Power battery terminal voltage prediction method based on multi-modal data fusion and charging pile

Abstract

The invention relates to the technical field of electric digital processing, in particular to a power battery terminal voltage prediction method based on multi-modal data fusion and a charging pile. The method comprises the steps that in the charging process of the electric vehicle, electrochemical parameters of a battery are collected in real time through a charging pile; collecting the charging noise of the battery through an acoustic sensor deployed at the electric vehicle end; preprocessing the electrochemical parameters and the charging noise to obtain input characteristics; inputting the input features into a hybrid neural network model, and predicting to obtain the total voltage of the power battery end at a future moment; wherein the hybrid neural network model comprises an input layer, a Transform layer, a convolutional neural network CNN, a long short-term memory network LSTM layer and an output layer. The total voltage of the battery end in the charging process can be accurately predicted, and charging safety is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of electrical digital processing technology, and in particular to a power battery terminal voltage prediction method and a charging pile based on multimodal data fusion. Background Art

[0002] With the transformation of the global energy structure and growing awareness of environmental protection, the new energy vehicle industry has experienced rapid development. As a key component of new energy vehicles, electric vehicles (EVs) have steadily increased their share of the global market. As a core component of EVs, the performance and condition of power batteries directly impact their operating efficiency and reliability. As a key performance indicator of battery status, total battery voltage reflects the battery's real-time condition and health.

[0003] Abnormal battery voltage can lead to various battery failures, making long-term prediction of total battery voltage crucial for monitoring the health of electric vehicles and ensuring their safe operation. However, to date, most research on total battery voltage prediction has focused on the operation of electric vehicles. However, numerous studies have shown that a significant proportion of vehicles experience failures or even fires during charging.

[0004] Therefore, the traditional prediction method may not be able to fully cover the battery voltage changes of electric vehicles under different working conditions. In view of this, the present invention is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a power battery terminal voltage prediction method and charging pile based on multimodal data fusion, so as to accurately predict the total battery terminal voltage during the charging process and ensure charging safety.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for predicting power battery terminal voltage based on multimodal data fusion, comprising: During the charging process of the electric vehicle, the electrochemical parameters of the battery are collected in real time through the charging pile; and the charging noise of the battery is collected through the acoustic sensor deployed at the electric vehicle end; Preprocessing the electrochemical parameters and charging noise to obtain input features; Inputting the input features into the hybrid neural network model to predict the total voltage of the power battery terminal at a future time; The hybrid neural network model includes an input layer, a Transformer layer, a convolutional neural network (CNN), a long short-term memory (LSTM) network, and an output layer.

[0007] In a second aspect, the present invention provides a charging pile, comprising: Charging pile body; at least one processor, and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors to enable at least one of the processors to execute the above-mentioned power battery terminal voltage prediction method based on multimodal data fusion.

[0008] Compared with the prior art, the present invention has the following beneficial effects: Traditional battery voltage prediction methods mostly target data collected from electric vehicles, lacking a grasp of actual vehicle charging data collected from charging piles, making it impossible for charging piles to provide voltage prediction capabilities. This paper proposes a new voltage prediction method based on acoustic signals and actual vehicle electrochemical parameters collected from charging piles to address these shortcomings. By using a hybrid neural network model, this method can predict battery voltage with higher accuracy than existing methods. Furthermore, because voltage prediction is based on acoustic signals and charging pile data, charging piles can provide timely warnings of battery anomalies, helping to maintain battery system safety and prevent safety risks such as failures and fires. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0010] Figure 1 This is a flow chart of a method for predicting terminal voltage of a power battery based on multimodal data fusion provided by an embodiment of the present invention; Figure 2 Schematic diagram of a hybrid neural network model provided by an embodiment of the present invention; Figure 3 This is a flow chart of another method for predicting power battery terminal voltage based on multimodal data fusion provided by an embodiment of the present invention; Figure 4 It is a structural schematic diagram of the charging pile provided by the present invention. DETAILED DESCRIPTION

[0011] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0012] Example 1 This embodiment is applicable to the following scenarios: by collecting the charging noise of the electric vehicle battery on the vehicle side and the battery charging data on the charging pile side, the total voltage of the power battery side is predicted, and abnormal voltage of the power battery is identified in a timely manner. Corresponding measures are taken for different situations to extend battery life and reduce maintenance costs.

[0013] Figure 1 This is a flow chart of a method for predicting the terminal voltage of a power battery based on multimodal data fusion provided by an embodiment of the present invention. This method is executed by a charging pile. Figure 1 , specifically including the following operations: S110. During the charging process of the electric vehicle, the electrochemical parameters of the battery are collected in real time through the charging pile; and the charging noise of the battery is collected through the acoustic sensor deployed at the electric vehicle end.

[0014] The electric vehicle is connected to the charging port of the charging pile. According to the transmission protocol of the charging port, the battery management system sends the battery's electrochemical parameters to the charging pile in real time, such as the remaining capacity (State Of Charge, abbreviated as SOC), total voltage, total current, temperature, maximum cell voltage, and maximum cell temperature.

[0015] In addition, the acoustic sensors deployed on the electric vehicle side collect the battery charging noise in real time and send it to the charging pile synchronously.

[0016] S120 , preprocessing the electrochemical parameters and charging noise to obtain input features.

[0017] The purpose of preprocessing is to unify the dimension of parameters, align the time series, and screen important parameters to meet the input requirements of the hybrid neural network model.

[0018] S130: Input the input features into the hybrid neural network model to predict the total voltage of the power battery terminal at a future moment.

[0019] See also Figure 2The hybrid neural network model in this embodiment includes an input layer, a Transformer layer, a convolutional neural network (CNN), a long short-term memory (LSTM) layer, and an output layer. By combining the global temporal modeling capabilities of the Transformer, the local feature extraction capabilities of the CNN, and the long-term dependency modeling capabilities of the LSTM, efficient processing of multimodal input data is achieved. The Transformer layer includes a self-attention mechanism layer, a position encoding layer, and a feedforward neural network; the CNN includes a convolution kernel and an activation function; and the LSTM includes a forget gate, an input gate, a memory unit, and an output gate.

[0020] For example, the preprocessed input features are fed into the input layer of the hybrid neural network model. After being processed sequentially by the Transformer layer, CNN, and LSTM, the output layer outputs the total voltage of the power battery terminal at a future moment. The future moment here can be a single moment or multiple consecutive moments in the future.

[0021] Traditional battery voltage prediction methods mostly target data collected from electric vehicles, lacking a grasp of actual vehicle charging data collected from charging piles, making it impossible for charging piles to provide voltage prediction capabilities. This paper proposes a new voltage prediction method based on acoustic signals and actual vehicle electrochemical parameters collected from charging piles to address these shortcomings. By using a hybrid neural network model, this method can predict battery voltage with higher accuracy than existing methods. Furthermore, because voltage prediction is based on acoustic signals and charging pile data, charging piles can provide timely warnings of battery anomalies, helping to maintain battery system safety and prevent safety risks such as failures and fires.

[0022] Example 2 This embodiment is optimized based on the above embodiment. Figure 3 , the method provided in this embodiment includes the following operations: S210. During the charging process of the electric vehicle, the electrochemical parameters of the battery are collected in real time through the charging pile; and the charging noise of the battery is collected through the acoustic sensor deployed at the electric vehicle end.

[0023] The charging noise undergoes preprocessing, Fourier transform, Mel filter bank mapping, logarithmic operation, and discrete cosine transform to obtain acoustic features. Specifically, the following steps are performed: ① Preprocessing: including framing, windowing, and pre-emphasis; ② Fourier transform: converting the time domain signal to the frequency domain; ③ Mel filter bank mapping: mapping the spectrum to the Mel scale; ④ Logarithmic operation: taking the logarithm of the Mel spectrum to compress the dynamic range; ⑤ Discrete Cosine Transform (DCT): extracting MFCC (Mel-Frequency Cepstral Coefficients). The MFCC feature dimension is 13 (i.e., C0-C13).

[0024] Finally, the acoustic features are downsampled to align the acoustic features with the electrochemical parameters time series.

[0025] S220 . Perform correlation analysis on the parameters, charging noise, and the total voltage of the power battery terminal, and select target features based on the results of the correlation analysis.

[0026] Optionally, the feature dimension of MFCC is 13, and the dimension of the electrochemical parameters of the electric vehicle collected by the charging pile is 6, totaling 19 types of data. However, some data is not highly correlated with the total voltage, and the charging curve of the electric vehicle battery is not linear, so the correlation of parameters at different stages is different. Therefore, at different charging stages, correlation analysis is performed on the parameters and charging noise with the total voltage at the power battery end to obtain the results of the correlation analysis at different charging stages and select target features for different charging stages. For example, a correlation analysis is performed with a 20% increase in SOC as a charging stage, and input features with a high correlation with the total voltage are selected for prediction.

[0027] Specifically, in each charging stage, the Pearson correlation coefficient is calculated for any one of the parameters and the charging noise and the total voltage at the power battery terminal to obtain a first correlation; a second correlation between any one of the parameters and the charging noise and the total voltage at the power battery terminal is determined; and the first correlation and the second correlation are weighted and summed to obtain a correlation analysis result.

[0028] The calculation formula of Pearson correlation coefficient (PCC) is as follows: ; in, is the Pearson correlation coefficient, the first degree of correlation; and The variables are and No. observations; and They are and The sample mean of ; ∑ represents the sum of all observations.

[0029] At the same moment in each charging stage, any one of the parameters and charging noise is selected as x, and the total voltage at that moment is selected as y. The PCC is calculated as the PCC of the selected parameter and the total voltage. After calculating the correlation coefficients for each charging stage, an adaptive feature selection algorithm based on the charging stage is developed to dynamically select the most relevant features according to the charging stage. The formula is as follows: ; in, It is the result of the weighted summation of the correlation analysis, indicating the importance of a selected option to the total voltage prediction; and is the weight coefficient, which is used to adjust the proportion of the two contributions (where ). and The degree of influence of parameters on total voltage prediction can be determined through controlled variable experiments. is the Pearson correlation coefficient, the first degree of correlation; The physical correlation reflects the physical relationship between the characteristic and voltage, i.e., the second correlation. This yields the correlation analysis results for any selected option at different charging stages. The second correlation is determined based on battery characteristics and charging principles. For example, battery engineers score the physical relationship between the parameter and voltage. For example, on a correlation scale of 0 to 1, the PC for SOC is 0.9 (high correlation), and the PC for temperature is 0.6 (moderate correlation).

[0030] S230: Normalize and window the target features to obtain input features.

[0031] Optionally, normalize the target features. In deep learning, especially when it comes to neural networks, the scale of the input data significantly impacts the model's training efficiency and ultimate performance. To address this, a MinMaxScaler is used to scale all features (i.e., the features with high correlation with the total voltage, selected based on the correlation analysis at each charging stage) to a uniform interval, such as [0, 1].

[0032] Then, the normalized target features are processed by sliding windows, that is, the normalized features are placed in the prediction window and training window. During the sliding window processing, the step size of the sliding window is adjusted according to the rate of change of the total voltage at the power battery end. In order to adapt to the model's processing requirements for time series data, the normalized data needs to be reorganized into a series of time windows, and a dynamic adjustment strategy for the sliding window is introduced. Based on the rate of change of the total voltage, the sliding window step size is adjusted. Automatically adjust the sliding window step size. When the voltage changes dramatically during the charging process, reduce the sliding window step size to enable the model to capture the voltage change trend more promptly. When the total voltage change rate is relatively stable, increase the sliding window step size to make full use of more historical data for prediction.

[0033] S240: Input the input features into the hybrid neural network model to predict the total voltage of the power battery terminal at a future moment.

[0034] Before using the hybrid neural network model to predict the future total voltage, the hybrid neural network model needs to be trained. This embodiment adopts a curriculum learning training method. The core idea is to let the model learn simple samples first, and then gradually learn complex samples, thereby improving model performance.

[0035] First, training samples are collected and divided into simple samples, medium samples and complex samples according to the constant current charging stage, constant voltage charging stage and trickle charging stage; in the first stage, the simple samples are used to train the hybrid neural network model; in the second stage, the simple samples and medium samples are used to train the hybrid neural network model; in the third stage, the simple samples, medium samples and complex samples are used to train the hybrid neural network model.

[0036] Specifically, the difficulty of training samples is divided into the following categories: ① Simple samples are collected during the constant-current charging phase (0%-20% SOC), where the current is constant and the voltage gradually increases; ② Medium samples are collected during the constant-voltage charging phase (20%-80% SOC), where the voltage is constant and the current gradually decreases; and ③ Complex samples are collected during the trickle charging phase (80%-100% SOC), where both the current and voltage decrease. Based on this difficulty classification, training is carried out in stages. The first stage is simple sample training: the model is trained using simple samples to gain a preliminary understanding of charging patterns. During this stage, the model can quickly learn the basic patterns of the charging process, namely the rising voltage pattern when the current is constant, laying the foundation for understanding the entire charging process. The second stage is medium sample training, where medium samples are added to continue training to improve the model's adaptability to complex scenarios. During the constant-voltage charging phase, the current gradually decreases, and the model needs to learn the stable relationship between current changes and voltage changes, thus enhancing its understanding of the dynamic changes in the charging process. The third stage is complex sample fine-tuning. The complex sample fine-tuning model is added to further improve the model performance. During the trickle charging stage, both the current and voltage decrease. The model needs to learn the complex current and voltage change patterns in this stage, which improves the model's generalization ability and adaptability to complex charging scenarios, ensuring that it can accurately predict the voltage at different charging stages.

[0037] After each training round, the performance of the hybrid neural network model is evaluated using the validation set, calculating the mean relative error (MRE) and root mean square error (RMSE) on the validation set. Based on the performance on the validation set, the model's hyperparameters are adjusted, such as the learning rate, regularization coefficient, number of network layers, and number of neurons.

[0038] After the model is trained and verified, a hybrid neural network model is used to predict total voltage. Specifically, the input features of the input layer are multimodal features filtered through correlation analysis (different input features correspond to different charging stages). The input feature dimensions are [batch size, time step, number of features]. For example, [16, 300, 19] represents 16 sets of input features (or samples), each set of input features includes 300 time steps and 19 features (this example is used for the following layers).

[0039] The Transformer layer captures the global temporal dependencies of the input sequence. Its architecture includes: ① a self-attention mechanism that computes global dependency weights for each time step in the input sequence; ② a positional encoding that adds positional information to the input sequence, preserving temporal properties; and ③ a feedforward neural network that performs a nonlinear transformation on the output of the self-attention mechanism. Its input shape is [16, 300, 19], and its output shape is [16, 300, 128].

[0040] The CNN layer extracts local mutation features from the input sequence. Its architecture includes: ① Convolution kernel size: 5; ② Number of convolution kernels: 64; ③ Activation function: ReLU. Its input shape is [16, 300, 128], and its output shape is [16, 300, 64].

[0041] The LSTM layer models long-term dependencies in the input sequence. Its structure includes: ① Forget gate: determines which information is discarded from the memory cell; ② Input gate: determines which new information is stored in the memory cell; ③ Memory cell: combines the outputs of the forget and input gates to update the memory cell value; ④ Output gate: determines which information is output from the memory cell. Its input shape is [16, 300, 64], and its output shape is [16, 64].

[0042] The output layer predicts the voltage value at the next moment. Its input shape is [16, 64], and its output shape is [16, 60]. This means that for each set of input features, the total voltage value for the next 60 time points is predicted.

[0043] S250: Compare the predicted total voltage at the power battery terminal with the normal total voltage; and take early warning and / or charging pile protection measures based on the comparison result.

[0044] When an electric vehicle is charging at a charging station, if the predicted total voltage at any future moment exceeds a certain threshold compared to the normal total voltage (i.e., the total voltage during normal charging), a warning signal will be issued immediately. For example, if the predicted total voltage exceeds 5% of the normal total voltage, a level 1 warning will be issued; if it exceeds 10%, a level 2 warning will be issued, requiring notification to safety personnel; and if it exceeds 20%, a level 3 warning will be issued, and the charging station power outage protection will be activated.

[0045] Example 3 like Figure 4 As shown, this embodiment provides a charging pile, comprising: a charging pile body, at least one processor; and a memory communicatively connected to the at least one processor. The charging pile body is used to charge electric vehicles, collect battery electrochemical parameters and charging noise in real time, and transmit them to the processor.

[0046] The memory stores instructions executable by at least one of the processors, and the instructions are executed by the at least one processor to enable the at least one processor to perform the above method. The at least one processor in the charging pile is capable of performing the above method, thereby having at least the same advantages as the above method.

[0047] Optionally, the charging pile also includes interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses and can be installed on a common motherboard or in other ways as needed. The processor can process instructions executed in the charging pile, including instructions stored in or on the memory to display graphical information of the GUI (Graphical User Interface) on an external input / output device (such as a display device coupled to the interface). In other embodiments, if necessary, multiple processors can be used with multiple memories, and / or multiple buses can be used with multiple memories. Similarly, multiple electronic devices can be connected (for example, as a server array, a group of blade servers, or a multi-processor system), and each device provides some necessary operations. Figure 4 A processor 301 is taken as an example.

[0048] Memory 302, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the power battery terminal voltage prediction method based on multimodal data fusion in the embodiments of the present invention. Processor 301 executes the software programs, instructions, and modules stored in memory 302 to perform various functional applications and data processing of the device, thereby implementing the aforementioned power battery terminal voltage prediction method based on multimodal data fusion.

[0049] The memory 302 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data created based on the use of the terminal, etc. Furthermore, the memory 302 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 302 may further include a memory remotely located relative to the processor 301, and these remote memories may be connected to the device via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0050] The charging pile may further include an input device 303 and an output device 304. The processor 301, the memory 302, the input device 303 and the output device 304 may be connected via a bus or other means, with the bus connection being used as an example in the figure.

[0051] The input device 303 can receive input digital or character information, and the output device 304 can include a display device, an auxiliary lighting device (e.g., an LED), a tactile feedback device (e.g., a vibration motor), etc. The display device may include, but is not limited to, a liquid crystal display (LCD), a light-emitting diode (LED) display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0052] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved. This is not limited herein.

[0053] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A power battery terminal voltage prediction method based on multimodal data fusion, characterized in that: The method comprises: During the charging process of the electric vehicle, the electrochemical parameters of the battery are collected in real time through the charging pile; and the charging noise of the battery is collected through the acoustic sensor deployed at the electric vehicle end; Preprocessing the electrochemical parameters and charging noise to obtain input features; Inputting the input features into the hybrid neural network model to predict the total voltage of the power battery terminal at a future time; The hybrid neural network model includes an input layer, a Transformer layer, a convolutional neural network (CNN), a long short-term memory (LSTM) network, and an output layer.

2. The method according to claim 1, characterized in that The electrochemical parameters and charging noise are preprocessed to obtain input features, including: Conduct correlation analysis on parameters, charging noise and total voltage of power battery terminals; Select target features based on the results of correlation analysis; The target features are normalized and windowed to obtain input features.

3. The method according to claim 2, characterized in that Correlation analysis of parameters, charging noise and total voltage of power battery terminals is performed, including: Calculating a Pearson correlation coefficient between any one of the parameter and the charging noise and the total voltage of the power battery terminal to obtain a first correlation; determining a second correlation between any one of the parameter and the charging noise and the total voltage at the power battery terminal; The first correlation and the second correlation are weighted and summed to obtain a result of correlation analysis.

4. The method according to claim 3, characterized in that The target features are normalized and windowed to obtain input features, including: Normalize the target features; The normalized target features are subjected to sliding window processing, and during the sliding window processing, the step size of the sliding window is adjusted according to the change rate of the total voltage at the power battery terminal.

5. The method according to claim 4, characterized in that The electrochemical parameters and charging noise are preprocessed to obtain input features, and the following steps are also included: The charging noise is sequentially preprocessed, Fourier transformed, mapped using a Mel filter bank, subjected to logarithmic operation, and subjected to discrete cosine transform to obtain acoustic features. The acoustic signature is downsampled to align the acoustic signature with the electrochemical parameter time series.

6. The method according to claim 2, characterized in that Correlation analysis of parameters, charging noise and total voltage of power battery terminals is performed, including: At different charging stages, correlation analysis is performed on the parameters, charging noise and the total voltage of the power battery terminal to obtain the results of the correlation analysis at different charging stages and select the target features at different charging stages.

7. The method according to any one of claims 1 to 6, characterized in that The Transformer layer includes a self-attention mechanism layer, a position encoding layer and a feedforward neural network; The CNN includes a convolution kernel and an activation function; The LSTM includes a forget gate, an input gate, a memory unit and an output gate.

8. The method according to any one of claims 1 to 6, characterized in that During the charging process of electric vehicles, before the real-time collection of battery electrochemical parameters through the charging pile, it also includes: Collect training samples and divide them into simple samples, medium samples and complex samples according to the constant current charging stage, constant voltage charging stage and trickle charging stage; In the first stage, the hybrid neural network model is trained using the simple samples; In the second stage, the hybrid neural network model is trained using simple samples and medium samples; In the third stage, simple samples, medium samples and complex samples are used to train the hybrid neural network model.

9. The method according to any one of claims 1 to 6, characterized in that After predicting the total voltage of the power battery at the future time, it also includes: Compare the predicted total voltage at the power battery terminal with the normal total voltage; Early warning and / or charging pile protection measures are taken based on the comparison results.

10. A charging pile, characterized in that: include: Charging pile body; at least one processor, and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors so that at least one of the processors can execute the power battery terminal voltage prediction method based on multimodal data fusion according to any one of claims 1 to 9.

Images