Multi-dimensional fusion battery health assessment method

Abstract

The invention provides a multi-dimensional fusion battery health assessment method, which comprises the following steps of: acquiring power consumption data of a battery in an energy storage power station, acquiring an SOH (state of health) predicted value at an edge end, and executing cloud estimation when a difference exceeds a threshold value; when the historical sequence exists, predicting the historical health data by using a recurrent neural network Bi-LSTM to obtain a battery SOH predicted value; and when the historical sequence does not exist, extracting battery parameter dimension feature data, inputting the trained ELM model, and outputting a battery SOH predicted value. According to the method, two data driving methods are combined, and health state estimation is carried out by using an ELM algorithm when historical health data are missing, so that a health state initial sequence with relatively high precision is obtained; when the health state historical sequence meets the Bi-LSTM requirement, health state prediction is carried out by using the Bi-LSTM; the method has the advantages of reliable precision, fewer monitoring parameters, low cost, better adaptability and easiness in online application.

Description

Technical Field

[0001] The present invention relates to the field of intelligent online evaluation of battery health status, and more specifically, to an algorithm for evaluating the health status of a lithium-ion battery. Background Art

[0002] During the application of electrochemical energy storage, the availability of the electrochemical energy storage system is reduced, the failure rate is frequent, and even serious fires may occur due to the lack of battery consistency, control system errors, and life detection methods. For example, an explosion occurred in the integrated photovoltaic storage and charging project of Beijing Fuweis Oil and Gas Technology Co., Ltd. The direct cause was an internal short circuit failure in the battery cell, which caused the battery to catch fire and then caused a tragedy, resulting in huge property losses and casualties.

[0003] The core of the healthy operation of new energy storage power stations is the safe operation of energy storage batteries. Therefore, it is of great significance to establish a complete theoretical system of battery health status assessment and detection algorithms and study the key technologies of the healthy operation and status of new energy storage batteries. At the same time, the battery status is an important factor affecting the economic operation of energy storage power stations. Battery aging causes an increase in the operating cost of the battery energy storage system of energy storage power stations.

[0004] However, the battery management system equipment and technology related to energy storage power stations are not yet mature, and the current measurement technology cannot obtain the parameters that characterize the internal health of the battery online. The precise parameters related to the internal health of the battery are only measured offline in the laboratory. Therefore, the research on efficient, robust and accurate online estimation technology of LIB health status has become a hot spot and difficulty in the current energy storage battery industry.

[0005] Since current measurement technology cannot obtain parameters that characterize the internal health of the battery online, the precise parameters related to the internal health of the battery need to be measured offline in the laboratory, so it is necessary to use some algorithm to obtain the internal health status of the battery. On the one hand, when the battery is working, it only responds to a few state quantities such as voltage, current, and temperature, which undoubtedly puts higher requirements on algorithm design. On the other hand, using fewer state quantities to evaluate more parameters and multiple states also increases the complexity and operation cost of the algorithm. As the battery health status deteriorates, algorithms based on circuit, electrochemical and other models will be distorted to varying degrees, affecting the safety monitoring results of energy storage power stations.

[0006] However, the various reactions and aging mechanisms inside energy storage batteries are complex, and existing research methods have certain shortcomings, or model parameters are difficult to collect and are greatly affected by the environment. Data-driven battery health characteristic monitoring requires a large amount of comprehensive offline data to establish a model, and the model is inefficient and relies on experimental data.

[0007] It can be seen that in order to improve the safety performance of energy storage power stations, reduce potential safety hazards of energy storage power stations, and ensure that the batteries in energy storage power stations emit normal power, it is necessary to propose a new algorithm for comprehensively evaluating the health status of batteries in multiple dimensions. Summary of the invention

[0008] The purpose of the present invention is to propose a multi-dimensional fusion battery health assessment method for the estimation of battery health status. The method of the present invention has reliable accuracy, high battery evaluation dimension, good adaptability, and is easy to apply online; it adopts a method combining a dual-scale extended Kalman filter algorithm, a long short-term memory recurrent neural network Bi-LSTM in machine learning, and an attention mechanism to evaluate the current health status of the battery in multi-dimensional fusion.

[0009] The technical solution of the present invention is:

[0010] A multi-dimensional integrated battery health assessment method, the method comprising:

[0011] Obtain the power consumption data of the batteries in the energy storage power station, and calculate the battery health status prediction value (SOH prediction value) online at the edge. When the difference between the current SOH prediction value and the previous SOH prediction value exceeds the threshold, perform cloud estimation;

[0012] Upload the battery data collected by the sensor to the cloud for data-driven SOH estimation;

[0013] When there is a historical battery health state sequence, use the recurrent neural network Bi-LSTM to predict the historical health data, obtain the battery SOH prediction value, and add it to the battery health state sequence;

[0014] When there is no historical battery health status sequence, the battery parameter dimension feature data collected by the sensor is extracted, input into the trained ELM model, output the battery SOH prediction value, and add it to the battery health status sequence.

[0015] Furthermore, a dual Kalman filter algorithm is used at the edge to evaluate the battery health status and obtain the battery SOH prediction value.

[0016] Furthermore, the recurrent neural network Bi-LSTM uses a logic gate structure to control and protect information.

[0017] Furthermore, the battery parameter dimension characteristic data includes the maximum peak temperature HF3, the peak-to-peak temperature rise HF4, the number of battery charge and discharge cycles HF5, and the cumulative number of battery charge and discharge hours HF6.

[0018] Furthermore, the maximum peak temperature HF3 and the peak-to-peak-to-valley temperature rise HF4 are obtained in the following manner:

[0019] Use sensors to obtain temperature data during battery cycle conditions;

[0020] Discrete wavelet transform DWT is used to reduce the noise of temperature data;

[0021] The maximum peak temperature HF3 and the peak-to-peak-valley temperature rise value HF4 are selected as parameter dimension data from the temperature data after noise reduction processing;

[0022] The discrete wavelet transform DWT adopts the Daubechies wavelet function.

[0023] Furthermore, the training process of the ELM model includes:

[0024] Providing a training data set, each training data in the training data set includes a battery SOH value and corresponding battery parameter dimension feature data;

[0025] Input the training data set into the ELM model;

[0026] The attention mechanism is used to fuse the battery parameter dimension feature data to obtain the fused Co-HF features;

[0027] The battery parameter dimension feature data and the fused Co-HF features are used as input vectors for training to obtain the ELM model.

[0028] Furthermore, the ELM model adopts a single hidden layer feedforward neural network, including an input layer, a hidden layer and an output layer.

[0029] Furthermore, data-driven SOH estimation also includes:

[0030] Get the remaining battery life RUL value HF2 and perform a health assessment on the battery time dimension; specifically:

[0031] Preset the length of the battery health status sequence and obtain the original battery health status sequence;

[0032] After obtaining the battery SOH prediction value, add it to the original sequence and delete the initial header data of the original sequence to keep the sequence length unchanged;

[0033] This is repeated until the predicted SOH value is less than 80%. The remaining battery life RUL value HF2 is obtained by subtracting the number of battery charge and discharge cycles of the last data in the original sequence from the current number of battery charge and discharge cycles.

[0034] Furthermore, data-driven SOH estimation also includes:

[0035] The voltage difference assessment SOH value HF7 is obtained using the equivalent difference model to perform a health assessment on the battery voltage difference dimension.

[0036] An electronic device, comprising:

[0037] A memory storing executable instructions;

[0038] A processor runs the executable instructions in the memory to implement the multi-dimensional integrated battery health assessment method.

[0039] Beneficial effects of the present invention:

[0040] The multi-dimensional integrated battery health assessment method of the present invention adopts a cloud-edge collaborative battery health status estimation solution;

[0041] At the edge, the battery dual Kalman filter algorithm is used to estimate the battery health status based on the battery voltage and current dimensions, and the estimation result is used as the basis for health status evaluation; when the Kalman filter algorithm has accuracy fluctuations, it is switched to the cloud;

[0042] In the cloud, data-driven battery SOH estimation is performed, which is divided into two models. One is to use the Bi-LSTM model trained with the historical health status in the cloud to predict the current SOH value and the current RUL as an important basis for evaluating the battery health status; the other is to use HF3, HF4, HF5, and HF6 for attention mechanism fusion, and the ELM model predicts the current SOH value. The use of historical health data for health prediction is highly robust and less affected by noise. When the difference algorithm is greatly affected by noise, the use of Bi-LSTM can reduce the estimation error of SOH; when the historical health status sequence is missing, the second method is used to estimate the health status, and the temperature characteristic values ​​are extracted from the entire operating cycle using data processing methods such as wavelet transform, which has higher accuracy.

[0043] The present invention combines two data-driven methods. When historical health data is missing, the Attention-ELM algorithm is used to estimate the health status to obtain a more accurate initial sequence of health status. When the health status historical sequence meets the Bi-LSTM requirements, the Bi-LSTM is used to predict the health status. When the battery life is close to EOL, the Attention-ELM algorithm with higher health estimation accuracy is used again on the battery.

[0044] The present invention selects seven characteristic data for comprehensive battery health assessment, namely, the battery historical sequence SOH prediction value HF1, the battery RUL value HF2, the maximum peak temperature HF3, the peak-to-peak temperature rise HF4, the battery charge and discharge cycle number HF5, the battery cumulative charge and discharge hours HF6 and the voltage difference estimated SOH value HF7, and extracts health characteristic values ​​from three different dimensions, namely, the inconsistency dimension, the parameter dimension, and the time dimension data.

[0045] In terms of parameter dimensions, BMS can monitor the number of battery charge and discharge cycles and the cumulative charge and discharge time, and the maximum peak temperature and peak-to-peak temperature rise as temperature characteristic data. Since the battery temperature data is real-time data, it can be measured by sensors;

[0046] For voltage inconsistency data, the battery SOH is calculated using the equivalent difference model, and the estimated data can be used as the characteristics of the voltage difference;

[0047] For battery time dimension health data, two data-driven methods are combined to estimate SOH and RUL.

[0048] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and other objects, features and advantages of the present invention will become more apparent through a more detailed description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present invention.

[0050] Figure 1 A schematic diagram of the multi-dimensional fusion battery health assessment method of the present invention is shown.

[0051] Figure 2 The flowchart of obtaining the remaining service life RUL value of the battery in the present invention is shown. DETAILED DESCRIPTION

[0052] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0053] Example 1

[0054] like Figure 1 As shown, this embodiment provides a multi-dimensional integrated battery health assessment method, the method comprising:

[0055] Obtain the power consumption data of the batteries in the energy storage power station, and use the dual Kalman filter algorithm on the edge to calculate the battery health status prediction value, that is, the SOH prediction value, online. When the difference between the current SOH prediction value and the previous SOH prediction value exceeds the threshold, it means that there is a large gap. It is considered that the Kalman filter algorithm has a large fluctuation, and cloud estimation is performed. The trigger conditions of cloud estimation are set as follows:

[0056]

[0057] Upload the battery data collected by the sensor to the cloud for data-driven SOH estimation;

[0058] When there is a historical battery health state sequence, use the recurrent neural network Bi-LSTM to predict the historical health data, obtain the battery SOH prediction value, and add it to the battery health state sequence;

[0059] When there is no historical battery health status sequence, the battery parameter dimension feature data collected by the sensor is extracted, input into the trained ELM model, output the battery SOH prediction value, and add it to the battery health status sequence.

[0060] Among them, the parameter identification part of the dual Kalman filter is divided into long time scale and short time scale, and the parameters such as ohmic internal resistance, polarization internal resistance, polarization capacitance, charge transfer internal resistance, charge transfer capacitance, battery capacity and so on with slowly changing parameters are identified in the long time scale.

[0061] In this embodiment, at the edge, starting from the battery voltage and current dimensions, a battery dual Kalman filter algorithm is used to estimate the battery health state, and the estimation result is used as the basis for health state evaluation; when the Kalman filter algorithm has accuracy fluctuations, it is switched to the cloud; in the cloud, data-driven battery SOH estimation is performed.

[0062] Specifically, data-driven battery SOH estimation can be divided into two methods;

[0063] First, when there is a historical battery health state sequence, use the recurrent neural network Bi-LSTM to predict the historical health data, obtain the battery SOH prediction value, and add it to the battery health state sequence;

[0064] The recurrent neural network Bi-LSTM uses a logic gate structure to control and protect information, and is specifically designed for long-term dependency information:

[0065] f c (t) =σ[W fX x (t) +W fh h (t-1) +b f ]

[0066] Among them, x(t) represents the neuron input at time t, and h(t-1) represents the hidden layer information at time t-1. The core of Bi-LSTM lies in its internal state quantity s c , which is located at the core of each neuron and activated by a linear function. The internal state of the Bi-LSTM RNN is the carrier for storing or forgetting historical information.

[0067] When the loss function Loss after training is less than the upper bound θ, that is:

[0068] Loss(U,I,T,SOC)<θ

[0069] The historical health status of a preset number of times is used as a vector input into the Bi-LSTM model, and the health information is output by the Bi-LSTM neural network to obtain the health status characteristics.

[0070] Using historical health data for health prediction is more robust and less affected by noise. When the difference algorithm is greatly affected by noise, using Bi-LSTM can reduce the estimation error of SOH.

[0071] Second: When there is no historical battery health state sequence, extract the battery parameter dimension feature data collected by the sensor, input the trained ELM model, output the battery SOH prediction value, and add it to the battery health state sequence;

[0072] Specifically, the maximum peak temperature HF3 of the battery, the peak-to-valley temperature rise HF4, the number of battery charge and discharge cycles HF5, and the cumulative charge and discharge hours HF6 of the battery are input into the trained Attention-ELM model to predict the SOH value at the current moment.

[0073] In this embodiment, the training process of the ELM model includes:

[0074] Providing a training data set, each training data in the training data set includes a battery SOH value and corresponding battery parameter dimension feature data;

[0075] Input the training data set into the ELM model;

[0076] The attention mechanism is used to fuse the battery parameter dimension feature data to obtain the fused Co-HF features;

[0077] For the battery input feature X = [x1, x2, x3, ..., x n ], first the attention mechanism introduces the query vector q, which represents the request value of the current task, and designs a score function to evaluate the correlation between the query vector q and the input vector X, and then normalizes this correlation data using the softmax function, as shown in the formula:

[0078]

[0079] Where: a i is feature x i The corresponding attention coefficient, (x i ,q) is the attention score function, and the additive model is used as the score function formula:

[0080] s(x i ,q)=v T tanh(UX+Vq)

[0081] Wherein, parameters v, U and V are parameters that need to be learned.

[0082] The battery parameter dimension feature data and the fused Co-HF features are used as input vectors for training to obtain the ELM model;

[0083] The ELM model adopts a single hidden layer feedforward neural network, including an input layer, a hidden layer and an output layer.

[0084] like Figure 2 As shown, it is a flowchart for obtaining the remaining service life RUL value of the battery in the present invention, obtaining the remaining service life RUL value HF2 of the battery, and performing a health assessment on the battery time dimension; specifically:

[0085] Preset the length of the battery health status sequence and obtain the original battery health status sequence;

[0086] After obtaining the battery SOH prediction value, add it to the original sequence and delete the initial header data of the original sequence to keep the sequence length unchanged;

[0087] This is repeated until the predicted SOH value is less than 80%. The remaining battery life RUL value HF2 is obtained by subtracting the number of battery charge and discharge cycles of the last data in the original sequence from the current number of battery charge and discharge cycles.

[0088] The present invention combines two data-driven methods. When historical health data is missing, the Attention-ELM algorithm is used to estimate the health status to obtain a more accurate initial sequence of health status. When the health status historical sequence meets the Bi-LSTM requirements, the Bi-LSTM is used to predict the health status. When the battery life is close to EOL, the Attention-ELM algorithm with higher health estimation accuracy is used again on the battery.

[0089] In one example, the data-driven SOH estimation of the present invention further includes: using an equivalent difference model to obtain a voltage difference assessment SOH value HF7, and performing a health assessment on the battery voltage difference dimension.

[0090] In one example, starting from the battery temperature, characteristic temperatures during battery cycle conditions are extracted: the maximum peak temperature and the peak-to-peak maximum temperature rise values ​​are used as health parameters; due to the accuracy limitation of the temperature sensor, there is a large amount of sensor noise in the battery temperature acquisition signal. The present invention uses discrete wavelet transform (DWT) to denoise the temperature data in the data set, so as to extract relevant battery SOH characteristics from the temperature dimension.

[0091] Discrete wavelet transform is a signal processing technology used to decompose a signal into components of different frequencies. DWT is a lossless transform and no information will be lost in the process of decomposing and reconstructing the signal. The inverse transform of the low-frequency and high-frequency components obtained by decomposition can reconstruct the original time domain signal. It has the advantages of multi-resolution analysis, localized features and compression performance, and is suitable for signal processing and task analysis in various fields.

[0092] Common wavelet functions include Haar wavelet, Daubechies wavelet, Symlet wavelet, Coiflet wavelet and Morlet wavelet. Daubechies wavelet has the characteristics of compact support and multi-scale analysis and is widely used in signal compression and denoising. Therefore, the present invention selects Daubechies wavelet for wavelet transform to obtain characteristic temperature parameters.

[0093] Embodiment 2:

[0094] This embodiment provides an electronic device, the electronic device comprising:

[0095] at least one processor; and,

[0096] a memory communicatively connected to the at least one processor; wherein,

[0097] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the multi-dimensional fusion battery health assessment method described in Example 1.

[0098] The electronic device according to an embodiment of the present disclosure includes a memory and a processor, and the memory is used to store non-temporary computer-readable instructions. Specifically, the memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may, for example, include a read-only memory (ROM), a hard disk, a flash memory, etc.

[0099] The processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of the present disclosure, the processor is used to run the computer-readable instructions stored in the memory.

[0100] Those skilled in the art should be able to understand that in order to solve the technical problem of how to obtain a good user experience, the present embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the protection scope of the present disclosure.

[0101] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0102] Embodiment 3:

[0103] This embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the multi-dimensional fusion battery health assessment method described in Example 1.

[0104] According to the computer-readable storage medium of the embodiment of the present disclosure, non-transitory computer-readable instructions are stored thereon. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the above-mentioned methods of each embodiment of the present disclosure are executed.

[0105] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0106] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A multi-dimensional integrated battery health assessment method, characterized in that: The method includes: Obtain the power consumption data of the batteries in the energy storage power station, and calculate the battery health status prediction value (SOH prediction value) online at the edge. When the difference between the current SOH prediction value and the previous SOH prediction value exceeds the threshold, perform cloud estimation; Upload the battery data collected by the sensor to the cloud for data-driven SOH estimation; When there is a historical battery health state sequence, use the recurrent neural network Bi-LSTM to predict the historical health data, obtain the battery SOH prediction value, and add it to the battery health state sequence; When there is no historical battery health status sequence, the battery parameter dimension feature data collected by the sensor is extracted, input into the trained ELM model, output the battery SOH prediction value, and add it to the battery health status sequence.

2. The multi-dimensional fusion battery health assessment method according to claim 1 is characterized in that: A dual Kalman filter algorithm is used at the edge to evaluate the battery health status and obtain the battery SOH prediction value.

3. The multi-dimensional fusion battery health assessment method according to claim 1 is characterized in that: The recurrent neural network Bi-LSTM uses a logic gate structure to control and protect information.

4. The multi-dimensional fusion battery health assessment method according to claim 1 is characterized in that: The battery parameter dimension characteristic data includes the maximum peak temperature HF3, the peak-to-peak temperature rise HF4, the number of battery charge and discharge cycles HF5, and the cumulative number of battery charge and discharge hours HF6.

5. The multi-dimensional fusion battery health assessment method according to claim 4 is characterized in that: The maximum peak temperature HF3 and the peak-to-peak-to-valley temperature rise HF4 are obtained in the following manner: Use sensors to obtain temperature data during battery cycle conditions; Discrete wavelet transform DWT is used to reduce the noise of temperature data; The maximum peak temperature HF3 and the peak-to-peak-valley temperature rise value HF4 are selected as parameter dimension data from the temperature data after noise reduction processing; The discrete wavelet transform DWT adopts the Daubechies wavelet function.

6. The multi-dimensional fusion battery health assessment method according to claim 1 is characterized in that: The training process of the ELM model includes: Providing a training data set, each training data in the training data set includes a battery SOH value and corresponding battery parameter dimension feature data; Input the training data set into the ELM model; The attention mechanism is used to fuse the battery parameter dimension feature data to obtain the fused Co-HF features; The battery parameter dimension feature data and the fused Co-HF features are used as input vectors for training to obtain the ELM model.

7. The multi-dimensional fusion battery health assessment method according to claim 6 is characterized in that: The ELM model adopts a single hidden layer feedforward neural network, including an input layer, a hidden layer and an output layer.

8. The multi-dimensional fusion battery health assessment method according to claim 1 is characterized in that: Data-driven SOH estimation also includes: Get the remaining battery life RUL value HF2 and perform a health assessment on the battery time dimension; specifically: Preset the length of the battery health status sequence and obtain the original battery health status sequence; After obtaining the battery SOH prediction value, add it to the original sequence and delete the initial header data of the original sequence to keep the sequence length unchanged; This is repeated until the predicted SOH value is less than 80%. The remaining battery life RUL value HF2 is obtained by subtracting the number of battery charge and discharge cycles of the last data in the original sequence from the current number of battery charge and discharge cycles.

9. The multi-dimensional fusion battery health assessment method according to claim 1 is characterized in that: Data-driven SOH estimation also includes: The voltage difference assessment SOH value HF7 is obtained using the equivalent difference model to perform a health assessment on the battery voltage difference dimension.

10. An electronic device, characterized in that: The electronic device comprises: A memory storing executable instructions; A processor runs the executable instructions in the memory to implement the multi-dimensional integrated battery health assessment method.

Images