and the constant voltage charging time feature (synonymous with CVT feature in the following text) are used to input the pre-built SOH prediction framework.

[0024] The battery incremental capacity characteristics include the incremental capacity peak position (hereinafter referred to as IC peak position) and the incremental capacity peak value (hereinafter referred to as IC peak value) of the incremental capacity curve (hereinafter referred to as IC curve). The calculation formula of the incremental capacity curve is:

[0025] Where IC represents the incremental capacity, Q represents the charge capacity, V represents the charge voltage, I represents the charge current, and t represents the sampling time.

[0026] In a specific embodiment, due to the presence of noise in the sensor in actual applications, and the differential method will amplify the noise, it is difficult to accurately obtain the incremental capacity peak value of the incremental capacity curve. In order to solve this problem, this embodiment uses the Savitzky-Golay (SG) filtering method to filter the IC curve, and its calculation formula is as follows:

[0027] in, is the smoothed signal value; k is the index of the filter coefficient, is the point of the original signal in the window; Is half the width of the window. is the coefficient related to the window weight, this series of weights It is obtained by fitting a polynomial using the least squares method in a sliding window. The IC curves before and after filtering are shown in the attached figure. Figure 2 .

[0028] To further illustrate, in the actual use of the power battery, the charging and discharging of the battery usually constitutes a cycle. However, since the battery swap station cannot obtain the discharge data of the power battery, this embodiment does not use the discharge data but only uses the charging data that the battery swap station can obtain to process the IC curve, such as Figure 2 As shown in the figure, the horizontal axis of the IC curve is the battery voltage. In addition, since the battery discharge current will change with the vehicle driving conditions during use, it is not conducive to using the differential calculation method to process IC features. In comparison, the stable charging condition of constant current-constant voltage charging used by power batteries in battery swap stations is more suitable for processing IC features.

[0029] like Figure 2 As shown in the figure, the IC curves before and after SG filtering show that the filtered IC curve is smoother, which is more conducive to extracting the battery incremental capacity characteristics that can characterize the degree of aging of the power battery. The IC curves processed by the power battery in each charging cycle in the battery swap station are different. As the number of charging times increases, the degree of aging of the power battery gradually deepens, and the IC curve also changes accordingly. The multiple IC curves corresponding to different charging cycles (or cycles) are shown in Figure 3 As shown. Combined Figure 3 , the IC curves obtained from the charging data of the power battery in different charging cycles have similar shapes, but the peaks in the IC curves have unique heights and positions, which reflect the changes in the electrode materials during the battery charging and discharging process. As the number of cycles (or charging times) of the power battery continues to increase, the peak of the IC curve decreases, and the peak position also moves toward a lower voltage. These changes are associated with the decline in the health of the power battery.

[0030] However, due to the high uncertainty of the remaining power (SOC) of the power batteries entering the battery swap station for charging, some power batteries may still have a high SOC. Therefore, the voltage range covered by the charging data of the power batteries in the battery swap station may be incomplete, and the charging data in the low-voltage area corresponding to the low SOC is often missing, which may make it impossible to extract suitable IC features. It is worth noting that, unlike the power batteries replenished by charging piles, there is battery rotation work in the battery swap station, and each power battery basically has sufficient time for maintenance without affecting the efficiency of vehicle replenishment. Therefore, the power battery can be fully charged in the battery swap station and have a relatively complete constant voltage charging process, which is difficult to achieve with charging piles.

[0031] Combination Figure 4The power battery in the figure is fast charged at a 3C rate current (15A). The battery switches from constant current charging to constant voltage charging in 890 seconds, and the battery SOC reaches 72%. It can be found that when constant voltage charging occurs, the battery power is high and very close to the end of charging, while the fast charging of the charging pile generally ends at 80% SOC, so the constant voltage charging time characteristic (CVT characteristic) is more suitable for the power battery of the battery swap station.

[0032] Specifically, the constant voltage charging time feature includes a constant voltage charging time, and the constant voltage charging time is determined according to a change curve of the charging voltage that changes with the charging time. The calculation formula of the constant voltage charging time is:

[0033] In the formula, CVT represents the constant voltage charging duration, CVT1 represents the time point when constant current charging is converted to constant voltage charging, and CVT2 represents the time point when constant voltage charging ends.

[0034] Combination Figure 5 , Figure 5 is the voltage curve of the power battery when it is fast charged at 5C rate (25A), Figure 5 It can be confirmed that even if a larger current rate (currently the maximum in the industry is 5C) is used for fast charging, there is a complete constant voltage charging interval in each charging cycle of the power battery. Figure 5 As shown, there is a clear correlation between the CVT feature and the degree of battery aging. As the battery ages, the constant voltage charging time is also extended accordingly. This is because the aging of the negative electrode material of the battery makes it impossible to quickly embed lithium ions, and the battery open circuit voltage rises faster, causing the constant voltage charging link to appear earlier. Therefore, this embodiment uses the charging data of the power battery obtained by the battery swap station to process the CVT feature and use it for the health status prediction of the power battery, which can obtain a higher prediction accuracy.

[0035] return Figure 1 Step S2: In the early stage of the life cycle of the power battery, a first feature data set is constructed using the feature data of each charging cycle arranged in chronological order and the corresponding battery maximum capacity label; in the middle stage of the life cycle of the power battery, a second feature data set is constructed using the feature data of each charging cycle arranged in chronological order and the corresponding battery maximum capacity label; the battery maximum capacity label is provided by the battery swap station and the battery owner.

[0036] Specifically, since the management method in this embodiment is mainly used for the late life cycle of the power battery, this embodiment also divides the acquired charging data into early, middle and late stages of the life cycle, and the charging data in different periods have different uses.

[0037] In a specific embodiment, the battery health state (SOH) is used to specifically divide the early, middle and late stages of the power battery life cycle. The early stage of the life cycle represents the first 40% of the power battery life cycle, the middle stage of the life cycle represents the middle 20% of the power battery life cycle, and the late stage of the life cycle represents the last 40% of the power battery life cycle. Of course, the specific division values ​​can also be adjusted according to actual conditions. For example, if the target power battery is planned to be retired and scrapped at 70% SOH, then the first 40% of the power battery life cycle is (100%-70%) * 40% = 12%, and the stage from 100% SOH to 88% SOH of the power battery is the early stage of the power battery life cycle. Correspondingly, the stage from 88% SOH to 82% SOH of the power battery is the middle stage of the power battery life cycle, and the stage from 82% SOH to 70% SOH of the power battery is the late stage of the power battery life cycle.

[0038] return Figure 1 , step S3: using the first feature data set as a training set of the SOH prediction framework, and using the second feature data set as a verification set of the SOH prediction framework.

[0039] Among them, the SOH prediction framework includes a BiLSTM neural network (bidirectional long short-term memory neural network) and a Transformer framework connected in series. The long short-term memory neural network (LSTM) performs well in processing time series data, especially in the field of predicting the evolution of battery capacity over time. The BiLSTM neural network is a bidirectional architecture based on LSTM, which processes the sequence bidirectionally through two LSTM layers, forward and reverse, so as to capture richer contextual information. It can be understood that in this embodiment, the charging cycle is the minimum time unit, the first charging cycle has corresponding IC peak value, IC peak position, CVT features, the second cycle has corresponding IC peak value, IC peak position, CVT features, ..., the Nth cycle has corresponding IC peak value, IC peak position, CVT features, and the feature data of each charging cycle belongs to time series data arranged in sequence. In addition, since the IC features and CVT features of this embodiment are calculated in advance, they do not need to be initially extracted by neural network convolution, which can effectively reduce the scale of training data.

[0040] Specifically, LSTM can capture long-term dependencies, which is critical for predicting the degree of attenuation of power batteries. Each LSTM unit consists of three gates (input, forgetting, and output) and a unit state, which can jointly regulate the flow of information. The BiLSTM neural network is better than the traditional LSTM network in capturing long-term dependencies and improving accuracy. In the prediction of battery aging, the capacity decay process of the battery is not only related to the current state, but also to multiple time steps in the past. The BiLSTM neural network can consider these previous and next dependencies at the same time. Therefore, in this embodiment, a BiLSTM neural network with a hidden layer dimension of 4 is added before the encoder of the Transformer framework. The BiLSTM neural network is used to process the initial time series data formed by the arrangement of the feature data of each charging cycle, and generate a more context-aware sequence, thereby providing a richer input for the subsequent Transformer framework.

[0041] In a specific embodiment, the BiLSTM neural network processes the input time series data through an input gate, a forget gate, a unit state, and an output gate.

[0042] Among them, the input gate controls the influence of the current input and the previous hidden state (responsible for short-term memory) on the state of the candidate unit. The input gate consists of two parts, and the mathematical expression is as follows:

[0043] in, represents the activation value of the input gate, represents the input of the current time step, represents the cell state at the previous time step, Indicates the candidate unit status, represents the Sigmoid activation function, is the weight matrix of the input gate, is the weight matrix of the candidate cell state, and The bias items representing the input gate and the candidate unit state respectively, and the initial time series data pre-sorted according to the charging data is input by the input gate.

[0044] The function of the forget gate is to control which information is deleted from the previous state. The mathematical expression of the forget gate is:

[0045] in: represents the input of the current time step, represents the cell state at the previous time step, represents the weight matrix of the forget gate, represents the Sigmoid activation function, represents the bias term of the forget gate, Represents the output of the forget gate.

[0046] The cell state (responsible for long-term memory) is the core of the BiLSTM neural network. The cell state determines the long-term memory. The neuron state is updated by merging the output of the forget gate and the input gate:

[0047] in, is the cell state at the previous time step, is the cell state at the current time step.

[0048] The output gate determines the next hidden state. The mathematical expression of the output gate is:

[0049] in, represents the weight matrix of the output gate, represents the bias term of the output gate, represents the output of the output gate, Represents the output of the LSTM unit at the current time step. The original time series data is passed through the BiLSTM neural network and then into the Transformer framework.

[0050] Furthermore, the structure of the Transformer framework consists of two main parts: encoder and decoder, which are composed of multiple stacked layers. Each layer includes the following main components: attention mechanism, feedforward neural network, residual connection and layer normalization.

[0051] The attention mechanism allows the model to focus on words at other locations in the time series data when encoding a word, thereby capturing global dependencies. (in, In this embodiment, the characteristic data corresponding to the nth charging cycle is mapped into an embedding vector and a position code is added to form a hidden state matrix ,in is the sequence length, is the embedding dimension. The attention mechanism maps the input to query (Query, Q), key (Key, K), and value (Value, V) through three matrices. In order to allow the model to capture different attention modes, the Transformer framework introduces a multi-head attention mechanism. The formula of the multi-head attention mechanism is as follows:

[0052] in, denote the learnable query weight, key weight, and value weight, respectively. represents the number of attention heads, represents the output transformation matrix, represents the similarity matrix, It represents the scaling factor to prevent the gradient from being too large, and softmax is used to normalize the similarity value into probability. The data processed by the BiLSTM neural network first enters the multi-head attention mechanism layer to learn the data processed by the neural network BiLSTM.

[0053] A separate feed-forward neural network is applied after each attention mechanism, and the feed-forward neural network has two fully connected layers:

[0054] in, is the first-layer weight matrix function (also known as the hidden layer dimension of the feedforward neural network), X represents the hidden state matrix, and the first-layer weight matrix function is used to map the input hidden state matrix X to a higher-dimensional space. is the second-layer weight matrix function, which is used to map the result after the activation function back to the original dimension. The feedforward neural network is used to connect the output of the multi-head attention mechanism layer with the output of the entire Transformer framework.

[0055] For each sublayer (attention mechanism or feedforward neural network), residual connection and layer normalization are applied. Residual connection helps gradient flow to avoid gradient disappearance. Layer normalization can speed up training and improve stability. Its expression is:

[0056] The Transformer encoder consists of N The encoder output is a context-dependent sequence representation. The decoder of the Transformer framework is similar to the encoder, but its multi-head attention is hidden to prevent the decoder from viewing information about future positions. The Transformer framework decoder obtains information from the output of the encoder. In a specific embodiment, the total output formula of the Transformer framework can be expressed as: The encoder generates a contextual representation of the sequence , the decoder is based on the target input Generate predictions , the calculation formula is as follows:

[0057] return Figure 1, step S4: in the late stage of the life cycle of the power battery, the characteristic data of each charging cycle is input into the trained and verified SOH prediction framework to estimate the maximum capacity of the battery after each charging cycle.

[0058] Specifically, this embodiment uses IC features and CVT features in combination, and both features are related to the aging of graphite at the negative electrode of the battery. Usually, the IC feature has a high correlation with the negative electrode of the battery, but the voltage needs to be differentially processed, which is easy to amplify signal noise; while the CVT feature does not require differential processing, and the feature fidelity is high. Therefore, this embodiment can better identify the battery aging process by combining IC features with CVT features to improve the capacity prediction accuracy in the late stage of the power battery life cycle. In addition, this embodiment uses the BiLSTM neural network and the Transformer framework connected in series as the SOH prediction framework. Among them, the BiLSTM neural network has good time series data processing capabilities, and the application effect in this embodiment is better. The multi-head attention mechanism in the Transformer framework can efficiently mine potential mapping relationships from training data.

[0059] The present invention combines the BiLSTM neural network with the Transformer framework, so that the SOH prediction framework can capture the relationship between the characteristic data of the power battery and the maximum capacity or aging degree of the battery from a relatively small training data. In actual operation, the first characteristic data set obtained in the early stage of the power battery life cycle is used as the training set to train the SOH prediction framework, and the SOH prediction framework is used in the late stage of the power battery life cycle to estimate the maximum capacity of the power battery after each charging cycle. It has been verified that a higher prediction accuracy can be achieved.

[0060] return Figure 1 , step S5: determining the SOH value according to the estimated ratio of the maximum capacity of the battery to the initial maximum capacity; determining the charging cut-off current of the power battery according to the SOH value and whether the battery has a capacity drop; wherein the initial maximum capacity is provided by the power battery manufacturer.

[0061] Specifically, power batteries in battery swap stations are generally charged with constant current and constant voltage. During constant voltage charging, the charging current will continue to decrease. When it is lower than a certain value, charging will stop. This is called the charging cut-off current. The industry usually takes 1C / 20, where 1C is the rated capacity of the battery pack. The larger the charging cut-off current, the earlier the battery charging process stops, the less power is charged into the battery, and the charging depth is low, which is conducive to extending the battery life. Therefore, in the late stage of the power battery's life cycle, fine management of the power battery through the battery charging cut-off current can increase the service life of the power battery. In the early stage of the power battery's life cycle, due to the high battery consistency, high capacity retention and relatively linear capacity decay, the demand for fine management is low. Therefore, from the perspective of management cost, fine management is not required.

[0062] Among them, in the late life cycle of the power battery, the SOH prediction framework predicts the maximum capacity of the power battery after each charging cycle. By connecting the predicted maximum capacities of each battery together, the future capacity decay curve of the battery can be obtained. The future capacity decay curve indicates how the capacity will decay in the future when the battery continues to use historical operating conditions. According to the capacity decay curve, for example, if accelerated capacity decay occurs, the charge and discharge depth of the battery can be reduced by increasing the charge cut-off current in management, thereby delaying battery decay. It can be understood that the core of the management method in this embodiment is: predicting future capacity decay through historical data (specifically, the degree of capacity decay is characterized by the SOH value); determining a suitable charging management strategy based on the predicted capacity decay curve; and extending the service life of the battery at the expense of a small amount of vehicle mileage reduction (reducing the charge and discharge depth of the power battery).

[0063] The SOH value in step S5 is predicted according to the above steps, and whether the capacity of the power battery drops is determined according to the battery capacity attenuation acceleration. The mathematical expression of the battery capacity attenuation acceleration is:

[0064] in, Indicates the acceleration of battery capacity attenuation, Capacity indicates the current capacity of the battery, and Cycle indicates the number of recorded charging cycles. The unit is , Generally, it is a negative number. Taking the bench test battery as an example, the nominal capacity of the bench test battery is 5Ah (i.e. 1C). When it is less than 0.1 (i.e., 1C / 50), it can be considered that the capacity attenuation has experienced a dive phenomenon. In actual operation, other standards can also be selected as the basis for determining whether the power battery has experienced a capacity dive phenomenon, which is not specifically limited in this embodiment.

[0065] In a specific embodiment, the power battery is divided into six levels for refined management according to the SOH value of the power battery and whether the power battery has a capacity dive. Specifically, the first level: the battery has a capacity dive, and the charging cut-off current is maintained at 0.5C; the second level: the battery has no capacity dive, and the SOH value>95%, the charging cut-off current is maintained at 0.05C; the third level: the battery has no capacity dive, and 95%>SOH value>90%, the charging cut-off current is maintained at 0.1C; the fourth level: the battery has no capacity dive, and 90%>SOH value>85%, the charging cut-off current is maintained at 0.2C; the fifth level: the battery has no capacity dive, and 85%>SOH value>80%, the charging cut-off current is maintained at 0.3C; the sixth level: the battery has no capacity dive, and 80%>SOH value, the charging cut-off current is maintained at 0.5C. When the SOH value is <70%, the battery is no longer finely managed, but it is recommended to exit the vehicle power battery application scenario.

[0066] In another specific embodiment, in order to verify the effect of the CVT feature proposed in this embodiment in battery SOH prediction, this embodiment uses the Pearson correlation coefficient to quantify the correlation between the feature and the battery capacity. The Pearson correlation coefficient is an indicator widely used to measure linear relationships, which defines the correlation coefficient between two variables as:

[0067] in, represents the correlation coefficient; and Respectively represent the average value of characteristic value and battery capacity value; x i represents the characteristic value of the i-th sample; y i represents the battery capacity value of the i-th sample; n represents the sample size. Correlation coefficient The range is -1 to 1, where values ​​close to 1 or -1 indicate a stronger relationship and values ​​close to 0 indicate a weaker relationship.

[0068] like Figure 6 As shown, the battery capacity, IC characteristics, and CVT characteristics show high correlation coefficients, which indicates that they can effectively capture the battery aging characteristics.

[0069] In another specific embodiment, in combination Figure 7In this embodiment, the first feature data set obtained from the charging data of the first 40% of the power battery life cycle is used as a training set for the SOH prediction framework training; the second feature data set obtained from the charging data of the middle 20% of the power battery life cycle is used as a validation set to reduce the risk of overfitting in training. And in the last 40% of the power battery life cycle, the maximum capacity of the power battery after each charging cycle is estimated, and the estimated capacity decay curve can be obtained by connecting the maximum capacity of the battery after each charging cycle in chronological order. After verification, the SOH prediction framework proposed in this embodiment performs well in the estimation of the capacity decay curve, and the RMSE value (root mean square error) under most conditions is less than 0.04Ah. Specifically, this embodiment is provided with 12 groups of power battery samples (#1~#12) for verification, and the charging conditions of each group of power batteries are specifically set as follows: the power battery samples #1~#6 are charged at 25°C, the power battery samples #7~#12 are charged at 0°C, the power battery samples #1, #2, #7, and #8 are charged with a charging current of 25 A, the power battery samples #3, #4, #9, and #10 are charged with a charging current of 10 A, and the power battery samples #5, #6, #11, and #12 are charged with a charging current of 15 A.

[0070] The best case, combined with Figure 7 (a), the RMSE value corresponding to the estimation result of the #5 power battery sample is equal to 0.0053Ah (normalized error 0.11%); in the worst case, the RMSE value corresponding to the estimation result of the #3 power battery sample is equal to 0.078Ah (normalized error 1.6%). This result verifies the effectiveness of the features selected in this embodiment and the robustness of the model training process. Figure 7 (b) Even for the #2 power battery sample that has accelerated aging in the late stage of its life cycle, the SOH prediction framework in this embodiment can achieve a high prediction accuracy; combined with Figure 7 (c) At 0°C, the #8 power battery sample shows accelerated decay in the late life cycle, and the SOH prediction framework in this embodiment can accurately identify this. In addition, it has been verified that if the battery capacity drops, the SOH prediction framework also captures the capacity drop behavior. This embodiment estimates the capacity decay curve of each power battery sample 4 times, combined with Figure 7(d), the horizontal axis represents the true value, the vertical axis represents the estimated value, the color closer to yellow represents the denser the data points, and the color closer to purple represents the sparser the data points. It is worth noting that the SOH prediction framework (BiLSTM-Transformer model) of this embodiment can still maintain high accuracy on power battery samples aged under multiple working conditions, demonstrating its generalization ability in complex scenarios, which verifies the effectiveness of the IC features, CVT features and SOH prediction framework proposed in this embodiment in predicting battery capacity attenuation.

[0071] In another specific embodiment, in combination Figure 8 As shown in Table 1, in order to evaluate the performance of the proposed BiLSTM-Transformer framework, this embodiment conducts comparative experiments and RMSE error analysis of different types of models. Support vector regression (SVR) is a machine learning algorithm commonly used for regression problems. It has been often used in the past for SOH prediction of power batteries. Therefore, this embodiment selects SVR as a comparative model for estimation effect. In addition, this embodiment selects the LSTM model and the Transformer model in the BiLSTM-Transformer model as the control group. This embodiment uses the above four networks to estimate the life of power battery samples with high SOH linearity (#5 power battery sample) and power battery samples with low SOH linearity (#2 power battery sample). All experimental groups use the same feature data, and the first feature data set obtained by using the charging data of the first 40% of the power battery life cycle is used as the training set for SOH prediction framework training, and the second feature data set obtained by using the charging data of the middle 20% of the power battery life cycle is used as the validation set to reduce the risk of overfitting in training. And after 40% of the power battery life cycle, the maximum capacity of the power battery after each charging cycle is estimated, and the capacity decay curve can be obtained by connecting the maximum capacity of the battery after each charging cycle.

[0072] The results show that the best battery capacity estimation effect can be obtained by using the SOH prediction framework (BiLSTM-Transformer model) in this embodiment. The Transformer model is close to the BiLSTM-Transformer model, but the estimation error is slightly higher than the BiLSTM-Transformer model, which shows that the network fusion method can improve the performance of the model. In addition, regardless of the linearity of the capacity decay curve of the battery capacity decay, a high estimation accuracy can be achieved, which shows that the fusion method effectively combines the advantages of different features and more comprehensively reflects the aging characteristics of the battery. The SVR model and the LSTM model achieved different prediction effects on the two batteries. The SVR prediction effect is better than the LSTM model for batteries with higher linearity (such as the #2 power battery sample), but it is slightly inferior to the LSTM model for batteries with capacity diving (such as the #5 power battery sample). This is also the reason why the LSTM model is more and more widely used in battery state estimation compared with the SVR model. Even though the LSTM model has a certain ability to capture capacity diving, there is still a large error at the end of the battery life.

[0073] Table 1 Comparison of estimation errors of SVR, LSTM, Transformer, and BiLSTM-Transformer

[0074] In another specific embodiment, this embodiment provides a power battery health status management system for a vehicle swap station, which includes: An acquisition module is used to acquire charging data of the power battery in the battery swap station; the charging data is used to process and obtain characteristic data of each charging cycle, the characteristic data including battery incremental capacity characteristics and constant voltage charging duration characteristics; A feature data set construction module is used to construct a first feature data set using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label in the early stage of the power battery life cycle; and to construct a second feature data set using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label in the middle stage of the power battery life cycle; A model training module, used to use the first feature data set as a training set of the SOH prediction framework, and use the second feature data set as a validation set of the SOH prediction framework; An estimation module, used for inputting the characteristic data of each charging cycle into the trained and verified SOH prediction framework in the late stage of the power battery life cycle to estimate the maximum capacity of the battery after each charging cycle respectively; A management module is used to determine the SOH value according to the ratio of the estimated maximum capacity of the battery to the initial maximum capacity; determine the charging cut-off current of the power battery according to the SOH value and whether the battery has a capacity drop phenomenon It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0076] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0078] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for managing the health status of a power battery in a vehicle battery swap station, characterized in that: include: Obtain charging data of power batteries in battery swap stations; The charging data is used to process and obtain characteristic data of each charging cycle, wherein the characteristic data includes battery incremental capacity characteristics and constant voltage charging duration characteristics; In the early stage of the life cycle of the power battery, a first feature data set is constructed using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label; In the middle of the life cycle of the power battery, a second feature data set is constructed using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label; Using the first feature data set as a training set of the SOH prediction framework, and using the second feature data set as a validation set of the SOH prediction framework; In the late stage of the life cycle of the power battery, the characteristic data of each charging cycle is input into the trained and verified SOH prediction framework to estimate the maximum capacity of the battery after each charging cycle; Determining the SOH value according to the ratio of the estimated maximum capacity of the battery to the initial maximum capacity; The charging cut-off current of the power battery is determined according to the SOH value and whether the battery capacity drops.

2. A method for managing the health status of a power battery in a vehicle battery swap station according to claim 1, characterized in that: The charging data includes the number of charging cycles, the charging time of each charging cycle, the charging current, the charging voltage, and the charging power.

3. The method for managing the health status of power batteries in a vehicle battery swap station according to claim 2, characterized in that: The battery incremental capacity characteristics include an incremental capacity peak position and an incremental capacity peak value of an incremental capacity curve; The calculation formula of the incremental capacity curve is: ; Where IC represents the incremental capacity, Q represents the charge capacity, V represents the charge voltage, I represents the charge current, and t represents the sampling time.

4. The method for managing the health status of power batteries in a vehicle battery swap station according to claim 2, characterized in that: The constant voltage charging time characteristic includes the constant voltage charging time, and the calculation formula of the constant voltage charging time is: ; In the formula, CVT represents the constant voltage charging duration, CVT1 represents the time point when constant current charging is converted to constant voltage charging, and CVT2 represents the time point when constant voltage charging ends.

5. The method for managing the health status of a power battery in a vehicle battery swap station according to claim 1, characterized in that: The incremental capacity curve is filtered using a Savitzky-Golay filtering method.

6. The method for managing the health status of a power battery in a vehicle battery swap station according to claim 1, characterized in that: The SOH prediction framework includes a BiLSTM neural network and a Transformer framework connected in series.

7. The method for managing the health status of a power battery in a vehicle battery swap station according to claim 1, characterized in that: The step of determining the charging cut-off current of the power battery according to the SOH value and whether the battery has a capacity drop phenomenon comprises: If the battery capacity drops, the charge cut-off current remains at 0.5C; If the battery capacity does not drop and SOH>95%, the charge cut-off current remains at 0.05C; If the battery capacity does not drop and 95%>SOH>90%, the charge cut-off current remains at 0.1C; If the battery capacity does not drop and 90%>SOH>85%, the charge cut-off current remains at 0.2C; If the battery capacity does not drop and 85%>SOH>80%, the charge cut-off current remains at 0.3C; If the battery capacity does not drop and 80%>SOH, the charge cut-off current remains at 0.5C.

8. A power battery health status management system for a vehicle battery swap station, characterized in that: include: An acquisition module is used to obtain charging data of the power battery in the battery swap station; The charging data is used to process and obtain characteristic data of each charging cycle, wherein the characteristic data includes battery incremental capacity characteristics and constant voltage charging duration characteristics; A feature data set construction module, used to construct a first feature data set in the early stage of the life cycle of the power battery by using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label; In the middle of the life cycle of the power battery, a second feature data set is constructed using the feature data of each charging cycle arranged in time sequence and the corresponding battery maximum capacity label; A model training and verification module, used to use the first feature data set as a training set of the SOH prediction framework, and use the second feature data set as a verification set of the SOH prediction framework; An estimation module, used for inputting the characteristic data of each charging cycle into the trained and verified SOH prediction framework in the late stage of the power battery life cycle to estimate the maximum capacity of the battery after each charging cycle respectively; A management module, used to determine the SOH value according to the ratio of the estimated maximum capacity of the battery to the initial maximum capacity; The charging cut-off current of the power battery is determined according to the SOH value and whether the battery capacity drops.

Images