Method and system for rapidly estimating state of health of battery based on frequency domain analysis

Abstract

The invention discloses a method and a system for rapidly estimating the state of health of a battery based on frequency domain analysis. The method comprises the following steps: acquiring lithium battery operation data in a plurality of discharge cycles; inputting the lithium battery operation data into a frequency domain prediction neural network to perform battery health state prediction so as to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by obtaining historical lithium battery operation data and corresponding lithium battery health state data as a sample set to train the neural network, and the neural network comprises a data processing layer, a Fourier transform layer, a filter, a complex value linear layer, a zero filling layer and an inverse Fourier transform layer which are connected in sequence; and outputting the prediction result. By implementing the method provided by the invention, effective combination of frequency domain analysis and low resource consumption can be realized, and the SOH prediction efficiency and accuracy of the battery management system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery health state management, and more specifically to a method and system for quickly estimating the health state of a battery based on frequency domain analysis. Background Art

[0002] As global energy shortages and environmental pollution become increasingly serious, lithium batteries have become an ideal power source for mobile devices, electric vehicles and other fields due to their environmental protection, high energy density, long life, low self-discharge, no memory effect, fast charging and wide temperature operating range. They have shown broad application prospects and strong market competitiveness. However, lithium batteries will age with the increase of usage time, which may cause safety hazards such as thermal runaway and fire. Therefore, it is particularly important to monitor the health of the battery in advance to identify potential risks and take necessary safety measures. SOH (State of Health) refers to the ratio of the performance of a battery at a specific point in time to the performance of a new battery. SOH prediction is not only the core function of the battery management system, but also an important link in promoting the advancement of battery technology, ensuring application safety and improving economic benefits.

[0003] Most of the existing SOH prediction methods focus on time domain analysis, such as the SD-crossformer algorithm, the whale optimization algorithm, and the traditional Gaussian process regression, long short-term memory network, etc. Although these methods are effective in some applications, they often lead to low prediction accuracy because they cannot fully capture the complex relationships in long time series. However, the battery has obvious periodic characteristics during use, that is, each process from charging to fully discharging constitutes a charge and discharge cycle. In standardized tests or actual use, the capacity and health of the battery will gradually decay, and this decay process often follows certain electrochemical periodic laws. From the SOH data, the change also shows periodic fluctuations. Therefore, it is possible to consider analyzing SOH from the frequency domain perspective, but the SOH prediction technology based on frequency domain analysis has not yet been widely used.

[0004] Therefore, it is necessary to design a new method to effectively combine frequency domain analysis with low resource consumption to improve the efficiency and accuracy of SOH prediction in battery management systems. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a method and system for quickly estimating the health status of a battery based on frequency domain analysis.

[0006] To achieve the above object, the present invention adopts the following technical solution: A method for rapidly estimating battery health status based on frequency domain analysis, comprising: Obtain lithium battery operation data over multiple discharge cycles; The lithium battery operation data is input into a frequency domain prediction neural network to predict the battery health state to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by obtaining historical lithium battery operation data and corresponding lithium battery health state data as a sample set to train a neural network, and the neural network includes a data processing layer, a Fourier transform layer, a filter, a complex linear layer, a zero filling layer and an inverse Fourier transform layer connected in sequence; The prediction result is output.

[0007] A further technical solution is: the lithium battery operation data is input into a frequency domain prediction neural network to predict the battery health status to obtain a prediction result, including: Inputting the lithium battery operation data into a frequency domain prediction neural network, and performing data processing on the lithium battery operation data to obtain a processing result; Using the Fourier transform layer to convert the processing result into a frequency domain representation to obtain a first transformation result; Using the filter to filter the first transformation result to obtain a filtering result; Performing frequency domain interpolation on the filtering result by using the complex-valued linear layer to obtain interpolated frequency domain data; Using the zero-filling layer to perform a zero-filling operation on the interpolated frequency domain data to obtain a zero-filling result; Converting the zero-filled result from the frequency domain back to the time domain using the inverse Fourier transform layer to obtain a second transform result; Energy compensation is performed on the second transformation result to obtain a prediction result.

[0008] A further technical solution is: inputting the lithium battery operation data into a frequency domain prediction neural network, and performing data processing on the lithium battery operation data to obtain a processing result, including: The lithium battery operation data is input into a frequency domain prediction neural network, and reversible instance normalization is performed on the lithium battery operation data to obtain a processing result.

[0009] A further technical solution is: performing energy compensation on the second transformation result to obtain a prediction result includes: The second transformation result is multiplied by the coefficient to obtain a prediction result.

[0010] A further technical solution is: using the complex-valued linear layer to perform frequency domain interpolation on the filtering result to obtain interpolated frequency domain data, including: The complex-valued linear layer is used to perform frequency domain interpolation on the filtering result through linear matrix multiplication and matrix addition to obtain interpolated frequency domain data.

[0011] A further technical solution is: using the zero-filling layer to perform a zero-filling operation on the interpolated frequency domain data to obtain a zero-filling result, including: The zero-filling layer is used to add zero values ​​to both ends of the interpolated frequency domain data to obtain a zero-filling result.

[0012] The further technical solution is: the frequency domain prediction neural network is obtained by obtaining historical lithium battery operation data and corresponding lithium battery health status data as a sample set to train the neural network, including: Obtain historical lithium battery operation data and corresponding lithium battery health status data to obtain a sample set; Build a neural network; The neural network is trained based on the historical lithium battery operation data in the sample set, and the parameters of the trained neural network are adjusted in combination with the corresponding lithium battery health status data to obtain a frequency domain prediction neural network.

[0013] A further technical solution is that the frequency domain prediction neural network is trained using a gradient descent mode.

[0014] Its further technical solution is: the frequency domain prediction neural network uses the root mean square error as an evaluation index of network performance.

[0015] The present invention also provides a battery health status rapid estimation system based on frequency domain analysis, comprising: A data acquisition unit, used to acquire lithium battery operation data within multiple discharge cycles; A prediction unit, used for inputting the lithium battery operation data into a frequency domain prediction neural network to perform battery health state prediction to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by training a neural network by acquiring historical lithium battery operation data and corresponding lithium battery health state data as a sample set, and the neural network includes a data processing layer, a Fourier transform layer, a filter, a complex-valued linear layer, a zero-filling layer, and an inverse Fourier transform layer connected in sequence; An output unit is used to output the prediction result.

[0016] The beneficial effects of the present invention compared with the prior art are as follows: the present invention obtains lithium battery operation data within multiple discharge cycles; the collected battery operation data is input into a frequency domain prediction neural network, which includes a Fourier transform layer, which can convert time domain data into frequency domain data; the frequency domain data is processed by a filter to extract important features and reduce redundant information; the design of the neural network includes a complex-valued linear layer and an inverse Fourier transform layer, so that the frequency domain information can be effectively converted and used to accurately predict SOH; compared with complex traditional models, the number of parameters of the frequency domain prediction neural network is significantly reduced, which is only one thousandth of the traditional model, and it can run efficiently on resource-constrained edge devices, thereby improving prediction efficiency and accuracy.

[0017] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0019] Figure 1 A schematic diagram of an application scenario of a method for rapidly estimating a battery health state based on frequency domain analysis provided by an embodiment of the present invention; Figure 2 A schematic diagram of a flow chart of a method for rapidly estimating a battery health state based on frequency domain analysis provided by an embodiment of the present invention; Figure 3 A schematic diagram of a sub-process of a method for rapidly estimating a battery health state based on frequency domain analysis provided by an embodiment of the present invention; Figure 4 A schematic block diagram of a system for quickly estimating a battery health state based on frequency domain analysis provided by an embodiment of the present invention; Figure 5 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0022] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0023] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0024] See also Figure 1 and Figure 2 , Figure 1 A schematic diagram of an application scenario of a method for rapidly estimating a battery health state based on frequency domain analysis provided in an embodiment of the present invention. Figure 2 A schematic flow chart of a method for rapidly estimating a battery health state based on frequency domain analysis provided by an embodiment of the present invention. The method for rapidly estimating a battery health state based on frequency domain analysis is applied to a server. The server interacts with a terminal for data exchange.

[0025] Figure 2 FIG. 1 is a flow chart of a method for rapidly estimating a battery health status based on frequency domain analysis provided by an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S130.

[0026] S110, obtaining lithium battery operation data in multiple discharge cycles.

[0027] In this embodiment, the lithium battery operation data includes four characteristics, including capacity, current, voltage and charge and discharge cycle.

[0028] Battery capacity refers to the total amount of electricity a battery can provide under specific discharge conditions. It is usually measured in ampere-hours (Ah), which represents the product of current intensity (amperes, A) and time (hours, h). Battery capacity is a key parameter for measuring battery performance, which is directly related to the endurance of the device. As the number of uses increases or time passes, the actual available capacity of the battery will gradually decrease, which reflects the degree of battery aging. The capacity can be determined by measuring the total charge that the battery can provide under a constant load until the termination voltage is reached after it is fully charged.

[0029] Current refers to the rate of charge flowing through a point in a circuit, measured in amperes (A). For batteries, it refers to the amount of current that passes through the battery during the charge and discharge process. The magnitude of the current affects the speed and efficiency of the battery's charge and discharge. Too much or too little current may have a negative impact on the battery's life. Use current sensors to monitor the input and output current values ​​of the battery in real time.

[0030] Voltage is the potential difference between the two ends of a power source, measured in volts (V). For batteries, this refers to its open circuit voltage or operating voltage. The voltage level not only determines whether the battery can provide power properly, but is also an important indicator of the battery's condition. Abnormal voltage may be a sign of battery failure or aging. Use a voltmeter or other built-in detection mechanism to regularly record the battery voltage.

[0031] A charge and discharge cycle refers to a complete charge and discharge process, from the time the battery is charged until it is fully charged again, and then discharged from the fully charged state until it needs to be recharged. Each charge and discharge cycle will cause some wear and tear on the internal structure of the battery, so the number of charge and discharge cycles is an important factor in evaluating the battery life. The number of charge and discharge times the battery has experienced is automatically tracked through a counter or software algorithm.

[0032] S120. Input the lithium battery operation data into a frequency domain prediction neural network to perform battery health status prediction to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by obtaining historical lithium battery operation data and corresponding lithium battery health status data as a sample set to train the neural network, and the neural network includes a data processing layer, a Fourier transform layer, a filter, a complex-valued linear layer, a zero-filling layer, and an inverse Fourier transform layer connected in sequence.

[0033] In this embodiment, the prediction result refers to the health status prediction result of the lithium battery.

[0034] In one embodiment, see Figure 3 , the above-mentioned step S120 may include steps S121~S127.

[0035] S121, inputting the lithium battery operation data into a frequency domain prediction neural network, and performing data processing on the lithium battery operation data to obtain a processing result.

[0036] In this embodiment, the processing result refers to the result obtained after reversible instance normalization is performed on the lithium battery operation data.

[0037] Specifically, the lithium battery operation data is input into a frequency domain prediction neural network, and reversible instance normalization is performed on the lithium battery operation data to obtain a processing result.

[0038] RevIN (Reversible Instance Normalization) is a data pre-processing and post-processing technique designed for time series forecasting tasks. It solves the problems existing in traditional multivariate time series data normalization methods. Traditional methods such as Min-Max Normalization and Z-Score Normalization apply uniform statistics (such as minimum, maximum, mean, and standard deviation) to the entire dataset, which may lead to inconsistencies when processing data during training and forecasting phases, especially in sliding window settings. To solve this problem, RevIN independently calculates and applies normalization parameters within each time step, which can preserve the local statistical characteristics of each sample and restore the original values ​​in the post-processing phase, thereby improving the prediction accuracy and model stability.

[0039] By independently calculating and applying normalization parameters at each time step, RevIN preserves the local statistical properties of each sample. Since the original values ​​can be accurately restored, the accuracy of model predictions is improved. Compared with traditional global normalization methods, RevIN provides a more stable model training process, especially under the sliding window setting. It is suitable for various types of time series data, especially those with complex dynamic characteristics, such as lithium battery operation data.

[0040] Specifically, reversible instance normalization includes the following steps: For each sample in the sliding window, calculate its own mean and standard deviation (or minimum and maximum). Use these calculated statistics to linearly normalize the samples to ensure that all features are on the same scale. The normalization formula can be expressed as , where R is the normalized result, D is the parameter data, Dmax is the maximum value, and Dmin is the minimum value of the corresponding parameter.

[0041] The above normalized data is used as input to train the frequency domain prediction neural network. This step helps the model learn the relationship between different frequency components and capture key patterns that may affect battery performance. After the prediction is completed, the predicted results are denormalized using the previously saved statistical information (i.e., the mean, standard deviation, or extreme value of each sliding window) to convert the predicted values ​​back to the original scale. This process ensures the authenticity and usability of the final output results.

[0042] By introducing RevIN technology, we can not only effectively solve the challenges brought by time series data normalization, but also significantly improve the performance of the model while maintaining the original characteristics of the data. This method is particularly suitable for analyzing and predicting the behavioral changes of complex systems such as lithium batteries, providing strong support for optimizing battery management and extending battery life.

[0043] S122. Use the Fourier transform layer to convert the processing result into a frequency domain representation to obtain a first transformation result.

[0044] In this embodiment, the first transformation result refers to a result obtained by converting the processing result into a frequency domain identifier.

[0045] Specifically, in order to gain a deeper understanding of the essential characteristics of the signal and to process the signal effectively, Fourier transform is used to convert the original signal from the time domain to the frequency domain. The core idea of ​​Fourier transform is to decompose a complex wave pattern, whether it is the change of voltage, current or other physical quantity in the time series, into a superposition of a series of simple sine waves and cosine waves. These basic frequency components not only reveal the internal structure of the signal, but also provide a new perspective for analyzing the signal characteristics.

[0046] Through Fourier transform, a seemingly random or complex time domain signal can be parsed into its basic frequency components. Each such frequency component is characterized by its amplitude and phase, which respectively describe the intensity of the frequency component and its position on the time axis. This representation makes it possible to understand and manipulate signals more intuitively, especially in application scenarios such as filtering, compression, and feature extraction.

[0047] According to the mathematical expression of Fourier transform: ; In this formula, F represents the signal set after all frequency components are synthesized in the complex frequency domain; |X(f)| refers to the amplitude at each specific frequency f, reflecting the contribution of this frequency component to the total signal; θ represents the phase angle of the corresponding frequency component, which is used to describe the relative time delay or lead between this component and other frequency components.

[0048] By breaking down a signal into its fundamental frequency components, the Fourier transform helps us identify important patterns and periodic behaviors that may not be apparent in the time domain. For certain types of signal processing tasks, such as filter design, it is often simpler and more efficient to operate directly in the frequency domain rather than in the time domain. Many signal processing algorithms, such as the Fast Fourier Transform (FFT), are able to be computed efficiently on computers, speeding up data analysis. The frequency domain view can provide deep insights into system response, noise characteristics, and other key properties, which is very valuable for engineering design and technology development.

[0049] S123: Utilize the filter to filter the first transformation result to obtain a filtering result.

[0050] In this embodiment, the filtering result refers to a result obtained by performing high-frequency filtering on the first transformation result.

[0051] Specifically, the frequency domain is analyzed and a high-frequency filter is used to filter out some of the interference. Frequency domain analysis is a method of converting a signal in the time domain to the frequency domain using mathematical tools such as Fourier transform. This is done to better understand the components of the signal, especially to identify different frequency components and their relative strengths. Compared to processing the signal directly in the time domain, frequency domain analysis provides a more intuitive way to observe and process specific frequency components in the signal.

[0052] The main advantages of frequency domain analysis include but are not limited to: Feature extraction: It is easier to identify periodic and non-periodic components in the signal.

[0053] Noise separation: helps distinguish useful signal information from useless or harmful noise.

[0054] System response: Being able to evaluate the frequency response characteristics of a system is important for design and debugging.

[0055] After completing the frequency domain analysis, a high-frequency filter is then applied to further process the signal. A high-frequency filter, strictly speaking, is a low-pass filter because it is used to "allow low-frequency signals to pass through while attenuating signals outside this range." This type of filter is designed to retain the low-frequency portion of the signal while reducing or eliminating the effects of the high-frequency components, which is very necessary in many cases, such as when it is necessary to remove high-frequency interference or prepare the signal for subsequent processing.

[0056] Features and benefits of high frequency filters include: A cut-off frequency point can be precisely set according to actual needs, so that all frequency components below this frequency can pass through intact, while components above this frequency are weakened to varying degrees.

[0057] Effectively reduce high-frequency noise introduced by the external environment or other devices and improve the quality of the target signal.

[0058] Users can flexibly adjust the parameters of the high-frequency filter, such as cutoff frequency, slope, etc., according to the specific frequency domain analysis results to adapt to different application scenarios.

[0059] Prevent high frequency components from causing adverse effects on subsequent processing stages, such as overheating or distortion issues.

[0060] Specifically, in step S123 of this example, through in-depth analysis of the frequency domain, it can be determined which components within the frequency range are unnecessary, which may be because they represent unnecessary noise or information that does not contribute to the final output. Then, using a high-frequency filter (assuming a low-pass filter LPF), an appropriate cutoff frequency can be set to ensure that only those low-frequency signals of interest can pass smoothly, while the remaining high-frequency components are effectively filtered out.

[0061] This not only improves the clarity and quality of the signal, but also makes it fully prepared for subsequent data processing or transmission. In addition, energy saving can be achieved in this way because after removing unnecessary high-frequency components, the energy consumption required in the subsequent processing process is reduced.

[0062] S124. Perform frequency domain interpolation on the filtering result using the complex-valued linear layer to obtain interpolated frequency domain data.

[0063] In this embodiment, the interpolated frequency domain data refers to the result obtained by interpolating the filtering result.

[0064] Specifically, the complex-valued linear layer is used to perform frequency domain interpolation on the filtering result through linear matrix multiplication and matrix addition to obtain interpolated frequency domain data.

[0065] The frequency-domain analyzed data is fed into a specially designed complex-valued linear layer. The key to this step is to predict the behavior of each discrete frequency component through frequency-domain interpolation and learn the amplitude scaling and phase shift characteristics on these frequency components.

[0066] Frequency domain interpolation is to estimate the unknown values ​​between certain discrete frequency points by mathematical methods after the information at these points is known. This method is particularly suitable for frequency domain signal processing because it can simplify the complex information extraction process in the time domain. Compared with direct operations in the time domain, the advantages of frequency domain processing are: It is easier to predict each discrete frequency component than to extract effective information directly from the time domain. For long time series dependencies, frequency domain processing can capture signal characteristics more accurately. It avoids the instability and noise amplification problems that may occur in time domain processing.

[0067] To achieve the above goals, a complex-valued linear layer is used. This layer is not only a sampling layer, but also responsible for learning how to scale the amplitude in the frequency domain and adjust the phase. Its working principle is based on a simple linear transformation formula: ; A is the weight matrix, which determines the degree of interaction between different frequency components; x is the input vector, which represents the frequency domain data within a certain time window; B is the bias vector, which is used to adjust the overall level of the output; y is the output vector. x is a vector within a lookback window.

[0068] In addition, it is worth noting that in the complex domain, the ratio of the length of the output sequence Y to the length of the input sequence X is equal to the ratio of the lookback window to the length of the output sequence in the time domain. This feature ensures that the time resolution before and after processing remains consistent, thereby maintaining the time characteristics of the original signal.

[0069] The neural network will automatically learn and optimize the weight matrix A and the bias vector B to better fit the actual data. The gradient descent method is used to update the parameters during the training process, and the root mean square error is selected as the performance evaluation indicator. The RMSE (Root Mean Squared Error) formula is , which can quantify the difference between the predicted value and the true value and measure the performance quality of the model.

[0070] The method of frequency domain interpolation using complex-valued linear layers has the following significant benefits: By converting the problem into frequency domain processing, the complex time domain operation requirements are reduced and the algorithm execution efficiency is improved. Since the frequency domain representation can better separate the signal components, the signal characteristics can be captured and predicted more accurately. It allows users to flexibly select the length of the output sequence according to the specific application scenario to meet the diverse data analysis requirements. The use of frequency domain characteristics helps to filter out unnecessary noise interference and enhances the model's ability to combat abnormal data. Compared with pure data-driven methods, frequency domain processing based on physical meaning makes the results easier to understand and interpret.

[0071] S125 . Perform a zero-filling operation on the interpolated frequency domain data using the zero-filling layer to obtain a zero-filling result.

[0072] In this embodiment, the zero-filling result refers to the result obtained after performing a zero-filling operation on the interpolated frequency domain data.

[0073] Specifically, the zero-filling layer is used to add zero values ​​to both ends of the interpolated frequency domain data to obtain a zero-filling result.

[0074] Zero padding is a commonly used technique in signal processing, especially in spectrum analysis. The basic idea is to add a certain number of zero-valued sample points at the end or both ends of the original signal. This practice does not change the content of the original signal, but will artificially increase the time length of the signal, thus affecting the subsequent frequency domain representation.

[0075] When performing a discrete Fourier transform or other form of spectral analysis on a time-domain signal of finite length, the resulting spectrum is usually discrete and fixed in intervals. If the original signal length is used directly for the transform, sufficiently fine spectral details may not be obtained. By adding extra zero values ​​to the signal, i.e., implementing zero padding, the signal can be effectively "stretched" so that the sample points on the spectrum become denser after a transform such as the DFT.

[0076] This more dense spectrum sampling brings higher frequency resolution, which means we can observe the spectral characteristics of the signal at a finer granularity. Although this does not really increase the frequency information content of the signal itself, because the original data has not changed, it does make the spectrum image look more continuous and smooth, which is convenient for visual analysis and further algorithm processing.

[0077] From a time domain perspective, a signal that has been zero-padding tends to show a smoother trend when it is inversely transformed back to the time domain. This is because zero-padding increases the number of data points used to reconstruct the signal and reduces the Gibbs phenomenon caused by insufficient data, that is, the oscillation effect that occurs near rapidly changing areas. Therefore, zero-padding helps generate a time domain waveform that is closer to the ideal state, especially when the signal needs to be accurately simulated or reconstructed.

[0078] Specifically, zero padding can be performed in the following ways: End padding: The most common way is to append several zero values ​​at the end of the signal sequence.

[0079] Padding at both ends: In addition to adding zero values ​​at the end, you can also choose to insert the same or different number of zero values ​​at the beginning of the signal sequence. This is sometimes called double-sided zero padding.

[0080] Internal padding: Although not common, in some special cases you may consider inserting zero values ​​at appropriate locations inside the signal, but this usually affects the true shape of the signal, so use it with caution.

[0081] S126. Use the inverse Fourier transform layer to convert the zero-filling result from the frequency domain back to the time domain to obtain a second transform result.

[0082] In this embodiment, the second transformation result refers to a result obtained by converting the zero-filling result from the frequency domain to the time domain.

[0083] The Inverse Fourier Transform (IFT) is a fundamental tool in signal processing and analysis that allows us to convert a function from the frequency domain back to the time domain. Specifically, if we have a signal X(f) represented in the complex frequency domain, we can restore it to the original time domain signal y(t) through the inverse Fourier transform.

[0084] For continuous-time signals, the mathematical expression for the inverse Fourier transform is as follows: ; represents the signal in the time domain; X(f) is the representation of the signal in the frequency domain, usually obtained by Fourier transform. t is the time variable. f is the frequency variable. j is the imaginary unit, . It is a complex exponential function, representing a sinusoidal waveform at different frequencies.

[0085] This integration operation means summing up all frequency components according to their corresponding amplitude and phase weights to reconstruct the original time domain signal. Each frequency component is determined by X(f), and The time variation pattern of the frequency component is defined.

[0086] S127: Perform energy compensation on the second transformation result to obtain a prediction result.

[0087] In this embodiment, the second transformation result is multiplied by a coefficient to obtain a prediction result.

[0088] During signal processing, especially after filtering, the energy of the original signal may be reduced because some frequency components are attenuated or completely removed. If this energy loss is not corrected, it may lead to inaccurate subsequent analysis or prediction results. In order to ensure the accuracy and reliability of the model, it is necessary to compensate the energy of the transformed signal.

[0089] In order to solve the energy loss problem mentioned above, we need to adjust the second transformation result. The method used here is to multiply the second transformation result by a specific coefficient. The coefficient is selected to compensate for the energy loss caused by filtering or other processing, thereby improving the prediction accuracy of the model.

[0090] In this embodiment, the coefficient used is determined based on the ratio of the output length to the input length. This is because when a signal passes through a filter, it not only experiences changes in amplitude, but may also experience compression or stretching on a time scale. Therefore, choosing the ratio of the output length to the input length as the compensation coefficient is a reasonable and direct way to effectively restore the energy level of the signal.

[0091] The formula is as follows: compensated signal = second transformation result × (output length / input length); the output length here refers to the length of the result signal obtained after all processing steps (including transformation and filtering); and the input length refers to the length of the original unprocessed signal.

[0092] By applying the above energy compensation mechanism to the transformed signal, the energy loss caused by the processing process is compensated, making the final prediction result closer to the actual situation. The adaptability and robustness of the model to different input data are enhanced, thereby improving its prediction ability and accuracy. Ensure that the signal can still maintain its original physical properties after a series of mathematical operations.

[0093] The result of all the above steps is the predicted battery health status data.

[0094] In one embodiment, the frequency domain prediction neural network is obtained by obtaining historical lithium battery operation data and corresponding lithium battery health status data as a sample set to train the neural network, including: Obtain historical lithium battery operation data and corresponding lithium battery health status data to obtain a sample set; Build a neural network; The neural network is trained based on the historical lithium battery operation data in the sample set, and the parameters of the trained neural network are adjusted in combination with the corresponding lithium battery health status data to obtain a frequency domain prediction neural network.

[0095] Specifically, the frequency domain prediction neural network is trained using a gradient descent mode. The frequency domain prediction neural network uses a root mean square error as an evaluation index of network performance.

[0096] The historical SOH data is used as training data to train the frequency domain prediction neural network FFS, and then the lithium battery operation data is input into the frequency domain prediction neural network FFS to predict the battery health status to obtain a prediction result.

[0097] S130: Output the prediction result.

[0098] In this embodiment, the prediction result is output to the terminal for display.

[0099] The method of this embodiment obtains the lithium battery operation data in multiple discharge cycles and processes it to obtain the battery health status, and then inputs this data into the frequency domain prediction network FFS based on Fourier transform and high-frequency filtering for prediction, outputs the prediction results and analyzes them. The prediction results are accurate, the prediction speed is very fast, the number of parameters is small, the cost is low, and it is suitable for edge device scenarios. The frequency domain prediction network FFS based on Fourier transform and high-frequency filtering has a much lower number of parameters than other models, which is only one thousandth of the number of parameters of the Autoformer model, so it is suitable for edge devices with limited resources and can reduce costs. At the same time, the prediction accuracy of this method is also quite good, with a root mean square error of only 0.2%.

[0100] The above-mentioned method for rapid estimation of battery health status based on frequency domain analysis obtains lithium battery operation data within multiple discharge cycles; the collected battery operation data is input into the frequency domain prediction neural network, which includes a Fourier transform layer, which can convert time domain data into frequency domain data; the frequency domain data is processed by a filter to extract important features and reduce redundant information; the design of the neural network includes a complex-valued linear layer and an inverse Fourier transform layer, so that the frequency domain information can be effectively converted and used to accurately predict SOH; compared with complex traditional models, the number of parameters of the frequency domain prediction neural network is significantly reduced, which is only one thousandth of the traditional model, and it can run efficiently on resource-constrained edge devices, improving prediction efficiency and accuracy.

[0101] Figure 4 FIG. 3 is a schematic block diagram of a battery health status rapid estimation system 300 based on frequency domain analysis provided by an embodiment of the present invention. Figure 4 As shown, corresponding to the above method for rapidly estimating the battery health state based on frequency domain analysis, the present invention also provides a system 300 for rapidly estimating the battery health state based on frequency domain analysis. The system 300 for rapidly estimating the battery health state based on frequency domain analysis includes a unit for executing the above method for rapidly estimating the battery health state based on frequency domain analysis, and the system can be configured in a server. Specifically, please refer to Figure 4 The battery health status rapid estimation system 300 based on frequency domain analysis includes a data acquisition unit 301, a prediction unit 302 and an output unit 303.

[0102] The data acquisition unit 301 is used to acquire the lithium battery operation data within multiple discharge cycles; the prediction unit 302 is used to input the lithium battery operation data into the frequency domain prediction neural network to predict the battery health status to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by acquiring historical lithium battery operation data and corresponding lithium battery health status data as a sample set to train the neural network, and the neural network includes a data processing layer, a Fourier transform layer, a filter, a complex-valued linear layer, a zero-filling layer and an inverse Fourier transform layer connected in sequence; the output unit 303 is used to output the prediction result.

[0103] In one embodiment, the prediction unit 302 includes: A data processing subunit is used to input the lithium battery operation data into the frequency domain prediction neural network, and perform data processing on the lithium battery operation data to obtain a processing result; a transformation subunit is used to use the Fourier transform layer to convert the processing result into a frequency domain representation to obtain a first transformation result; a filtering subunit is used to use the filter to filter the first transformation result to obtain a filtering result; an interpolation subunit is used to use the complex-valued linear layer to perform frequency domain interpolation on the filtering result to obtain interpolated frequency domain data; a zero-filling subunit is used to use the zero-filling layer to perform a zero-filling operation on the interpolated frequency domain data to obtain a zero-filling result; an inverse transformation subunit is used to use the inverse Fourier transform layer to convert the zero-filling result from the frequency domain back to the time domain to obtain a second transformation result; a compensation subunit is used to perform energy compensation on the second transformation result to obtain a prediction result.

[0104] In one embodiment, the data processing subunit is used to input the lithium battery operation data into a frequency domain prediction neural network, and perform reversible instance normalization on the lithium battery operation data to obtain a processing result.

[0105] In one embodiment, the compensation subunit is used to multiply the second transformation result by a coefficient to obtain a prediction result.

[0106] In one embodiment, the interpolation subunit is used to use the complex-valued linear layer to perform frequency domain interpolation on the filtering result through linear matrix multiplication and matrix addition to obtain interpolated frequency domain data.

[0107] In one embodiment, the zero-filling subunit is used to add zero values ​​to both ends of the interpolated frequency domain data using the zero-filling layer to obtain a zero-filling result.

[0108] In one embodiment, the system also includes a network training unit for acquiring historical lithium battery operation data and corresponding lithium battery health status data to obtain a sample set; constructing a neural network; training the neural network with the historical lithium battery operation data in the sample set, and adjusting the parameters of the trained neural network in combination with the corresponding lithium battery health status data to obtain a frequency domain prediction neural network.

[0109] It should be noted that technicians in the relevant field can clearly understand that the specific implementation process of the above-mentioned battery health status rapid estimation system 300 based on frequency domain analysis and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and conciseness of the description, it will not be repeated here.

[0110] The above-mentioned battery health status fast estimation system 300 based on frequency domain analysis can be implemented in the form of a computer program. The computer program can be used in Figure 5 Runs on the computer device shown.

[0111] See also Figure 5 , Figure 5 5 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0112] See also Figure 5 The computer device 500 includes a processor 502 , a memory and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0113] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, and when the program instructions are executed, the processor 502 may execute a method for quickly estimating a battery health state based on frequency domain analysis.

[0114] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .

[0115] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a method for quickly estimating the battery health status based on frequency domain analysis.

[0116] The network interface 505 is used to communicate with other devices over the network. Figure 5The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0117] The processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps: Acquire lithium battery operation data within multiple discharge cycles; input the lithium battery operation data into a frequency domain prediction neural network to predict the battery health status to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by acquiring historical lithium battery operation data and corresponding lithium battery health status data as a sample set to train the neural network, and the neural network includes a data processing layer, a Fourier transform layer, a filter, a complex-valued linear layer, a zero-filling layer and an inverse Fourier transform layer connected in sequence; output the prediction result.

[0118] In one embodiment, when the processor 502 implements the step of inputting the lithium battery operation data into the frequency domain prediction neural network to predict the battery health state to obtain a prediction result, the processor 502 specifically implements the following steps: The lithium battery operation data is input into the frequency domain prediction neural network, and the lithium battery operation data is processed to obtain a processing result; the processing result is converted into a frequency domain representation by using the Fourier transform layer to obtain a first transformation result; the first transformation result is filtered by using the filter to obtain a filtering result; the filtering result is interpolated in the frequency domain by using the complex-valued linear layer to obtain interpolated frequency domain data; the interpolated frequency domain data is zero-filled by using the zero-filling layer to obtain a zero-filled result; the zero-filled result is converted from the frequency domain back to the time domain by using the inverse Fourier transform layer to obtain a second transformation result; and the second transformation result is energy compensated to obtain a prediction result.

[0119] In one embodiment, when the processor 502 implements the step of inputting the lithium battery operation data into the frequency domain prediction neural network and performing data processing on the lithium battery operation data to obtain a processing result, the processor 502 specifically implements the following steps: The lithium battery operation data is input into a frequency domain prediction neural network, and reversible instance normalization is performed on the lithium battery operation data to obtain a processing result.

[0120] In one embodiment, when the processor 502 implements the step of performing energy compensation on the second transformation result to obtain a prediction result, the processor 502 specifically implements the following steps: The second transformation result is multiplied by the coefficient to obtain a prediction result.

[0121] In one embodiment, when the processor 502 implements the step of performing frequency domain interpolation on the filtering result using the complex-valued linear layer to obtain interpolated frequency domain data, the processor 502 specifically implements the following steps: The complex-valued linear layer is used to perform frequency domain interpolation on the filtering result through linear matrix multiplication and matrix addition to obtain interpolated frequency domain data.

[0122] In one embodiment, when the processor 502 implements the step of performing a zero-filling operation on the interpolated frequency domain data using the zero-filling layer to obtain a zero-filling result, the processor 502 specifically implements the following steps: The zero-filling layer is used to add zero values ​​to both ends of the interpolated frequency domain data to obtain a zero-filling result.

[0123] In one embodiment, when the processor 502 implements the step of obtaining the historical lithium battery operation data and the corresponding lithium battery health status data as a sample set to train the neural network, the processor 502 specifically implements the following steps: Acquire historical lithium battery operation data and corresponding lithium battery health status data to obtain a sample set; construct a neural network; train the neural network with the historical lithium battery operation data in the sample set, and adjust the parameters of the trained neural network in combination with the corresponding lithium battery health status data to obtain a frequency domain prediction neural network.

[0124] Wherein, the frequency domain prediction neural network is trained using a gradient descent mode.

[0125] The frequency domain prediction neural network uses the root mean square error as an evaluation index of the network performance.

[0126] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0127] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0128] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps: The lithium battery operation data is input into the frequency domain prediction neural network, and the lithium battery operation data is processed to obtain a processing result; the processing result is converted into a frequency domain representation by using the Fourier transform layer to obtain a first transformation result; the first transformation result is filtered by using the filter to obtain a filtering result; the filtering result is interpolated in the frequency domain by using the complex-valued linear layer to obtain interpolated frequency domain data; the interpolated frequency domain data is zero-filled by using the zero-filling layer to obtain a zero-filled result; the zero-filled result is converted from the frequency domain back to the time domain by using the inverse Fourier transform layer to obtain a second transformation result; and the second transformation result is energy compensated to obtain a prediction result.

[0129] In one embodiment, when the processor executes the computer program to implement the step of inputting the lithium battery operation data into the frequency domain prediction neural network to predict the battery health state to obtain a prediction result, the following steps are specifically implemented: The lithium battery operation data is input into the frequency domain prediction neural network, and the lithium battery operation data is processed to obtain a processing result; the processing result is converted into a frequency domain representation by using the Fourier transform layer to obtain a first transformation result; the first transformation result is filtered by using the filter to obtain a filtering result; the filtering result is interpolated in the frequency domain by using the complex-valued linear layer to obtain interpolated frequency domain data; the interpolated frequency domain data is zero-filled by using the zero-filling layer to obtain a zero-filled result; the zero-filled result is converted from the frequency domain back to the time domain by using the inverse Fourier transform layer to obtain a second transformation result; and the second transformation result is energy compensated to obtain a prediction result.

[0130] In one embodiment, when the processor executes the computer program to implement the step of inputting the lithium battery operation data into the frequency domain prediction neural network and performing data processing on the lithium battery operation data to obtain a processing result, the processor specifically implements the following steps: The lithium battery operation data is input into a frequency domain prediction neural network, and reversible instance normalization is performed on the lithium battery operation data to obtain a processing result.

[0131] In one embodiment, when the processor executes the computer program to implement the step of performing energy compensation on the second transformation result to obtain a prediction result, the processor specifically implements the following steps: The second transformation result is multiplied by the coefficient to obtain a prediction result.

[0132] In one embodiment, when the processor executes the computer program to implement the step of using the complex-valued linear layer to perform frequency domain interpolation on the filtering result to obtain interpolated frequency domain data, the processor specifically implements the following steps: The complex-valued linear layer is used to perform frequency domain interpolation on the filtering result through linear matrix multiplication and matrix addition to obtain interpolated frequency domain data.

[0133] In one embodiment, when the processor executes the computer program to implement the step of performing a zero-filling operation on the interpolated frequency domain data using the zero-filling layer to obtain a zero-filling result, the processor specifically implements the following steps: The zero-filling layer is used to add zero values ​​to both ends of the interpolated frequency domain data to obtain a zero-filling result.

[0134] In one embodiment, when the processor executes the computer program to implement the step that the frequency domain prediction neural network is obtained by training the neural network by acquiring historical lithium battery operation data and corresponding lithium battery health status data as a sample set, the processor specifically implements the following steps: Acquire historical lithium battery operation data and corresponding lithium battery health status data to obtain a sample set; construct a neural network; train the neural network with the historical lithium battery operation data in the sample set, and adjust the parameters of the trained neural network in combination with the corresponding lithium battery health status data to obtain a frequency domain prediction neural network.

[0135] Wherein, the frequency domain prediction neural network is trained using a gradient descent mode.

[0136] The frequency domain prediction neural network uses the root mean square error as an evaluation index of the network performance.

[0137] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which are computer-readable storage media that can store program codes.

[0138] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0139] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0140] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the system of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0141] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0142] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A fast estimation method of battery health status based on frequency domain analysis, characterized in that: include: Obtain lithium battery operation data over multiple discharge cycles; The lithium battery operation data is input into a frequency domain prediction neural network to predict the battery health state to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by obtaining historical lithium battery operation data and corresponding lithium battery health state data as a sample set to train a neural network, and the neural network includes a data processing layer, a Fourier transform layer, a filter, a complex linear layer, a zero filling layer and an inverse Fourier transform layer connected in sequence; the lithium battery operation data includes battery capacity, current, voltage and charge and discharge cycle; The prediction result is output.

2. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 1, characterized in that: The step of inputting the lithium battery operation data into a frequency domain prediction neural network to predict the battery health status to obtain a prediction result includes: Inputting the lithium battery operation data into a frequency domain prediction neural network, and performing data processing on the lithium battery operation data to obtain a processing result; Using the Fourier transform layer to convert the processing result into a frequency domain representation to obtain a first transformation result; Using the filter to filter the first transformation result to obtain a filtering result; Performing frequency domain interpolation on the filtering result by using the complex-valued linear layer to obtain interpolated frequency domain data; Using the zero-filling layer to perform a zero-filling operation on the interpolated frequency domain data to obtain a zero-filling result; Converting the zero-filled result from the frequency domain back to the time domain using the inverse Fourier transform layer to obtain a second transform result; Energy compensation is performed on the second transformation result to obtain a prediction result.

3. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 2, characterized in that: The step of inputting the lithium battery operation data into a frequency domain prediction neural network and performing data processing on the lithium battery operation data to obtain a processing result includes: The lithium battery operation data is input into a frequency domain prediction neural network, and reversible instance normalization is performed on the lithium battery operation data to obtain a processing result.

4. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 2, characterized in that: The performing energy compensation on the second transformation result to obtain a prediction result includes: The second transformation result is multiplied by the coefficient to obtain a prediction result.

5. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 2, characterized in that: The using the complex-valued linear layer to perform frequency domain interpolation on the filtering result to obtain interpolated frequency domain data includes: The complex-valued linear layer is used to perform frequency domain interpolation on the filtering result through linear matrix multiplication and matrix addition to obtain interpolated frequency domain data.

6. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 2, characterized in that: The using the zero-filling layer to perform a zero-filling operation on the interpolated frequency domain data to obtain a zero-filling result includes: The zero-filling layer is used to add zero values ​​to both ends of the interpolated frequency domain data to obtain a zero-filling result.

7. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 1, characterized in that: The frequency domain prediction neural network is obtained by obtaining historical lithium battery operation data and corresponding lithium battery health status data as a sample set to train the neural network, including: Obtain historical lithium battery operation data and corresponding lithium battery health status data to obtain a sample set; Build a neural network; The neural network is trained based on the historical lithium battery operation data in the sample set, and the parameters of the trained neural network are adjusted in combination with the corresponding lithium battery health status data to obtain a frequency domain prediction neural network.

8. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 7, characterized in that: The frequency domain prediction neural network is trained using a gradient descent mode.

9. The method for rapid estimation of battery health status based on frequency domain analysis according to claim 7, characterized in that: The frequency domain prediction neural network uses the root mean square error as an evaluation index of the network performance.

10. A fast battery health status estimation system based on frequency domain analysis, characterized in that: include: A data acquisition unit, used to acquire lithium battery operation data within multiple discharge cycles; A prediction unit, used for inputting the lithium battery operation data into a frequency domain prediction neural network to perform battery health state prediction to obtain a prediction result; wherein the frequency domain prediction neural network is obtained by training a neural network by acquiring historical lithium battery operation data and corresponding lithium battery health state data as a sample set, and the neural network includes a data processing layer, a Fourier transform layer, a filter, a complex-valued linear layer, a zero-filling layer, and an inverse Fourier transform layer connected in sequence; An output unit is used to output the prediction result.

Images