A life prediction method of a super capacitor monitoring system based on sparse data

Abstract

A kind of life prediction method of super capacitor monitoring system based on sparse data, comprising the following steps: parameter identification is carried out to sparse monitoring data based on interior point method;Old age characteristic parameter is predicted based on GRU neural network, and more sample data conforming to the evolution trend of original old age characteristic parameter is generated;Time series decomposition result of characteristic value is determined using time series model, and aging factor reflecting the recession trend of vehicle-mounted super capacitor is extracted;The end of life of vehicle-mounted super capacitor is predicted by combining super capacitor life termination criterion.The present application provides a method for predicting the life of vehicle-mounted super capacitor by a small amount of sample data, solves the problem of directly extracting aging factor representing the recession trend of super capacitor from sparse and discontinuous low-quality vehicle-mounted monitoring data;Without disassembling vehicle-mounted super capacitor energy storage system, online life prediction can be carried out, which is of great significance for the optimization operation and predictive maintenance of vehicle-mounted super capacitor system.

Description

Technical Field

[0001] This invention relates to a method for extracting aging factors and predicting lifespan of supercapacitor energy storage systems, which can be applied to the technical fields of digital operation and maintenance of large-scale supercapacitor energy storage systems. Background Technology

[0002] Supercapacitors are energy storage devices with high power density, long cycle life, wide operating temperature range, and the ability to charge and discharge at high currents. They are widely used in urban buses, electric ferries, and trams. In actual operation, onboard monitoring systems monitor the voltage, current, temperature, and other information of the supercapacitor system in real time and transmit the data back to the data center via wireless communication. By extracting and analyzing aging characteristics from historical monitoring data, the degree of aging of the supercapacitor can be determined. Existing research typically uses data-driven methods to predict the evolution of aging characteristics of onboard supercapacitors, thereby achieving lifespan prediction. Accurate lifespan prediction can provide a basis for predictive maintenance of onboard supercapacitors, thus ensuring their safe and reliable operation.

[0003] On the one hand, the operating conditions of vehicle-mounted supercapacitors are complex and variable, with large random fluctuations in voltage and current. The sampling frequency of their on-board monitoring data is typically 0.1Hz, lower than the 2Hz sampling frequency required by common parameter identification methods. On the other hand, the sample size of on-board monitoring data is relatively small, only one or two years' worth of data. Therefore, extracting aging characteristics from the monitoring data using traditional methods is difficult. Furthermore, due to cost and other constraints, the temperature inside the vehicle-mounted supercapacitor's casing is not constant and is easily affected by external temperatures. Its aging characteristic parameters exhibit periodic characteristics that change seasonally with temperature, making the extracted aging characteristic parameters unsuitable for directly characterizing supercapacitor aging. Traditional supercapacitor lifespan prediction studies mostly use laboratory cyclic charge-discharge experiments as the source of aging data. The aging characteristics rely on static test conditions and cannot be directly used to characterize the aging of vehicle-mounted supercapacitors, making it difficult to apply to the lifespan prediction of vehicle-mounted supercapacitors. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a method for extracting aging factors and predicting the lifespan of supercapacitors suitable for vehicle-mounted operation. This method employs a GRU neural network to generate more sample data, solving the problem of insufficient sample data in lifespan prediction. Simultaneously, it extracts the supercapacitor aging factor from the feature parameter sequence, mitigating the impact of periodic variations on the feature parameters. Combined with a lifespan termination criterion, this method enables lifespan prediction for vehicle-mounted supercapacitors.

[0005] Technical solution:

[0006] A method for predicting the lifetime of a supercapacitor monitoring system based on sparse data, comprising the following steps:

[0007] Step 1: After preprocessing the acquired vehicle-mounted supercapacitor monitoring data, perform parameter identification on the sparse monitoring data based on the interior point method.

[0008] Step 2: Predict aging characteristic parameters based on GRU (Gate Recurrent Unit) neural network to generate more sample data that conforms to the evolution trend of the original aging characteristic parameters;

[0009] Step 3: Based on the feature parameter prediction results of Step 2, establish a time series model, use the time series model to determine the time series decomposition results of the feature values, and extract the aging factor that can reflect the degradation trend of the vehicle supercapacitor.

[0010] Step 4: Based on the aging factors extracted in Step 3, and combined with the supercapacitor lifespan termination criteria, the lifespan end point of the vehicle-mounted supercapacitor is predicted.

[0011] Compared with existing methods, the present invention has the following characteristics and beneficial effects:

[0012] This invention provides a method for extracting aging factors and predicting the lifespan of sparse, real-world monitoring data of vehicle-mounted supercapacitors. The invention utilizes vehicle-mounted operating condition data, characterized by sparseness and discontinuity, and employs the interior-point method to identify parameters. A GRU neural network is then used to predict the identified aging parameters, generating more aging characteristic parameters that conform to historical variation patterns, thus addressing the problem of insufficient original sample size. Furthermore, a time-series model is established, and the predicted characteristic parameter sequence is decomposed using a time-series decomposition method to eliminate the influence of seasonal temperature periodic changes on the characteristic parameters, obtaining aging factors that characterize the degradation trend of vehicle-mounted supercapacitors. Finally, combined with supercapacitor lifespan termination criteria, the lifespan of the vehicle-mounted supercapacitor is predicted and validated.

[0013] This invention presents a method for predicting the lifespan of vehicle-mounted supercapacitors using a small amount of sample data, and solves the problem of directly extracting aging factors characterizing the degradation trend of supercapacitors from sparse, discontinuous, low-quality vehicle-mounted monitoring data. This method does not require disassembly of the vehicle-mounted supercapacitor energy storage system and enables online lifespan prediction, which is of great significance for the optimized operation and predictive maintenance of vehicle-mounted supercapacitor systems. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention;

[0015] Figure 2 This is a schematic diagram of the equivalent circuit model of a supercapacitor selected in an embodiment of the present invention;

[0016] Figure 3This is a diagram showing the extraction results of aging characteristic parameters of the vehicle-mounted supercapacitor in the embodiment;

[0017] Figure 4 This is a schematic diagram of the thermal stability range of the vehicle-mounted supercapacitor in the embodiment;

[0018] Figure 5 In this example, a GRU neural network is used to predict aging characteristic parameter curves.

[0019] Figure 6 This example compares the prediction results of supercapacitor lifetime using whether or not the GRU neural network is used to predict feature parameters.

[0020] Figure 7 This is a schematic diagram of the on-board supercapacitor life prediction steps in the embodiment. Detailed Implementation

[0021] The inventors proposed a method for identifying supercapacitor parameters under sparse data conditions in their accepted patent (Acceptance No.: 202210886369.1, Patent Title: A Method for Identifying Equivalent Circuit Parameters of Supercapacitors Applicable to Sparse Data). This method solves the problem of large errors in traditional parameter identification methods under sparse data conditions and achieves the identification of aging characteristic parameters (characteristic capacitance C and characteristic internal resistance) based on vehicle-mounted supercapacitor monitoring data at a sampling rate of 0.1Hz. Based on this, the present invention further proposes a method for extracting aging factors and predicting the lifespan of vehicle-mounted supercapacitors using aging characteristic parameters from a small amount of sample data.

[0022] The technical solutions provided in this application will be further described below with reference to specific embodiments and accompanying drawings. The advantages and features of this application will become clearer from the following description.

[0023] This invention proposes a method for predicting the lifetime of a supercapacitor monitoring system based on sparse data, the processing flow of which is as follows: Figure 1 As shown.

[0024] Step 1: Sparse data processing and parameter identification.

[0025] The original sparse data undergoes preprocessing such as data cleaning and selection, and is then based on the first-order equivalent circuit model of a supercapacitor (e.g., Figure 2 Based on the voltage and current data (as shown), a fitness function characterizing the error of the identification results was constructed. An optimization algorithm minimizing the fitness function using the interior point penalty function method was employed to achieve accurate identification of supercapacitor parameters under sparse data. For detailed process, please refer to the accepted patent (Acceptance No.: 202210886369.1). However, the ambient temperature of the vehicle-mounted supercapacitor is not constant during operation, but rather varies periodically with the seasons. The identified characteristic capacitance value C and characteristic internal resistance... It is not a monotonous change, but rather exhibits obvious periodic fluctuations accompanying time and seasons, such as... Figure 3 As shown, it cannot be directly used to characterize the aging of vehicle-mounted supercapacitors.

[0026] Step 2: Predict feature parameters using a GRU neural network

[0027] To address the issue of small original sample data volume, this invention uses a GRU neural network to adjust the feature tolerance. and characteristic internal resistance To generate more sample data that conforms to the evolution trend of aging characteristic parameters, predictions are made. Considering the influence of temperature on the characteristic parameters, the daily thermal stability interval temperature is introduced into the input of the GRU network. The internal temperature of the vehicle-mounted supercapacitor fluctuates during operation, but there is always a temperature range that tends to stabilize on each day of operation. This range is called the thermal stability range. Figure 4 As shown. Simultaneously, to enable the GRU model to learn the temporal characteristics of the feature parameters, historical values ​​of the feature parameters are also introduced into the model input. This is used to predict the... Characteristic capacitance value of each operating day For example, the input to the GRU model consists of the thermal stability range temperature of the previous operating day. and the characteristic capacitance value of the previous operating day Composition, that is The output of the GRU model is .in, , M This represents the current total number of operating days.

[0028] The GRU neural network constructed in this invention comprises one GRU layer and one fully connected layer, with 12 and 6 hidden neurons respectively. The learning rate used during the training of the neural network model is... Number of training cycles The loss function is the mean absolute error (MAE). This GRU neural network is used to assess the feature tolerance. and characteristic internal resistance The prediction error is approximately 2%.

[0029] Step 3: Based on the feature parameter prediction results of Step 2, establish a time series model, decompose the feature parameter sequence, and extract the aging factor that can reflect the degradation trend of the vehicle-mounted supercapacitor.

[0030] 3.1 Establishing a time series model

[0031] To address the periodicity issue, this invention proposes a time series model incorporating sinusoidal and polynomial components. The model decomposes the characteristic parameter sequence, using sinusoidal components to characterize the fluctuations in characteristic parameters caused by seasonal temperature variations, and using polynomial components to characterize the performance degradation trend of the supercapacitor. The model is shown in equation (1):

[0032] (1)

[0033] t represents the decomposition result of the characteristic parameters. D For the operating days. Among them, The periodic term characterizes the effect of temperature fluctuations on characteristic parameters, and its amplitude and phase are determined by undetermined coefficients. and The decision has been made to set the cycle at 365 days. The polynomial characterizes the overall performance degradation trend of the vehicle-mounted supercapacitor, with undetermined coefficients being... .

[0034] 3.2 Solving for Time Series Model Parameters

[0035] In order to obtain the decomposition results of the time series characteristic parameters, in order to solve the eigenvalue time series of polynomial degree m=3 Sequence decomposition results For example,

[0036] Represented as:

[0037] (2)

[0038] Represented as:

[0039] (3)

[0040] By minimizing and Sum of squares of errors It can solve for the undetermined coefficients. , and As shown in (4). Among them This marks the end of the runtime.

[0041] (4) In the formula, and Coupling in the sine function, minimize This is a nonlinear least squares problem, and direct solution involves a large amount of computation. Therefore, it is advisable to decouple the two components before solving. The decoupling is shown in equation (5), where... Let A and It is denoted as B.

[0042] (5)

[0043] Sum of squared errors Transformed into:

[0044] (6)

[0045] The principle of the least squares method is as follows:

[0046] (7)

[0047] (8)

[0048] For polynomial fitting of degree n, Let X be a parameter, T be a matrix, and T be the matrix transpose.

[0049] (9)

[0050] (10)

[0051] (11)

[0052] At this point, the least squares method can be used to solve equation (6). The undetermined coefficients A, B, and ... when minimizing Furthermore, the coefficients of the sine term can be solved inversely. and : and .

[0053] 3.3 Determine the aging factor, i.e., the polynomial trend term function.

[0054] To determine the polynomial trend term function, the fitting results for different values ​​of m were considered to determine the sequence decomposition results of the characteristic capacitance time series. The fitting results show that the time series decomposition method can basically reflect the periodic changes of the characteristic capacitance series and give the overall decay trend of the characteristic capacitance series. Among them, the aging factor, which characterizes the decay trend of the vehicle-mounted supercapacitor, is a monotonic linear function.

[0055] (12)

[0056] Step 4: Predict the lifespan of the vehicle-mounted supercapacitor based on the extracted aging factors

[0057] To make the aging factor of characteristic capacitance C This invention further improves upon the performance of actual vehicle-mounted supercapacitors, ensuring consistency with their actual performance. A reduction was performed (unification of initial values), and the initial values ​​of the trend term were adjusted. Reduced to factory inspection tolerance value The result of the reduction is recorded as :

[0058] (13)

[0059] Construct a composite sequence based on the prediction results of the GRU neural network in step 2. The reduced capacity of the composite sequence is calculated using equation (13). ,by Reduced to factory test tolerance 80% of the nodes are considered the end of their lifespan.

[0060] (14)

[0061] (15)

[0062] Therefore, the estimated end-of-life results of on-board supercapacitors can be obtained. (End of Life) is:

[0063] (16)

[0064] It should be noted that the above decomposition method is also applicable to characteristic internal resistance. Characteristic internal resistance complex sequence The recalculated value Reaching twice the internal resistance during factory testing can be considered the end of the lifespan of an on-board supercapacitor.

[0065] In embodiments of the present invention, a characteristic capacitance value C is selected to extract the aging factor. Capacitance is calculated from current charging and discharging data; its uncertainty is relatively small, mainly stemming from the selection of the voltage range involved in the capacitance calculation. However, the uncertainty of internal resistance is greater because it changes dynamically during charging and discharging, only stabilizing after the charge distribution process is completed within the supercapacitor. Therefore, embodiments of the present invention select a characteristic capacitance value C to extract the aging factor for further analysis.

[0066] In an embodiment of the present invention, the research object is a vehicle driven by a supercapacitor. Due to the uncertainty and complexity of its operating conditions, the operating current of this supercapacitor system is relatively random.

[0067] like Figure 5 , Figure 6As shown, according to step 2 of the specific implementation method, the feature parameters were predicted using a GRU neural network to obtain more sample data. The error between the predicted and actual values ​​demonstrates the accuracy of the GRU method in predicting the feature parameters. Then, according to step 3, the aging factor that can characterize the degradation trend of the studied on-board supercapacitor system was obtained. Finally, based on the reduction result of the feature capacitance composite sequence, the end-of-life of the tram supercapacitor system was estimated using equation (16). The detailed steps for predicting the lifespan of the on-board supercapacitor are as follows: Figure 7 As shown. It should be noted that, considering that the field of supercapacitor research pays more attention to capacitance degradation, the end-of-life estimation of the tram supercapacitor system in this embodiment is only based on the characteristic capacitance value.

[0068] The above description is merely a description of preferred embodiments of this application and is not intended to limit the scope of this application in any way. Any changes or modifications made by those skilled in the art based on the above-disclosed technical content should be considered as equivalent and valid embodiments and fall within the scope of protection of the technical solution of this application.

Claims

1. A method for predicting the lifetime of a supercapacitor monitoring system based on sparse data, characterized in that, Including the following steps: Step 1: After preprocessing the acquired vehicle-mounted supercapacitor monitoring data, perform parameter identification on the sparse monitoring data based on the interior point method. Step 2: Predict aging characteristic parameters based on GRU neural network to generate more sample data that conforms to the evolution trend of the original aging characteristic parameters; Step 3: Based on the feature parameter prediction results of Step 2, establish a time series model, use the time series model to determine the time series decomposition results of the feature values, and extract the aging factor that can reflect the degradation trend of the vehicle supercapacitor. Step 4: Based on the aging factors extracted in Step 3, and combined with the supercapacitor lifespan termination criteria, predict the lifespan end point of the vehicle-mounted supercapacitor. Step 3 includes: 3.1 Establishing a time series model To address the periodicity issue, a time series model incorporating sinusoidal and polynomial components is established. The characteristic parameter sequence is decomposed, with the sinusoidal component representing the fluctuations in characteristic parameters caused by seasonal temperature changes, and the polynomial component representing the performance degradation trend of the supercapacitor. The time series model is shown in equation (1). (1) t represents the decomposition result of the characteristic parameters. D For operation days; The periodic term characterizes the effect of temperature fluctuations on characteristic parameters, and its amplitude and phase are determined by undetermined coefficients. and The decision has been made to set the cycle at 365 days. The polynomial characterizes the overall performance degradation trend of the vehicle-mounted supercapacitor, with undetermined coefficients being... ; 3.2 Solving for Time Series Model Parameters The undetermined coefficients are solved by minimizing the sum of squared errors between the time series of the characteristic parameters and the time series decomposition results of the characteristic parameters. and ,as well as ; 3.3 Determine the aging factor, i.e., the polynomial trend term function. To determine the polynomial trend term function, the fitting results for different values ​​of m are considered to determine the sequence decomposition results of the time series. The time series decomposition method basically reflects the periodic changes of the characteristic capacitor series and gives the overall decay trend of the characteristic capacitor series. Among them, the aging factor that can characterize the decay trend of the vehicle supercapacitor is a monotonic linear function. Step 4: The lifespan of on-board supercapacitors is predicted based on the aging factor extracted from the characteristic capacitance value C. Specifically: To make the aging factor of characteristic capacitance C It can match the performance of actual vehicle-mounted supercapacitors, and is able to The reduction was performed, and the result is recorded as follows: : (13) in, This is the initial value for the trend term. This is the tolerance value for factory inspection. Construct a composite sequence based on the prediction results of the GRU neural network in step 2. The reduced capacity of the composite sequence is calculated using equation (13). ,by Reduced to factory test tolerance 80% of the nodes are considered the end of their lifetime; (14) (15) Therefore, the estimated end-of-life results of on-board supercapacitors can be obtained. (End of Life) is: (16)。 2. The method as described in claim 1, characterized in that, Step 2: Using a GRU neural network to evaluate eigenvalues and characteristic internal resistance Make predictions to generate more sample data that conform to the evolution trend of aging characteristic parameters; To consider the effect of temperature on feature parameters, the daily thermal stability interval temperature is introduced into the input of the GRU network. The temperature inside the supercapacitor fluctuates during operation, but there is a temperature range that tends to stabilize on each day of operation, meaning that the supercapacitor has entered a thermally stable state. This range is called the thermally stable range. At the same time, in order to enable the GRU model to learn the temporal characteristics of the feature parameters, the historical values ​​of the feature parameters are also introduced into the model input.

3. The method as described in claim 2, characterized in that: The GRU neural network comprises one GRU layer and one fully connected layer, with 12 and 6 hidden neurons respectively; the learning rate used during the training of the neural network model is... Number of training cycles The loss function is the mean absolute error (MAE).

4. The method as described in claim 1, characterized in that, Step 3: By decomposing the time series of the characteristic capacitance C, the aging factor is extracted, specifically: Solving the eigenvalue time series when the polynomial degree m=3 Sequence decomposition results ,in: Represented as: (2) Represented as: (3) By minimizing and Sum of squares of errors Solving for undetermined coefficients , and ,in This marks the end of the runtime; (4) In the formula, and Coupling in the sine function, minimize This is a nonlinear least squares problem, and direct solution involves a large amount of computation. Therefore, the two are decoupled before solution. The decoupling is shown in equation (5), where, Let A and Let it be B; (5) Sum of squared errors Transformed into: (6) The principle of the least squares method is as follows: (7) (8) For polynomial fitting of degree n, Let X be a parameter, X be a matrix, and T represent the matrix transpose. (9) (10) (11) At this point, the least squares method is used to solve equation (6). The undetermined coefficients A, B, and ... when minimizing Furthermore, the coefficients of the sine term are obtained by inverse solution. and : and ; The aging factor, which characterizes the degradation trend of on-board supercapacitors, is a monotonic linear function: (12)。

Images