Electrochemical impedance spectroscopy testing system based on multi-model combination

Abstract

The invention relates to the technical field of electrochemical testing, in particular to an electrochemical impedance spectroscopy testing system based on multi-model combination, which comprises a data acquisition unit, a deep learning analysis unit, a frequency control unit and a feedback regulation unit. The deep learning analysis unit adopts a convolutional neural network and a long-short-term memory network to construct a model, after training, features are extracted, trends are analyzed, and dynamic frequency priority ranking is carried out, the frequency control unit drives the electrochemical workstation to adjust the frequency according to a weight generation strategy, and the feedback adjustment unit calculates predicted and actually measured impedance errors. When the threshold value is exceeded, parameter correction is carried out in each unit, the test frequency is intelligently optimized according to the dynamic change of the electrochemical process, key information is accurately captured, resource waste is avoided, the test efficiency and accuracy are improved, the stability and reliability of the system are ensured through a feedback regulation mechanism, and the research and analysis capability of the electrochemical system is effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of electrochemical testing, in particular to an electrochemical impedance spectrum testing system based on multi-model combination. Background Art

[0002] Electrochemical testing is an important technology. In electrochemical research and applications, electrochemical impedance spectroscopy testing is a key means to analyze the internal mechanism of the electrochemical system. However, the traditional electrochemical impedance spectroscopy testing system faces many difficulties.

[0003] Previous test systems usually use fixed frequency scanning or simple manual frequency adjustment, which makes it difficult to optimize the test frequency in real time according to the dynamic changes of the electrochemical process. When studying the charging and discharging process of lithium-ion batteries and the progress of metal corrosion, the electrochemical characteristics of different stages change significantly, and key information is often scattered in different frequency bands. Fixed frequency testing may miss important electrochemical change details, resulting in an incomplete and inaccurate understanding of the system. At the same time, due to the lack of effective frequency screening and dynamic adjustment mechanisms, a lot of test time and resources are wasted on scanning irrelevant or secondary frequency areas, which seriously reduces the test efficiency. In order to solve this technical problem, we provide an electrochemical impedance spectroscopy test system based on multi-model combination. Summary of the invention

[0004] The purpose of the present invention is to provide an electrochemical impedance spectroscopy testing system based on a multi-model combination to solve the problems raised in the above background technology.

[0005] To achieve the above purpose, an electrochemical impedance spectroscopy test system based on multi-model combination is provided, including a data acquisition unit, a deep learning analysis unit, a frequency control unit and a feedback adjustment unit; The data acquisition unit collects electrochemical impedance spectroscopy data through an electrochemical workstation and acquires test environment data through a multi-source sensor; The deep learning analysis unit constructs an electrochemical process analysis model, automatically extracts local features of electrochemical impedance spectroscopy data and obtains dynamic change rules in time series, collects electrochemical impedance spectroscopy data and corresponding real state information to train the model, and then uses the trained model to extract and analyze features, determines the key stages and feature change trends of the current electrochemical process based on the analysis results, and assigns time-varying priority weights to different frequency points through a dynamic frequency priority sorting algorithm; The frequency control unit receives the frequency priority weight information, generates a frequency adjustment amount in combination with the current test frequency setting, and then sends a control instruction to the electrochemical workstation according to the generated frequency adjustment amount to drive it to change the frequency scanning parameters, and at the same time, monitors the working state and feedback information of the electrochemical workstation; During the test process after each frequency adjustment of the feedback regulation unit, the data acquisition unit transmits the newly collected electrochemical impedance data to the deep learning analysis unit in real time. The model predicts the impedance change trend that should be obtained in this test based on the current test frequency and analysis results, compares it with the actual measured data, calculates the error index between the two, and triggers feedback regulation based on the error index.

[0006] As a further improvement of the present technical solution, the deep learning analysis unit includes a model building training module. In the model building training module, the specific method of building an electrochemical process analysis model is as follows: The architecture combines a convolutional neural network with a long short-term memory network, wherein the convolutional neural network is used to perform a convolution operation on the input electrochemical impedance spectroscopy data to extract the local feature map of the electrochemical impedance spectroscopy data; The feature map output by the convolutional neural network is expanded in time series as the input of the long short-term memory network. The long short-term memory network calculates according to the preset formula to obtain the dynamic change law and finally constructs the electrochemical process analysis model.

[0007] As a further improvement of the technical solution, the specific steps of the model construction training module using the trained model to perform feature extraction and analysis are as follows: The collected electrochemical impedance spectroscopy data is input into the trained model. After the forward propagation process of the model, the feature vector representation is obtained. For the convolutional neural network part, the feature vector is the output after multi-layer convolution and pooling operations. For the long short-term memory network part, the feature vector is the hidden state at the last moment. The principal component analysis algorithm is used to reduce the dimensionality of the eigenvector, calculate its covariance matrix, and perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The eigenvector corresponding to the largest eigenvalue is used as the new basis vector, and the original eigenvector is projected into the new basis vector space to obtain the eigenvector after dimensionality reduction; According to the eigenvector after dimensionality reduction, the clustering analysis algorithm is used to divide the electrochemical process into stages, set the number of cluster centers, and randomly initialize the cluster centers. The distance from each sample to the cluster center is calculated, and the sample is assigned to the class to which the nearest cluster center belongs. Then the cluster center is updated and the above process is repeated until the cluster center no longer changes, that is, the key stages and characteristic change trends of the current electrochemical process are determined.

[0008] As a further improvement of the present technical solution, the deep learning analysis unit includes a frequency priority determination module, in which the specific implementation of the dynamic frequency priority sorting algorithm is as follows: According to the key stages and characteristic change trends of the electrochemical process obtained by analysis, the key frequency range associated with each stage and trend is determined; For each critical frequency range, an initial priority weight is first assigned, and then the priority weight is dynamically adjusted based on the real-time collected data and model analysis results; Set the current time to , the last adjustment time was During this period of time, if the impedance change rate within a certain critical frequency range exceeds a preset threshold , then increase its priority weight. If the impedance change rate is less than , then reduce its priority weight. If the impedance change rate is The priority weight remains unchanged.

[0009] As a further improvement of the technical solution, the frequency control unit includes a frequency strategy generation module, in which the specific algorithm for generating the frequency adjustment amount is as follows: Receive frequency priority weight information, obtain the test frequency currently set by the electrochemical workstation, and calculate the adjustment amplitude of each frequency point; A new test frequency is calculated according to the adjustment amplitude and the current test frequency, and the new test frequency is within the frequency range allowed by the electrochemical workstation.

[0010] As a further improvement of the technical solution, the frequency control unit includes an instruction driving module, in which the specific method of driving the electrochemical workstation to change the frequency scanning parameter according to the frequency adjustment amount is as follows: Extract the target frequency value of each frequency point, the frequency scanning mode and the scanning speed from the frequency adjustment amount, and calculate the time required to reach the target frequency from the current frequency; According to the calculated time and target frequency, a frequency scanning parameter adjustment instruction sequence is generated, and within each time step, the current frequency value is calculated and updated by linear interpolation; The instruction containing the updated frequency value and other related scanning parameters is sent to the electrochemical workstation through the communication interface, and the electrochemical workstation works according to the new frequency scanning parameters.

[0011] As a further improvement of the technical solution, the feedback adjustment unit includes an error index calculation module. In the error index calculation module, the specific algorithm for calculating the error index between the two is as follows: Assume that the predicted impedance change trend is ,in represents the time series, and the measured impedance data is , using the root mean square error algorithm to calculate the error index , the calculation formula is: ; in is the number of data points. In the calculation process, the predicted value and the measured value are first aligned in time series, that is, the difference is calculated at the same time point. Then, the sum of the squares of the differences at each time point is calculated in turn according to the formula, and the average is taken and the square root is taken to obtain the final error index.

[0012] As a further improvement of the present technical solution, the feedback adjustment unit includes a parameter adjustment module. In the parameter adjustment module, a parameter adjustment method involved in triggering feedback adjustment according to an error indicator is as follows: When the calculated error index is greater than a preset threshold, feedback adjustment is initiated, and in the deep learning analysis unit, the model parameters are updated using an adaptive learning rate adjustment method; At the same time, in the frequency control unit, the frequency adjustment range and the priority weight in the frequency adjustment amount are corrected according to the size of the error index, and the priority weight is redistributed.

[0013] Compared with the prior art, the present invention has the following beneficial effects: In the electrochemical impedance spectroscopy test system based on multi-model combination, the deep learning analysis unit uses the architecture that combines convolutional neural networks with long short-term memory networks. It first uses the convolution layer to extract local features of the electrochemical impedance spectroscopy data, and then uses the long short-term memory network to capture the dynamic change rules to build a model. After training, the unit can accurately extract features, analyze trends and perform dynamic frequency priority sorting, effectively identify the key stages of the electrochemical process, and improve analysis accuracy. The frequency control unit calculates the adjustment amplitude and new frequency based on the frequency priority weight of the deep learning analysis unit and the current test frequency through a specific algorithm, and generates instructions to drive the electrochemical workstation to change the frequency scanning parameters, thereby realizing intelligent and precise adjustment of the frequency and improving test efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is an overall block diagram of the present invention.

[0015] The meaning of each number in the figure is: 1. Data acquisition unit; 2. Deep learning analysis unit; 21. Model building and training module; 22. Frequency priority determination module; 3. Frequency control unit; 31. Frequency strategy generation module; 32. Instruction drive module; 4. Feedback adjustment unit; 41. Error index calculation module; 42. Parameter adjustment module. DETAILED DESCRIPTION

[0016] 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 only 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.

[0017] The present invention provides an electrochemical impedance spectroscopy test system based on a multi-model combination, please refer to Figure 1 As shown, it includes a data acquisition unit 1, a deep learning analysis unit 2, a frequency control unit 3 and a feedback adjustment unit 4; The data acquisition unit 1 collects electrochemical impedance spectroscopy data through an electrochemical workstation and acquires test environment data through a multi-source sensor.

[0018] The deep learning analysis unit 2 constructs an electrochemical process analysis model, automatically extracts local features of electrochemical impedance spectroscopy data and obtains dynamic change patterns in time series, collects electrochemical impedance spectroscopy data and corresponding real-state information to train the model, and then uses the trained model to extract and analyze features. Based on the analysis results, the key stages and characteristic change trends of the current electrochemical process are determined, and a dynamic frequency priority sorting algorithm is used to assign priority weights that change over time to different frequency points.

[0019] The deep learning analysis unit 2 includes a model building training module 21. In the model building training module 21, the specific method of building an electrochemical process analysis model is as follows: First, the structure of the convolutional neural network is determined, and multiple convolutional layers are set. Such a setting can capture the local detail information in the electrochemical impedance spectroscopy data in the initial stage. For the convolution kernel weight matrix, the random initialization method is used for initialization, which helps to provide a reasonable starting point for the network in the early stage of training, avoid the problem of gradient disappearance or explosion, and enable the network to effectively learn features.

[0020] For the input electrochemical impedance spectroscopy data, a convolution operation is performed so that the network can learn more complex feature relationships. The convolution operation performs weighted summation on the data through a sliding convolution kernel to extract local features. These features can better reflect the essential characteristics of the electrochemical process and provide strong support for subsequent analysis.

[0021] The feature map output by the convolutional neural network is expanded in time series as the input of the long short-term memory network. Since the electrochemical process is a dynamic process that changes over time, the long short-term memory network is very suitable for processing this kind of time series data. It can capture the long-term dependencies in the data, thereby better handling the long-term dependencies and dynamic changes in the time series.

[0022] By processing time series data through the long short-term memory network, the hidden state finally obtained contains information about the dynamic changes of the electrochemical process. These hidden states can be used as the basis for subsequent analysis. This architecture that combines convolutional neural networks and long short-term memory networks can give full play to the advantages of both. The local features extracted by the convolutional neural network provide more targeted input for the long short-term memory network, and the long short-term memory network can integrate the information of these features in the time dimension, thereby constructing a powerful electrochemical process analysis model, effectively improving the understanding and analysis capabilities of electrochemical processes.

[0023] The specific steps of the model building training module 21 using the trained model to extract and analyze features are as follows: The collected electrochemical impedance spectroscopy data is input into the trained model. The electrochemical impedance spectroscopy data contains rich information about the electrochemical process, but the original data has high dimensions and complex information, making it difficult to analyze directly. After being trained with a large amount of data in the early stage, the model has learned the inherent laws and characteristic patterns in the data.

[0024] When the data passes through the convolutional neural network part, it goes through multiple layers of convolution and pooling operations in sequence to extract local features. Subsequent layers continue to deepen the feature extraction. The pooling operation reduces the data dimension and reduces the amount of calculation while retaining key feature information. After these operations, the convolutional neural network outputs a feature map. Then, the feature map output by the convolutional neural network is expanded in time series as the input of the long short-term memory network. The long short-term memory network integrates information in the time dimension and captures the dynamic changes of features over time. Finally, the hidden state of the long short-term memory network at the last moment is used as the feature vector. This feature vector integrates the dynamic information in the time series and reflects the evolution trend of the electrochemical process. It is crucial for the subsequent analysis of the stage changes of the electrochemical process.

[0025] The principal component analysis algorithm is used to reduce the dimensionality of the eigenvectors obtained above and calculate the covariance. The covariance matrix reflects the correlation between the feature dimensions. By analyzing it, we can find the main change direction in the data. In the analysis of electrochemical processes, due to the existence of multiple interrelated features, the covariance matrix can help us sort out these relationships, remove redundant information, focus on key change patterns, reduce subsequent calculations, and improve analysis efficiency.

[0026] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors. The eigenvalue size reflects the variance of the data in the direction represented by the corresponding eigenvector, that is, the information content. The eigenvector corresponding to the largest eigenvalue is used as the new basis vector, and the original eigenvector is projected into the new basis vector space to obtain the eigenvector after dimensionality reduction. This can greatly reduce the data dimension while retaining most of the key information, so that subsequent clustering analysis and other operations can be carried out more efficiently, avoiding the dimensionality disaster and highlighting the main direction of change of electrochemical process characteristics.

[0027] According to the eigenvector after dimension reduction, the electrochemical process is divided into stages using the cluster analysis algorithm. First, the number of cluster centers is set. These initial centers provide starting points for subsequent clustering iterations. They will guide the classification of samples. The initial randomness helps avoid falling into local optimal solutions and makes the clustering results more universal.

[0028] Calculate the distance from each sample to the cluster center , assign the samples to the class to which the nearest cluster center belongs. This step clusters samples with similar electrochemical process characteristics according to the distance, so that the characteristics of different stages can be distinguished. Then, update the cluster center. The formula is ;in Is a cluster The number of samples, It is clustering The sample set is repeated and the above sample allocation and cluster center update process is repeated until the cluster center no longer changes. At this time, the key stage and characteristic change trend of the current electrochemical process are determined.

[0029] Through cluster analysis, the electrochemical process can be clearly divided into different stages, and the characteristic patterns of each stage can be understood, providing strong support for further studying the mechanism of the electrochemical process and optimizing experimental conditions.

[0030] The deep learning analysis unit 2 includes a frequency priority determination module 22. In the frequency priority determination module 22, the specific implementation of the dynamic frequency priority sorting algorithm is as follows: Based on the key stages and characteristic change trends of the electrochemical process obtained through analysis, the key frequency range associated with each stage and trend is determined. At different stages of the electrochemical process, its internal chemical reactions, material transfer and other mechanisms will change, and these changes will be reflected in different frequency bands of the electrochemical impedance spectrum. Through in-depth research on the physicochemical nature of the electrochemical process, combined with a large amount of previous experimental data and model analysis, these key frequency ranges can be accurately located, providing a basis for subsequent frequency priority allocation.

[0031] For each key frequency range, an initial priority weight is first assigned. The setting of this initial weight is based on historical experience, which guides subsequent dynamic adjustments in a more optimized direction, quickly focusing on key monitoring areas and reducing blind testing in the early stages.

[0032] Set the current time to , the last adjustment time was During this period of time, the electrochemical impedance data is continuously collected in real time through the data acquisition unit 1. For each key frequency range, the impedance change rate is calculated. Suppose a key frequency range is arrive During the time period, the average initial impedance is The average end impedance is , then the impedance change rate ; Real-time data collection can promptly reflect the latest status of the electrochemical process. As a dynamic indicator, the impedance change rate can keenly capture the fluctuation of impedance in each key frequency range, thereby providing a basis for priority weight adjustment and ensuring that the test frequency always closely follows the key changes of the electrochemical process.

[0033] If the impedance change rate within a critical frequency range exceeds a preset threshold , then increase its priority weight. If the impedance change rate is less than , then reduce its priority weight, indicating that the electrochemical process corresponding to this frequency band is relatively stable or enters a secondary stage. Lowering the priority can allocate more test resources to more variable key areas. If the impedance change rate is In between, the priority weights are kept unchanged, the current test focus is maintained, unnecessary frequent adjustments are avoided, and the stability of the test process is ensured.

[0034] Through the above dynamic frequency priority sorting algorithm, the priority of the test frequency can be reasonably allocated according to the real-time dynamics of the electrochemical process, making the entire electrochemical impedance spectroscopy test system run more intelligently and efficiently, accurately capturing the key information of the electrochemical process, and providing strong support for in-depth research and optimization of the electrochemical process.

[0035] The frequency control unit 3 receives the frequency priority weight information, and generates a frequency adjustment amount in combination with the current test frequency setting, and then sends a control instruction to the electrochemical workstation based on the generated frequency adjustment amount to drive it to change the frequency scanning parameters. At the same time, it monitors the working status and feedback information of the electrochemical workstation.

[0036] The frequency control unit 3 includes a frequency strategy generation module 31. In the frequency strategy generation module 31, a specific algorithm for generating a frequency adjustment amount is as follows: The frequency strategy generation module 31 first receives the frequency priority weight information from the frequency priority determination module 22 ,in For the The priority weight of each frequency point reflects the urgency of the current electrochemical process for monitoring each frequency point. At the same time, it accurately obtains the test frequency set by the current electrochemical workstation These basic data are the basis for subsequent calculation of the adjustment range.

[0037] Calculate the adjustment amplitude for each frequency point , assuming the total frequency adjustment range coefficient is , which is pre-set according to the performance and test requirements of the electrochemical workstation. The calculation formula is: Here, the total adjustment range is reasonably allocated through priority weights, which means that frequency points with high priority will receive larger adjustment ranges, enabling them to focus more quickly on frequency areas where electrochemical processes change dramatically or are critical, thereby giving priority to capturing important information and improving test efficiency.

[0038] Calculate the new test frequency based on the adjustment amplitude and the current test frequency , the formula is ; This step intuitively combines the current frequency with the adjustment amplitude to obtain the adjusted target frequency, so that the frequency changes in a direction that is more in line with the needs of electrochemical process monitoring.

[0039] Make sure the new test frequency is within the frequency range allowed by the electrochemical workstation, that is, ,in and The lower and upper limits of the frequency are determined by the hardware characteristics of the electrochemical workstation. Through the above steps, the frequency strategy generation module 31 can generate a reasonable and effective frequency adjustment amount according to the dynamic requirements of the electrochemical process and the actual situation of the equipment, guide the electrochemical workstation to accurately adjust the test frequency, and enhance the ability of the electrochemical impedance spectroscopy test system to monitor and analyze the electrochemical process.

[0040] The frequency control unit 3 includes an instruction driving module 32. In the instruction driving module 32, a specific method for driving the electrochemical workstation to change the frequency scanning parameter according to the frequency adjustment amount is as follows: The instruction driving module 32 first accurately extracts the target frequency value of each frequency point from the frequency adjustment amount. And frequency sweep mode and sweep speed These parameters completely define the goals and methods of frequency adjustment. Accurately extracting them is the prerequisite for the subsequent smooth driving of the workstation, ensuring that the adjustment strategy can be accurately implemented and that the workstation clearly knows the frequency state to be achieved and how to achieve it.

[0041] Calculate from the current frequency Reach target frequency Time required Through this simple but critical calculation, the time process of frequency adjustment can be clearly planned, so that the entire frequency scanning parameter adjustment can be carried out in an orderly manner, avoiding incomplete data collection or slow device response caused by too fast adjustment.

[0042] According to the calculated time and target frequency , generating a frequency sweep parameter adjustment instruction sequence. This sequence is like a detailed operation guide, which specifies the frequency sweep parameter adjustment instruction sequence at each time step. It determines how the workstation should change the frequency. It provides guarantee for subsequent stable and accurate frequency adjustment, ensures that the workstation executes according to the predetermined plan, and realizes smooth transition from the current state to the target state.

[0043] In each time step, the current frequency value is calculated and updated by linear interpolation. , the formula is ;in is the frequency value at the current moment. The linear interpolation method is simple and effective. It can adjust the frequency evenly in each time period to make the frequency change continuous and stable, avoid the interference of sudden changes on the electrochemical test process, and ensure that the collected electrochemical impedance data is stable and reliable, truly reflecting the electrochemical process.

[0044] The instruction containing the updated frequency value and other relevant scanning parameters is sent to the electrochemical workstation through the communication interface. After receiving the instruction, the electrochemical workstation works according to the new frequency scanning parameters. At this time, the workstation accurately changes the frequency scanning behavior according to the instruction and starts to collect electrochemical impedance data at the new frequency, realizing dynamic monitoring of the electrochemical process and providing real-time and accurate data support for subsequent in-depth analysis of the electrochemical process.

[0045] During the test process after each frequency adjustment of the feedback regulation unit 4, the data acquisition unit 1 transmits the newly collected electrochemical impedance data to the deep learning analysis unit 2 in real time. The model predicts the impedance change trend that should be obtained in this test based on the current test frequency and analysis results, and compares it with the measured data, calculates the error index between the two, and triggers feedback regulation based on the error index.

[0046] The feedback adjustment unit 4 includes an error index calculation module 41. In the error index calculation module 41, a specific algorithm for calculating the error index between the two is as follows: Assume that the predicted impedance change trend is ,in represents a time series, which is based on the electrochemical process analysis model built and trained in the early stage. According to the current test frequency and the electrochemical process law that has been mastered, it makes an estimate of the upcoming impedance change. The measured impedance data is This is first-hand information reflecting the real state of the electrochemical system, which is obtained in real time by the data acquisition unit 1 during the actual test process. The clear definition of these two variables provides a clear operation object for the subsequent error calculation.

[0047] In the calculation process, the predicted value and the measured value are first aligned in time series. In actual testing, factors such as the starting time of data collection, sampling frequency, and possible transmission delays may cause the predicted value and the measured value to be misaligned on the time axis. Through a specific time synchronization algorithm or data processing flow, the two are adjusted to the same time point for difference calculation.

[0048] According to the root mean square error algorithm calculation formula The purpose of calculating the average value is to measure the average deviation between the predicted value and the measured value as a whole, eliminate the impact caused by the different number of data points, and obtain a representative intermediate indicator to reflect the average accuracy of the model's prediction over the entire time period. Finally, the average value is squared to obtain the final error indicator. The square root operation makes the dimension of the error indicator consistent with the original impedance data, which is convenient for intuitive understanding and application in subsequent decision-making, thereby providing a reliable basis for whether to start feedback regulation.

[0049] The feedback adjustment unit 4 includes a parameter adjustment module 42. In the parameter adjustment module 42, the parameter adjustment method involved in triggering feedback adjustment according to the error index is as follows: The error index calculation module 41 monitors and calculates the error index between the predicted impedance change trend and the measured impedance data in real time This indicator is a key quantitative value to measure the prediction accuracy of the model. It comprehensively reflects the degree of deviation of the model's understanding of the electrochemical process. Greater than the preset threshold When the system determines that the current model prediction deviates too much from the actual situation and needs to be adjusted urgently, the feedback adjustment process is started. Based on historical experience.

[0050] When feedback regulation is started, the deep learning analysis unit 2 responds and enables the adaptive learning rate adjustment method to optimize the model parameters. The learning rate of the current model is set to , dynamically adjust the learning rate according to the size of the error indicator, and the adjustment formula is ;in is the adjustment factor.

[0051] Similarly, the frequency adjustment range in the frequency adjustment amount is adjusted according to the size of the error index. , let the adjustment coefficient be , when the error index Threshold exceeded When , the expression of the new frequency adjustment range is The expanded adjustment range allows subsequent tests to reach more potential key frequencies, avoiding missing important electrochemical information due to the initial frequency setting being too narrow, and improving the comprehensiveness of the test.

[0052] First get the current frequency priority weight vector At the same time, the error value corresponding to each frequency point is obtained from the error index calculation module 41, and the frequency point index set with an error value greater than a certain proportion is found. , for the weights in the set , updated to ; For those not in the set Weight , updated to The weight increase allows subsequent tests to focus on these problem frequency points, improve test efficiency, quickly lock in key frequency changes in the electrochemical process, and enhance monitoring accuracy.

[0053] In the present invention, the electrochemical impedance spectrum and environmental data are collected by the data acquisition unit 1, and the deep learning analysis unit 2 uses a convolutional neural network and a long short-term memory network to build a model. After training, features are extracted, trends are analyzed, and dynamic frequency priority is sorted. The frequency control unit 3 drives the electrochemical workstation to adjust the frequency according to this weight generation strategy. The feedback adjustment unit 4 calculates the error between the predicted and measured impedance. When the threshold is exceeded, parameter correction is performed in each unit, and the test frequency is intelligently optimized according to the dynamic changes of the electrochemical process, so as to accurately capture key information, avoid waste of resources, and improve test efficiency and accuracy.

[0054] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An electrochemical impedance spectroscopy test system based on a multi-model combination, characterized in that: It comprises a data acquisition unit (1), a deep learning analysis unit (2), a frequency control unit (3) and a feedback adjustment unit (4); The data acquisition unit (1) uses an electrochemical workstation as an acquisition device and is equipped with a multi-channel data acquisition card to collect electrochemical impedance spectroscopy data, and obtains test environment related parameters through a multi-source sensor; The deep learning analysis unit (2) inputs the electrochemical impedance spectrum data into a network architecture based on a combination of a convolutional neural network and a long short-term memory network. The convolutional neural network performs a convolution operation on the input electrochemical impedance spectrum data to extract a local feature map of the electrochemical impedance spectrum data, and expands the feature map output by the convolutional neural network in a time series as an input of the long short-term memory network to obtain a dynamic change law in the time series, thereby constructing a primary electrochemical process analysis model. The primary electrochemical process analysis model is then trained using the electrochemical impedance spectrum data collected by the data acquisition unit (1) to obtain an electrochemical process analysis model. The electrochemical process analysis model then analyzes the electrochemical impedance spectrum data, determines the key stage and feature change trend of the current electrochemical process based on the analysis results and in combination with a built-in electrochemical knowledge rule base, and then assigns time-varying priority weights to different frequency points in the key stage using a dynamic frequency priority sorting algorithm. The frequency control unit (3) receives the frequency priority weight information, generates a frequency adjustment amount in combination with the currently set test frequency, calculates a new frequency based on the generated frequency adjustment amount, and sends a control instruction to the electrochemical workstation to drive it to change the scanning frequency, while monitoring the working state and feedback information of the electrochemical workstation; During each test after the frequency adjustment of the feedback adjustment unit (4), the data acquisition unit (1) transmits the newly acquired electrochemical impedance data to the deep learning analysis unit (2) in real time. The model predicts the impedance change trend that should be obtained in this test based on the current test frequency and analysis results, compares it with the actual measured data, calculates the error index between the two, and triggers feedback adjustment based on the error index.

2. The electrochemical impedance spectroscopy testing system based on multi-model combination according to claim 1, characterized in that: The deep learning analysis unit (2) comprises a model building training module (21), wherein the specific method for building an electrochemical process analysis model in the model building training module (21) is as follows: The architecture combines a convolutional neural network with a long short-term memory network, wherein the convolutional neural network is used to perform a convolution operation on the input electrochemical impedance spectroscopy data to extract the local feature map of the electrochemical impedance spectroscopy data; The feature map output by the convolutional neural network is expanded in time series as the input of the long short-term memory network. The long short-term memory network calculates according to the preset formula to obtain the dynamic change law and finally constructs the electrochemical process analysis model.

3. The electrochemical impedance spectroscopy testing system based on multi-model combination according to claim 2, characterized in that: The specific steps of the model building training module (21) using the trained model to perform feature extraction and analysis are as follows: The collected electrochemical impedance spectroscopy data is input into the trained model. After the forward propagation process of the model, the feature vector representation is obtained. For the convolutional neural network part, the feature vector is the output after multi-layer convolution and pooling operations. For the long short-term memory network part, the feature vector is the hidden state at the last moment. The principal component analysis algorithm is used to reduce the dimensionality of the eigenvector, calculate its covariance matrix, and perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The eigenvector corresponding to the largest eigenvalue is used as the new basis vector, and the original eigenvector is projected into the new basis vector space to obtain the eigenvector after dimensionality reduction; According to the eigenvector after dimensionality reduction, the clustering analysis algorithm is used to divide the electrochemical process into stages, set the number of cluster centers, and randomly initialize the cluster centers. The distance from each sample to the cluster center is calculated, and the sample is assigned to the class to which the nearest cluster center belongs. Then the cluster center is updated and the above process is repeated until the cluster center no longer changes, that is, the key stages and characteristic change trends of the current electrochemical process are determined.

4. The electrochemical impedance spectroscopy testing system based on multi-model combination according to claim 3, characterized in that: The deep learning analysis unit (2) comprises a frequency priority determination module (22), in which the specific implementation of the dynamic frequency priority sorting algorithm is as follows: According to the key stages and characteristic change trends of the electrochemical process obtained by analysis, the key frequency range associated with each stage and trend is determined; For each critical frequency range, an initial priority weight is first assigned, and then the priority weight is dynamically adjusted based on the real-time collected data and model analysis results; Set the current time to , the last adjustment time was During this period of time, if the impedance change rate within a certain critical frequency range exceeds a preset threshold , then increase its priority weight. If the impedance change rate is less than , then reduce its priority weight. If the impedance change rate is The priority weight remains unchanged.

5. The electrochemical impedance spectroscopy testing system based on multi-model combination according to claim 4, characterized in that: The frequency control unit (3) comprises a frequency strategy generation module (31), in which a specific algorithm for generating a frequency adjustment amount is as follows: Receive frequency priority weight information, obtain the test frequency currently set by the electrochemical workstation, and calculate the adjustment amplitude of each frequency point; A new test frequency is calculated according to the adjustment amplitude and the current test frequency, and the new test frequency is within the frequency range allowed by the electrochemical workstation.

6. The electrochemical impedance spectroscopy testing system based on multi-model combination according to claim 5, characterized in that: The frequency control unit (3) comprises an instruction driving module (32), wherein the specific method of driving the electrochemical workstation to change the frequency scanning parameter according to the frequency adjustment amount in the instruction driving module (32) is as follows: Extract the target frequency value of each frequency point, the frequency scanning mode and the scanning speed from the frequency adjustment amount, and calculate the time required to reach the target frequency from the current frequency; According to the calculated time and target frequency, a frequency scanning parameter adjustment instruction sequence is generated, and within each time step, the current frequency value is calculated and updated by linear interpolation; The instruction containing the updated frequency value and other related scanning parameters is sent to the electrochemical workstation through the communication interface, and the electrochemical workstation works according to the new frequency scanning parameters.

7. The electrochemical impedance spectroscopy testing system based on multi-model combination according to claim 6, characterized in that: The feedback adjustment unit (4) comprises an error index calculation module (41). In the error index calculation module (41), a specific algorithm for calculating the error index between the two is as follows: Assume that the predicted impedance change trend is ,in represents the time series, and the measured impedance data is , using the root mean square error algorithm to calculate the error index , the calculation formula is: ； in is the number of data points. In the calculation process, the predicted value and the measured value are first aligned in time series, that is, the difference is calculated at the same time point. Then, the sum of the squares of the differences at each time point is calculated in turn according to the formula, and the average is taken and the square root is taken to obtain the final error index.

8. The electrochemical impedance spectroscopy testing system based on multi-model combination according to claim 7, characterized in that: The feedback adjustment unit (4) comprises a parameter adjustment module (42). In the parameter adjustment module (42), a parameter adjustment method involved in triggering feedback adjustment according to an error index is as follows: When the calculated error index is greater than a preset threshold, feedback adjustment is initiated, and in the deep learning analysis unit (2), the model parameters are updated using an adaptive learning rate adjustment method; At the same time, in the frequency control unit (3), the frequency adjustment range and the priority weight in the frequency adjustment amount are corrected according to the size of the error index, and the priority weight is redistributed.

Images