An electrochemical energy storage power station fault diagnosis method, system, medium and product

Abstract

This invention relates to the field of electrochemical energy storage power station technology, and discloses a method, system, medium, and product for fault diagnosis of electrochemical energy storage power stations. This invention overcomes the limitations of traditional technologies that rely solely on single data such as battery voltage and current by acquiring a target multi-source dataset. Furthermore, it utilizes an improved DS evidence theory for data fusion, resolving the conflicts and uncertainties between multi-source heterogeneous data. Further, by determining multiple real-time health assessment indicators and evaluating the health of the electrochemical energy storage power station to be diagnosed using a target health assessment model, the accuracy and reliability of the health assessment results are improved. Finally, through a trained target fault diagnosis model based on support vector machines, the existence and specific type of fault in the power station can be quickly and accurately determined, solving the problem that traditional threshold alarms or pattern matching cannot accurately locate the fault type, thus improving fault handling efficiency.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage power station technology, specifically to a fault diagnosis method, system, medium, and product for an electrochemical energy storage power station. Background Technology

[0002] With the rapid development of the new energy industry, electrochemical energy storage power stations are increasingly widely used in power systems. Accurately assessing and diagnosing the health of electrochemical energy storage power stations is crucial for ensuring their safe and stable operation, extending their service life, and optimizing energy utilization efficiency. However, current technologies for assessing and diagnosing the health of electrochemical energy storage power stations have several shortcomings. On the one hand, the data sources relied upon for assessment and diagnosis are relatively limited, relying solely on basic data such as battery voltage and current, which cannot comprehensively reflect the actual operating status of the energy storage power station. On the other hand, the algorithms used are mostly traditional, single algorithms, which struggle to effectively handle complex and variable data, resulting in low accuracy and reliability of the assessment and diagnostic results. Summary of the Invention

[0003] This invention provides a method, system, medium, and product for fault diagnosis of electrochemical energy storage power stations, in order to solve the problems of low accuracy in assessment and diagnosis caused by single data sources and insufficient algorithm processing capabilities in the prior art.

[0004] In a first aspect, the present invention provides a fault diagnosis method for an electrochemical energy storage power station, the method comprising: A target multi-source dataset for the electrochemical energy storage power station to be diagnosed is obtained. This dataset includes data from multiple sensors, historical battery operation data, environmental temperature and humidity data, and power performance data. The target multi-source dataset is then fused using an improved DS evidence theory to obtain a real-time fused dataset. Based on this real-time fused dataset, multiple real-time health assessment indicators are determined. These indicators are then input into a target health assessment model to obtain multiple real-time health assessment results for the electrochemical energy storage power station to be diagnosed. Finally, these results are input into a target fault diagnosis model based on a support vector machine to obtain the fault diagnosis results for the electrochemical energy storage power station to be diagnosed.

[0005] The fault diagnosis method for electrochemical energy storage power stations provided by this invention overcomes the limitations of traditional technologies that rely solely on single data such as battery voltage and current by acquiring a multi-source dataset of the target electrochemical energy storage power station to be diagnosed. Furthermore, by utilizing an improved DS evidence theory for data fusion, it integrates effective information from different data sources, improving data credibility and consistency, eliminating the impact of errors from a single data source on subsequent analysis, and resolving conflicts and uncertainties between multi-source heterogeneous data. Furthermore, by determining multiple real-time health assessment indicators, it comprehensively reflects the overall health status of the electrochemical energy storage power station, avoiding the shortcomings of traditional single indicators that cannot cover complex operating scenarios, resulting in a more accurate health description. Furthermore, by evaluating the health of the electrochemical energy storage power station to be diagnosed using a target health assessment model, it improves the accuracy and reliability of the health assessment results. Simultaneously, the multiple real-time health assessment results provide a more intuitive way to judge the health level of the power station. Furthermore, by using a well-trained target fault diagnosis model based on support vector machines, it is possible to quickly and accurately determine whether a power plant has a fault and the specific type of fault. This solves the problem that traditional threshold alarms or pattern matching cannot accurately locate the fault type, thereby providing maintenance personnel with a clear fault indication, improving fault handling efficiency, and solving the problem of low assessment and diagnosis accuracy caused by single data sources and insufficient algorithm processing capabilities in existing technologies.

[0006] In one optional implementation, a target multi-source dataset of the chemical energy storage power station to be diagnosed is obtained, including: Obtain the initial multi-source dataset of the chemical energy storage power station to be diagnosed; perform outlier detection and processing on the initial multi-source dataset to obtain the first multi-source dataset; perform normalization processing on the first multi-source dataset to obtain the target multi-source dataset.

[0007] The fault diagnosis method for electrochemical energy storage power stations provided by this invention can identify and repair abnormal data caused by sensor failures, data transmission interference, etc., by performing outlier detection and processing on the initial multi-source dataset. This avoids the interference of abnormal data on subsequent fusion, evaluation, and diagnosis results, ensuring data quality and improving the reliability of subsequent analysis results. Furthermore, by normalizing the first multi-source dataset, the influence of numerical differences between data of different dimensions can be eliminated, ensuring that the weights of each data point are reasonable in subsequent fusion and model training processes. This avoids the model's over-reliance on a certain type of data due to different data dimensions, improving the fairness and accuracy of model analysis.

[0008] In one optional implementation, outlier detection and processing are performed on the initial multi-source dataset to obtain a first multi-source dataset, including: Based on the initial multi-source dataset, multiple local outliers are obtained through outlier detection model based on local outlier algorithm. Based on the multiple local outliers, abnormal multi-source datasets are identified in the initial multi-source dataset. The abnormal multi-source datasets are repaired using cubic spline interpolation method, and the first multi-source dataset is determined.

[0009] The fault diagnosis method for electrochemical energy storage power stations provided by this invention, through an outlier detection model based on the local outlier factor algorithm, can accurately identify outlier data points in the initial dataset. Compared with traditional simple threshold judgment, the local outlier factor algorithm, by calculating the distance difference between data points and neighboring data, can more sensitively and accurately discover hidden outlier data, reducing the possibility of missed or false positives for outlier data. Furthermore, by using multiple local outlier factors, it can distinguish between normal and outlier data in the initial multi-source dataset, providing a clear basis for subsequent targeted repair and preventing outlier data from being mixed with normal data and affecting the analysis results, thus ensuring the accuracy of data screening. Furthermore, by using cubic spline interpolation to repair the outlier multi-source dataset, it can more smoothly and accurately fill in the gaps in outlier data, restore the true trend of data change, avoid introducing new errors into the repaired data, and ensure the continuity and reliability of the data.

[0010] In one alternative implementation, the target multi-source dataset is fused using an improved DS evidence theory to obtain a fused dataset, including: Multiple battery health indicators are obtained; based on these indicators, historical data analysis results and expert experience are used to determine multiple basic probability assignment values ​​for the target multi-source dataset; based on these basic probability assignment values, the target multi-source dataset is fused using the synthesis rules of the improved DS evidence theory to obtain a fused dataset.

[0011] The fault diagnosis method for electrochemical energy storage power stations provided by this invention, by combining historical operating patterns and expert knowledge, assigns reasonable trust weights to different data sources. This ensures that indicators with a greater impact on health assessment have higher weights, improving the rationality of evidence synthesis and avoiding fusion bias caused by blind weight allocation. Furthermore, by using the synthesis rules of the improved DS evidence theory for data fusion, information from different weighted data sources can be effectively integrated. This overcomes the limitations of traditional DS evidence theory in handling highly conflicting data, improving the stability and credibility of the fusion results, and ensuring that the fused data more accurately reflects the battery health status.

[0012] In one alternative implementation, the method further includes: The process involves: acquiring a historical fusion dataset of the electrochemical energy storage power station to be diagnosed; determining multiple historical health assessment indicators based on the historical fusion dataset; training a hybrid neural network using the historical fusion dataset and multiple historical health assessment indicators to obtain an initial health assessment model, which includes a long short-term memory network, a convolutional neural network, a fully connected layer, and an output layer; and optimizing the initial health assessment model using a stochastic gradient descent algorithm based on mean squared error to obtain the target health assessment model.

[0013] The fault diagnosis method for electrochemical energy storage power stations provided by this invention acquires a historical fusion dataset of the electrochemical energy storage power station to be diagnosed, providing rich historical samples for training the health assessment model. This ensures that the model can learn comprehensive health change patterns and avoids the problem of poor model generalization ability due to insufficient samples. Furthermore, by constructing a multi-dimensional historical health indicator sample library, precise feature inputs are provided for model training, enabling the model to learn the correlation between various health indicators and the actual health status of the power station, thus improving the model's ability to judge health status. Furthermore, by combining the advantages of long short-term memory networks and convolutional neural networks, the model can simultaneously grasp the temporal and spatial characteristics of the data. Furthermore, by continuously adjusting the model parameters using the stochastic gradient descent algorithm, the error between the model's prediction results and the actual health status is reduced, improving the accuracy and stability of the model's assessment. This allows the final target health assessment model to more accurately output the power station's health results, meeting practical application needs.

[0014] In one alternative implementation, the method further includes: Acquire multiple historical health assessment results for the electrochemical energy storage power station to be diagnosed under multiple fault types; use multiple fault types as labels and the multiple historical health assessment results to train a support vector machine to obtain an initial fault diagnosis model based on the support vector machine; use radial basis functions as kernel functions and cross-validation method to optimize the initial fault diagnosis model based on the support vector machine to obtain a target fault diagnosis model based on the support vector machine.

[0015] The fault diagnosis method for electrochemical energy storage power stations provided by this invention constructs a sample library covering multiple fault scenarios, providing sufficient fault feature samples for the training of the support vector machine model. This ensures that the model can learn the health status change patterns corresponding to different faults, avoiding missed faults due to a single sample. Furthermore, training with multiple fault types as labels enables the support vector machine model to establish a correlation between health status assessment results and fault types, thus initially possessing fault identification capabilities and solving the problem that traditional methods struggle to establish a correspondence between health data and fault types. Furthermore, using radial basis functions as kernel functions effectively handles nonlinear data, improving the model's ability to fit complex fault features. Simultaneously, model optimization through cross-validation avoids overfitting or underfitting, resulting in a final support vector machine-based target fault diagnosis model that can more accurately and stably determine fault types, enhancing the practicality of fault diagnosis.

[0016] Secondly, the present invention provides an electrochemical energy storage power station fault diagnosis system, the system comprising: an electrochemical energy storage power station and a fault diagnosis module, the electrochemical energy storage power station comprising: a battery management system, an ambient temperature and humidity sensor and a power change system, and multiple sensors being installed in the battery compartment of the electrochemical energy storage power station; Multiple sensors are used to acquire multiple sensor data and send the multiple sensor data to the fault diagnosis module. The multiple sensor data includes battery voltage, current, and temperature. A battery management system is used to acquire battery operation history dataset and send the battery operation history dataset to the fault diagnosis module. The battery operation history dataset includes battery charge and discharge cycles, historical state of charge data, and historical health data. An environmental temperature and humidity sensor is used to acquire multiple environmental temperature and humidity data and send the multiple environmental temperature and humidity data to the fault diagnosis module. A power change system is used to acquire multiple power performance data and send the multiple power performance data to the fault diagnosis module. The fault diagnosis module is used to execute the electrochemical energy storage power station fault diagnosis method of the first aspect or any corresponding embodiment described above.

[0017] The electrochemical energy storage power station fault diagnosis system provided by this invention can directly capture the real-time status of the battery by collecting key operational data such as battery voltage, current, and temperature in real time, avoiding misjudgments caused by untimely or incomplete data collection. Furthermore, by acquiring historical information about the battery's long-term operation, it helps analyze the long-term evolution trend of battery health, providing a basis for judging the degree of battery aging and predicting future health status, thus compensating for the inability of real-time data to reflect historical changes. Furthermore, by supplementing data on the impact of environmental factors on battery health, it avoids the bias in health assessment caused by the traditional neglect of environmental factors. Furthermore, by acquiring information on the impact of grid interaction on the battery, it helps analyze the cumulative effect of grid dispatching behavior on power station health, enabling fault diagnosis to take into account grid-side factors, thereby improving the comprehensiveness and accuracy of diagnostic results. Furthermore, by combining the electrochemical energy storage power station with a fault diagnosis module executing the fault diagnosis method of the electrochemical energy storage power station described in the first aspect or any corresponding embodiment, it achieves systematic and automated fault diagnosis of the power station, improving the intelligent level of power station operation and maintenance, and ensuring the safe and stable operation of the power station.

[0018] In one optional implementation, the fault diagnosis module includes: a storage module and a display module; The storage module is used to store fault diagnosis results, multiple sensor data, multiple battery operation history data, multiple ambient temperature and humidity data, and multiple power performance data of the electrochemical energy storage power station; the display module is used to display multiple real-time health assessment results and fault diagnosis results of the electrochemical energy storage power station.

[0019] The electrochemical energy storage power station fault diagnosis system provided by this invention solves the problems of data loss and difficulty in traceability caused by the scattered storage of traditional energy storage power stations through its storage module. It achieves centralized management of multi-dimensional data, providing complete data support for subsequent review of fault causes and analysis of power station health trends. Furthermore, the stored data can serve as a sample basis for model iteration and optimization, helping to continuously improve the accuracy of the target health assessment model and the target fault diagnosis model. It also meets the compliance requirements for data retention, thus avoiding the impact of data loss on the entire lifecycle health management of the power station. Furthermore, the display module breaks through the limitations of traditional technologies that require complex data reports to interpret health and fault information, significantly improving information acquisition efficiency. Simultaneously, the visualization of real-time data and diagnostic results helps maintenance personnel to promptly identify power station health anomalies and faults, shortening fault response time and providing an intuitive basis for accurately formulating maintenance strategies, further ensuring the safe and stable operation of the power station.

[0020] Thirdly, the present invention provides a computer-readable storage medium storing computer instructions, which are used to cause a computer to execute the electrochemical energy storage power station fault diagnosis method of the first aspect or any corresponding embodiment described above.

[0021] Fourthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the electrochemical energy storage power station fault diagnosis method described in the first aspect or any corresponding embodiment. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first process of the fault diagnosis method for an electrochemical energy storage power station according to an embodiment of the present invention; Figure 3 This is a structural block diagram of an electrochemical energy storage power station fault diagnosis system according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0026] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0027] As an optional application scenario of this invention, considering the specific application environment architecture or specific hardware architecture upon which the electrochemical energy storage power station fault diagnosis method depends, the specific application environment architecture or specific hardware architecture is described here. For example... Figure 1 As shown, the architecture system may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0028] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0029] This invention provides a fault diagnosis method for electrochemical energy storage power stations. By acquiring multi-source datasets and combining them with an improved DS evidence theory, a target health assessment model, and a target fault diagnosis model based on support vector machines, the method can quickly and accurately determine whether a fault exists in the power station and the specific type of fault. This solves the problem of low accuracy in assessment and diagnosis caused by single data sources and insufficient algorithm processing capabilities in existing technologies.

[0030] According to an embodiment of the present invention, an embodiment of a fault diagnosis method for an electrochemical energy storage power station is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] This embodiment provides a fault diagnosis method for electrochemical energy storage power stations, which can be used in the aforementioned mobile terminals, such as mobile phones and tablets. Figure 2 This is a flowchart of a fault diagnosis method for an electrochemical energy storage power station according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the target multi-source dataset of the electrochemical energy storage power station to be diagnosed.

[0032] Specifically, the electrochemical energy storage power station to be diagnosed refers to a power station that uses electrochemical principles to store and release energy and currently requires health assessment and fault diagnosis. It can cover core components such as battery packs, battery management systems, and supporting sensors.

[0033] The target multi-source dataset represents a preprocessed collection of data related to the multi-dimensional operation of the electrochemical energy storage power station to be diagnosed. It may include preprocessed data from multiple sensors, historical battery operation data, environmental temperature and humidity data, and power performance data.

[0034] Among them, multiple sensor data can be collected in real time by various sensors deployed in key locations of the electrochemical energy storage power station to be diagnosed (such as near the battery compartment or battery cluster), which reflect the battery's operating status and may include battery voltage, current and temperature.

[0035] Furthermore, multiple battery operation history data refer to historical data recorded by the power station battery management system (BMS) that reflects changes in the long-term operating status of the battery, which may include the number of charge and discharge cycles, historical data on state of charge, historical data on health status, etc.

[0036] Furthermore, multiple environmental temperature and humidity data indicate that data collected by environmental temperature and humidity sensors deployed within the electrochemical energy storage power station to be diagnosed, reflecting the operating environment conditions of the power station, can be used to analyze the impact of environmental factors on battery health.

[0037] Furthermore, multiple power performance data represent data collected through the power conversion system (PCS) that reflect the operating status and performance level of the electrochemical energy storage power station to be diagnosed, and may include power data, output data, etc.

[0038] Furthermore, by acquiring the aforementioned multi-source datasets, the limitations of traditional energy storage power stations that only acquire single data such as battery voltage and current are overcome, thus avoiding analytical biases caused by incomplete data.

[0039] Step S202: The target multi-source dataset is fused using the improved DS evidence theory to obtain a real-time fused dataset.

[0040] Among them, the improved DS evidence theory represents an evidence fusion method that is based on the traditional DS evidence theory and optimized for the characteristics of multi-source heterogeneous data from electrochemical energy storage power stations.

[0041] Specifically, existing technologies rely on a single data source for assessing the health of energy storage power stations. Even when multiple data sources are introduced, uncertainties (such as data anomalies caused by a momentary malfunction of a temperature sensor) and conflicts (such as voltage data indicating "healthy" batteries, but gas sensor data indicating "potential risks") can easily arise due to differences in the characteristics of different data sources.

[0042] Traditional data fusion methods (such as simple weighted averaging and single threshold screening) cannot quantify this uncertainty or reasonably resolve data conflicts, easily leading to biased fusion results and affecting the accuracy of health assessment. In this embodiment of the invention, by introducing an improved DS evidence theory, effective information from different data sources can be integrated, improving data credibility and consistency, eliminating the impact of errors from a single data source on subsequent analysis, and resolving the conflict and uncertainty issues between multi-source heterogeneous data.

[0043] Step S203: Based on the real-time fusion dataset, determine multiple real-time health assessment indicators.

[0044] Specifically, based on the fused real-time dataset, battery voltage consistency, battery capacity decay rate, charge / discharge efficiency, and temperature uniformity can be used as health assessment indicators.

[0045] Among them, battery voltage consistency is measured by calculating the standard deviation of the voltage of each battery in the battery pack; battery capacity decay rate is calculated based on the ratio of the difference between the initial capacity and the current capacity to the initial capacity; charge and discharge efficiency is calculated based on the ratio of the amount of charge received to the amount of charge discharged; and temperature uniformity is measured by calculating the standard deviation of the temperature of each battery in the battery pack.

[0046] Furthermore, the real-time voltage data of all individual cells in the battery pack of the electrochemical energy storage power station to be diagnosed can be extracted from the real-time fused dataset. At the same time, the average voltage data of the battery pack at the current moment can be obtained through the statistical characteristics after data fusion. Then, the standard deviation of the voltage of each cell in the battery pack can be calculated using the following relationship (1): (1) In the formula: This represents the standard deviation of the voltage of each battery in the battery pack; Indicates the number of batteries; Indicates the first The voltage of each battery; This indicates the average voltage of the battery pack.

[0047] Furthermore, the standard deviation of the voltage of each battery in the battery pack can be calculated to measure the consistency of battery voltage and obtain the corresponding battery voltage consistency index data.

[0048] Furthermore, the initial capacity data of the electrochemical energy storage power station to be diagnosed, as well as the actual capacity data calibrated by the fusion data at the current moment, can be extracted from the real-time fused dataset. Then, the battery capacity decay rate can be calculated using the following relationship (2): (2) In the formula: Indicates the battery capacity decay rate; Indicates the current capacity; This indicates the initial capacity of the battery.

[0049] Furthermore, the charged energy data and discharged energy data of the electrochemical energy storage power station to be diagnosed in the current charge-discharge cycle can be extracted from the real-time fusion dataset, and the charge-discharge efficiency can be calculated using the following relationship (3): (3) In the formula: Indicates charge / discharge efficiency; Indicates the amount of electricity discharged; This indicates the amount of electricity charged.

[0050] Furthermore, the real-time temperature data of all individual cells in the battery pack of the electrochemical energy storage power station to be diagnosed can be extracted from the real-time fusion dataset at the current moment, and the average temperature data of the battery pack at the current moment can be obtained simultaneously. Then, the standard deviation of the temperature of each cell in the battery pack can be calculated using the following relationship (4): (4) In the formula: This represents the standard deviation of the temperature of each battery in the battery pack. Indicates the first The temperature of each battery; This indicates the average temperature of the battery pack.

[0051] Furthermore, temperature uniformity can be measured by calculating the standard deviation of the temperature of each battery in the battery pack, and the corresponding temperature uniformity index data can be obtained.

[0052] Furthermore, through the above process, multiple real-time health assessment indicators can be calculated.

[0053] Step S204: Input multiple real-time health assessment index data into the target health assessment model for processing to obtain multiple real-time health assessment result data of the electrochemical energy storage power station to be diagnosed.

[0054] Among them, the target health assessment model represents a hybrid neural network model that integrates long short-term memory networks and convolutional neural networks, used to assess the health of electrochemical energy storage power stations.

[0055] Multiple real-time health assessment results represent a set of quantitative results output by the target health assessment model, reflecting the health status of the electrochemical energy storage power station to be diagnosed in different dimensions or at different levels. The results are represented by a value from 0 to 100, with a higher value indicating a higher level of health.

[0056] Specifically, multiple real-time health assessment indicators (such as battery voltage consistency, capacity decay rate, charge and discharge efficiency, and temperature uniformity) are input into the target health assessment model. The feature processing and quantitative assessment capabilities of the target health assessment model can then be used to perform in-depth analysis and integration of the input data, thereby outputting multiple real-time health assessment results that reflect the health status of the electrochemical energy storage power station to be diagnosed in different dimensions or at different levels.

[0057] Step S205: Input multiple real-time health assessment results into the target fault diagnosis model based on support vector machine to obtain the fault diagnosis results of the electrochemical energy storage power station to be diagnosed.

[0058] Among them, the target fault diagnosis model based on support vector machine (SVM) represents a classification model built on the support vector machine algorithm, which is used to identify the fault types of electrochemical energy storage power stations.

[0059] Specifically, a target fault diagnosis model based on support vector machines can be used to match and classify multiple real-time health assessment results, ultimately outputting two types of core diagnostic results: one is the judgment of "whether a fault exists", that is, whether the current health status of the power station is abnormal; the other is the location of "specific fault type", such as determining whether the fault is battery overcharging, excessive temperature, or excessive capacity decay.

[0060] The fault diagnosis method for electrochemical energy storage power stations provided in this embodiment overcomes the limitations of traditional technologies that rely solely on single data such as battery voltage and current by acquiring a multi-source dataset of the target electrochemical energy storage power station to be diagnosed. Furthermore, by utilizing an improved DS evidence theory for data fusion, effective information from different data sources can be integrated, improving data credibility and consistency, eliminating the impact of errors from a single data source on subsequent analysis, and resolving conflicts and uncertainties between multi-source heterogeneous data. Furthermore, by determining multiple real-time health assessment indicators, the overall health status of the electrochemical energy storage power station can be comprehensively reflected, avoiding the shortcomings of traditional single indicators that cannot cover complex operating scenarios of the power station, resulting in a more accurate health description. Furthermore, by evaluating the health of the electrochemical energy storage power station to be diagnosed using a target health assessment model, the accuracy and reliability of the health assessment results are improved. Simultaneously, the multiple real-time health assessment results obtained make it easier to intuitively judge the health level of the power station. Furthermore, by using a well-trained target fault diagnosis model based on support vector machines, it is possible to quickly and accurately determine whether a power plant has a fault and the specific type of fault. This solves the problem that traditional threshold alarms or pattern matching cannot accurately locate the fault type, thereby providing maintenance personnel with a clear fault indication, improving fault handling efficiency, and solving the problem of low assessment and diagnosis accuracy caused by single data sources and insufficient algorithm processing capabilities in existing technologies.

[0061] In some optional implementations, step S201 above includes: Step S2011: Obtain the initial multi-source dataset of the chemical energy storage power station to be diagnosed.

[0062] The initial multi-source dataset refers to the unprocessed data set covering the multi-dimensional operation of the electrochemical energy storage power station to be diagnosed. For details and processes, please refer to the description of the target multi-source dataset in step S201 above, which will not be repeated here.

[0063] Step S2012: Perform outlier detection and processing on the initial multi-source dataset to obtain the first multi-source dataset.

[0064] Specifically, step S2012 includes: Step a1: Based on the initial multi-source dataset, multiple local outliers are obtained through outlier detection model based on the local outlier algorithm.

[0065] Among them, the Local Outlier Factor (LOF) algorithm represents a density-based outlier detection method. Its core is to determine whether a data point deviates from the normal data distribution by calculating the density difference between a data point and other data points in its neighborhood.

[0066] Furthermore, the local outlier factor is used to reflect the degree of anomaly of the data points.

[0067] Specifically, the neighborhood size is set to 10. Each data point in the initial multi-source dataset is taken as the target data point, and each neighborhood consists of 10 adjacent data points. The adjacency can be defined by time order or data correlation. For example, for battery voltage data collected in time series, the neighborhood consists of 5 data points before and after the target data point.

[0068] Furthermore, for each target data point, the distance to all data points in its neighborhood is calculated, and the "k-distance" (i.e., the distance to the 10th nearest data point in the neighborhood) is taken as the benchmark, so that the local reachability density of the target data point in the neighborhood can be calculated.

[0069] Furthermore, by comparing the local reachability density of the target data point with the average local reachability density of all data points in its neighborhood, the local outlier factor of the target data point is obtained. If the factor value is close to 1, it indicates that the target data point has the same density as the neighborhood data and is considered normal data; if the factor value is much greater than 1, it indicates that the density of the target data point is much lower than that of the neighborhood data, suggesting a possible anomaly.

[0070] Furthermore, the above calculation is performed on all data points in the initial multi-source dataset one by one, and finally the local outlier factor corresponding to each data point is obtained, that is, multiple local outlier factors.

[0071] Furthermore, compared with traditional threshold judgment methods (such as setting a fixed voltage range to identify anomalies), the LOF algorithm identifies anomalies through density differences, which can more accurately capture hidden abnormal data (such as single voltage jumps caused by instantaneous sensor failures, and abnormal temperature values ​​caused by data transmission interference), avoiding "missed judgments" or "false judgments" caused by fixed thresholds being unable to adapt to dynamic changes in data.

[0072] Step a2: Identify the anomalous multi-source dataset in the initial multi-source dataset based on multiple local outlier factors.

[0073] Specifically, an anomaly detection threshold can be set, and local outliers can be compared with the threshold one by one to filter out all data points whose local outliers are greater than the threshold.

[0074] Furthermore, the selected abnormal data points are categorized and organized according to data type (such as abnormal battery voltage data, abnormal temperature data, and abnormal power grid fluctuation data) to form an abnormal multi-source dataset containing all abnormal data from the initial multi-source dataset. Simultaneously, the acquisition time, data type, and corresponding sensor number of each abnormal data point can be recorded to facilitate subsequent repair and tracing.

[0075] Step a3: Use cubic spline interpolation to repair the abnormal multi-source dataset and determine the first multi-source dataset.

[0076] Among them, cubic spline interpolation is a method that uses normal data points before and after outlier data points to accurately fill in the gaps in outlier data by constructing a smooth polynomial curve.

[0077] Specifically, for each anomalous data point in the multi-source anomalous dataset, multiple normal data points adjacent to it in the time series are extracted to ensure that the extracted normal data points can reflect the changing trend of this data type.

[0078] Furthermore, by using time as the independent variable and data values ​​as the dependent variable, a cubic polynomial interpolation function can be constructed based on the extracted normal data points before and after the data points. This function satisfies the requirements of smooth connection between adjacent intervals and continuity of the second derivative. The cubic polynomial interpolation function can accurately fit the changing trend of normal data, avoiding trend deviations caused by simple linear interpolation.

[0079] Furthermore, by substituting the time parameter of the outlier data point into the cubic spline interpolation function, the corrected value of the outlier data point can be obtained. Then, by replacing the outlier data points in the initial multi-source dataset with the corrected values, and then fully integrating the corrected data with the normal data in the initial dataset, a first multi-source dataset without outlier data can be finally formed.

[0080] Furthermore, the data repaired using cubic spline interpolation better reflects the changing trends of the original data (such as the nonlinear variation of battery voltage with charging and discharging), avoiding the introduction of new errors into the repaired data and ensuring the continuity and authenticity of the data. Simultaneously, by repairing anomalous data rather than directly deleting it, information loss due to missing data is avoided, ensuring the integrity of the first multi-source dataset.

[0081] In some alternative implementations, when performing outlier detection on the initial multi-source dataset, for each data point, its distance to other data points in the neighborhood can also be calculated, and by comparing it with the global average distance, it can be determined whether the data point is an outlier.

[0082] Specifically, the neighborhood range of each data point in the initial multi-source dataset can be determined based on its data characteristics (such as time series and spatial correlation). Simultaneously, all data points of the same type as the target data point in the initial multi-source dataset are used as a global dataset.

[0083] Furthermore, for each data point, calculate its Euclidean distance to every data point in its neighborhood. For multi-dimensional data, Manhattan distance or cosine distance can also be used.

[0084] Furthermore, the average of all the calculated individual distances is taken to obtain the average distance between the target data point and the data in its neighborhood.

[0085] Furthermore, it is possible to traverse all data points in the global dataset and calculate the distance between any two data points. Further, the average of all pairwise distances within the global dataset can be taken to reflect the overall dispersion of this type of data. Furthermore, a threshold ratio of the neighborhood average distance to the global average distance can be set based on historical data validation and expert experience.

[0086] Furthermore, for a target data point, the ratio of its neighborhood average distance to the global average distance is calculated. If the calculated ratio is less than or equal to a ratio threshold, it indicates that the dispersion of the data point from its neighborhood data conforms to the overall global characteristics, and it is determined to be a normal data point; if the calculated ratio is greater than the ratio threshold, it indicates that the difference between the data point and its neighborhood data is much greater than the global average level, deviating from the normal distribution, and it is determined to be an abnormal data point.

[0087] Step S2013: Normalize the first multi-source dataset to obtain the target multi-source dataset.

[0088] Specifically, by normalizing the first multi-source dataset, data of different dimensions can be mapped to the same interval.

[0089] For example, firstly, the first multi-source dataset is divided into categories according to data type (such as battery voltage data, current data, temperature data, charge / discharge cycles data, ambient temperature and humidity data, and power grid fluctuation data). For each category of data, all data points are traversed to determine the minimum value of that category of data. With the maximum value .

[0090] Secondly, the min-max normalization method can be used to perform calculations on each data point in each data class to obtain the normalized data. Furthermore, min-max normalization calculations are performed on all data types to obtain the normalized results for each data class.

[0091] The minimum-maximum normalization method is shown in the following relation (5): (5) In the formula: This represents the normalized data; This represents the original data.

[0092] Finally, the normalization results of all types of data can be integrated according to the "data type-collection time" to ensure that the temporal sequence and correlation of the data are not lost, and ultimately form the corresponding target multi-source dataset.

[0093] In some optional implementations, step S202 above includes: Step S2021: Obtain multiple battery health indicators.

[0094] Specifically, multiple battery health indicators are used to intuitively define the current performance level, risk level, and maintenance needs of the battery. These can include health indicators that reflect stable battery performance and no potential risks, sub-health indicators that reflect slight performance degradation and potential risks, and failure indicators that reflect severe performance degradation and inability to operate normally.

[0095] For example, obtain multiple battery health indicators .in, This indicates a health indicator, reflecting that the battery capacity and cycle life are normal and there is no abnormal degradation; This indicates a sub-healthy state, reflecting a capacity decay of 10%-20%, requiring attention to charging strategies. This indicates a failure indicator, reflecting a capacity degradation exceeding 20%, requiring battery replacement.

[0096] Step S2022: Based on multiple battery health indicators, and utilizing historical data analysis results and expert experience, determine multiple basic probability assignment values ​​for the target multi-source dataset.

[0097] In this context, the Basic Probability Assignment (BPA) represents the quantitative representation of the trust level of each data source in the improved DS evidence theory, which is the "trust level assignment" of each data source (evidence) to a subset in the framework. The sum of the BPAs of all subsets is 1, and the empty set BPA is 0.

[0098] Specifically, based on battery health indicators, combined with the patterns presented by historical data and the experience of experts in the battery field, a basic probability assignment value (BPA) can be assigned to different data in the target multi-source dataset, thereby quantifying the degree of trust of each data point in different battery health states.

[0099] For example, firstly, historical data is analyzed to statistically determine the correlation between different health indicators and the actual health status of the battery (such as the occurrence of faults and the degree of capacity degradation), and the contribution weight of each data point to different health statuses is initially determined. Then, battery experts are invited to revise the initial weights based on their professional experience, and finally, a corresponding BPA is assigned to each data source (such as voltage sensor data, temperature sensor data, etc.).

[0100] Furthermore, through the above process, the trust level of different data sources is differentiated and quantified, avoiding the defects of equal weighting in traditional data fusion. This allows data that is more important for battery health assessment to have higher trust levels, making the basis for subsequent data fusion more scientific and reasonable, and improving the reliability of the fusion results.

[0101] In an alternative embodiment, it is assumed that after data preprocessing from the three sensors, the output BPA (BPA(A) represents the degree of confidence in "state A") is shown in Table 1 below.

[0102] Table 1, BPA

[0103] Here, m(Θ) represents the degree of "uncertainty," such as a voltage sensor having a degree of 0.1 of uncertainty about the specific state, rather than distrusting any state.

[0104] Step S2023: Based on multiple basic probability assignment values, the target multi-source dataset is fused using the synthesis rules of the improved DS evidence theory to obtain a fused dataset.

[0105] Among them, the improved Dempster evidence theory synthesis rule represents the evidence integration logic optimized for the conflict and uncertainty of multi-source heterogeneous data (such as battery voltage, ambient temperature and humidity, and power grid fluctuation data) based on the traditional Dempster synthesis rule. Its core is to integrate the basic probability allocation (BPA) evidence bodies corresponding to multiple data sources into a single, credible fused evidence through conflict quantification, conflict handling, weighted synthesis, and result normalization.

[0106] Specifically, the data (evidence) in the target multi-source dataset are fused pairwise using the synthesis rules of the improved DS evidence theory, and the corresponding fused dataset is finally obtained. Its core is "calculating the conflict coefficient → normalizing the fusion result".

[0107] In some alternative embodiments, for Table 1 above, evidence 1 and evidence 2 are first fused together, and then the result is fused together with evidence 3.

[0108] 1. First step of fusion: Fuse evidence 1 (voltage) + evidence 2 (temperature).

[0109] First, calculate the conflict coefficient. (Quantifying the degree of contradiction between the two pieces of evidence). The conflict coefficient formula is shown in the following relation (6): (6) In the formula: Indicates the conflict coefficient; Indicates the summation symbol; Indicates the subscript, i.e. Indicates a conditional subscript for summation; The symbol for the intersection of sets; Symbol for the empty set; Represents the multiplication symbol; and middle, and The subscript indicates the basic probability assignment value, which is used to distinguish different basic probability assignment values. That is, the above relation (6) represents the sum of the BPA products of all subset pairs whose intersection is an empty set. Among them, the empty set represents two pieces of evidence that are completely contradictory, such as , .

[0110] Furthermore, the specific calculation process includes: (1) Contradictory combination 1: ; (2) Contradictory combination 2: ; (3) Contradictory combination 3: ; (4) Contradictory combination 4: ; (5) Contradictory combination 5: ; (6) Contradictory combination 6: .

[0111] Furthermore, by summing, we can obtain: .

[0112] Furthermore, This indicates that evidence 2 and evidence 3 do not completely conflict and can be integrated.

[0113] Secondly, calculate the fused BPA ( The fusion formula (normalized) is shown in the following relation (7): (7) in, Representation Frame Any subset of; denominator This represents the normalization coefficient.

[0114] Further, calculate separately , , , : (1) :all The sum of the products of ... , , Furthermore, the sum of the products = 0.5 × 0.2 + 0.5 × 0.1 + 0.1 × 0.2 = 0.1 + 0.05 + 0.02 = 0.17, from which we can calculate... ; (2) The combinations that meet the criteria are: , , Furthermore, the sum of the products = 0.3 × 0.5 + 0.3 × 0.1 + 0.1 × 0.5 = 0.15 + 0.03 + 0.05 = 0.23, from which we can calculate... ; (3) The combinations that meet the criteria are: , , Furthermore, the sum of the products = 0.1 × 0.2 + 0.1 × 0.1 + 0.1 × 0.2 = 0.02 + 0.01 + 0.02 = 0.05, from which we can calculate... ; (4) The combinations that meet the criteria are: Furthermore, the sum of the products = 0.1 × 0.1 = 0.01, from which we can calculate... .

[0115] Finally, we can obtain fusion result 1: .

[0116] 2. Second step of fusion: Fusion result 1 ( Evidence 3 (Gas): Repeat the above steps, first calculating the conflict factor. Then calculate the final fusion BPA ( ).

[0117] Specifically, the conflict coefficient is calculated as shown in the following equation (8): (8) Furthermore, the sum of the products of contradictory combinations is: 0.37×0.3+0.37×0.4+0.5×0.1+0.5×0.4+0.11×0.1+0.11×0.3+0.02×0.1+0.02×0.3=0.435.

[0118] Furthermore, the normalization coefficient = 1 - 0.435 = 0.565. Then calculate separately... , , , : (1) : (0.37×0.1+0.37×0.2+0.02×0.1) / 0.565≈0.18; (2) : (0.5×0.3+0.5×0.2+0.37×0.3+0.02×0.3) / 0.565≈0.55; (3) : (0.11×0.4+0.11×0.2+0.5×0.4+0.02×0.4) / 0.565≈0.26; (4) : (0.02×0.2) / 0.565≈0.01.

[0119] Further, calculate the final fused BPA: .

[0120] Furthermore, DS decision-making typically employs the "maximum confidence principle": selecting the state with the highest BPA value, while also ensuring a "significant difference from the second-highest value" (to avoid ambiguity). In this embodiment, the BPA ranking for each state is: m_final(S2) = 0.55 (highest) > m_final(S3) = 0.26 (second highest) > m_final(S1) = 0.18 > m_final(Θ) = 0.01. This means that the confidence level of S2 (0.55) is much higher than that of the other states, and the "uncertainty" confidence level (0.01) is extremely low, indicating that the result is reliable.

[0121] Furthermore, the final diagnostic conclusion can be obtained: the current health status of the battery is "sub-healthy (S2)", and the charging cut-off voltage needs to be adjusted and the fast charging frequency reduced to slow down the degradation.

[0122] In some optional implementations, the target health assessment model in step S204 above is obtained through the following steps: Step b1: Obtain the historical fusion dataset of the electrochemical energy storage power station to be diagnosed.

[0123] For the specific process, please refer to the detailed acquisition process of the real-time fused dataset in step S202 above, which will not be repeated here.

[0124] Step b2: Based on the historical fusion dataset, determine multiple historical health assessment indicator data.

[0125] For details on the specific process, please refer to the detailed acquisition process of multiple real-time health assessment indicator data in step S203 above, which will not be repeated here.

[0126] Step b3: Using the historical fusion dataset and multiple historical health assessment index data, train the hybrid neural network to obtain the initial health assessment model.

[0127] Hybrid neural networks include Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), fully connected layers, and output layers.

[0128] Specifically, by using a historical fusion dataset (containing features derived from multi-source data fusion) as input and corresponding historical health assessment index data (such as historical health measurement results) as output labels, a hybrid neural network consisting of a Long Short-Term Memory (LSTM) network, a Convolutional Neural Network (CNN), fully connected layers, and an output layer is iteratively trained. This allows the model to grasp the mapping relationship between the input data and the health assessment results, ultimately generating an initial health assessment model with basic assessment capabilities. LSTM can learn the temporal dynamic features of the data; CNN can mine the spatial features of the data.

[0129] For example, the acquired historical fusion dataset is divided into a training set (70%-80%) and a validation set (10%-15%) according to time series or working condition type. At the same time, multiple historical health assessment index data (such as health values ​​of 0-100 at various historical moments, and healthy / sub-healthy / failure classification labels) are used as the true value labels of the model output and associated with the training set and validation set data one by one.

[0130] Furthermore, a hybrid neural network is constructed: (1) LSTM layer: Input the time series features in the historical fusion dataset (such as voltage changes and capacity decay trend data for one consecutive hour), set 2-3 layers of LSTM units, and capture the long-term dependencies of the data (such as the cumulative impact of charge and discharge cycles on health) through the gating mechanism. (2) CNN layer: Receive the features output by the LSTM layer or directly input spatial correlation features (such as the temperature distribution data of each cell in the battery cluster), set 1-2 convolutional layers (using 3×3 convolutional kernels) and pooling layers to extract local spatial features (such as the correlation between temperature hotspots and health status). (3) Fully connected layer: After flattening the features output by the LSTM layer and CNN layer, connect 2-3 fully connected layers (the number of neurons decreases layer by layer, such as 128→64→32) to achieve deep fusion and non-linear transformation of features; (4) Output layer: Use a linear activation function (for predicting health scores) or a softmax activation function (for classifying health status) to output prediction results that are consistent with the dimensions of historical health assessment index data (such as health scores of 0-100).

[0131] Furthermore, set the training parameters: number of iterations (e.g., 100-200 rounds), batch size (e.g., 32-64), and initial learning rate (e.g., 0.001).

[0132] Furthermore, the training set data is input into the hybrid neural network, features are calculated layer by layer, and corresponding health assessment predictions are output. Further, the mean squared error (MSE) is used as the loss function to calculate the error between the predicted value and the true label. This error is then propagated back to each layer using a chain rule, updating the gating parameters of the LSTM, the convolutional kernel parameters of the CNN, and the weights of the fully connected layers.

[0133] Furthermore, after each training round, the model performance is evaluated using a validation set. If the validation set loss no longer decreases, training is stopped early to avoid overfitting. Finally, after training is complete, the current model parameters are saved, resulting in an initial health assessment model with basic predictive capabilities.

[0134] Step b4: Based on the mean square error, the initial health assessment model is optimized using the stochastic gradient descent algorithm to obtain the target health assessment model.

[0135] Among them, the Stochastic Gradient Descent (SGD) algorithm is an algorithm used to optimize the parameters of a model. It updates the parameters by randomly selecting a small number of data points to calculate gradients, thereby speeding up training and adapting to large-scale datasets.

[0136] Specifically, using mean squared error (MSE) as the optimization objective, the parameters (such as weights and biases) of the initial health assessment model are iteratively adjusted through the stochastic gradient descent (SGD) algorithm to reduce the deviation between the model's predicted values ​​and the actual health assessment results, so that the model converges to a better state, and ultimately a target health assessment model with higher accuracy and stronger generalization ability can be generated.

[0137] For example, with mean squared error (MSE) as the optimization objective, stochastic gradient descent (SGD) algorithm is selected as the optimizer, and gradient is calculated by randomly sampling training samples to achieve efficient parameter updates.

[0138] Furthermore, setting the parameters for the stochastic gradient descent (SGD) algorithm can include: (1) Learning rate: Initially set to 0.001, and adopt a learning rate decay strategy (e.g., decay to 0.9 every 10 rounds) to avoid parameter oscillation due to excessively large learning rate or slow convergence due to excessively small learning rate; (2) Momentum: Set to 0.9 (recommended value in the document) to accelerate the convergence of parameters in the correct direction by accumulating the gradient directions of the previous rounds and reduce oscillations; (3) Weight decay: set to 1e-5 to suppress excessively large model parameters and alleviate overfitting.

[0139] Furthermore, a batch of samples (e.g., 32 samples) is randomly selected from the training set and input into the initial health assessment model to calculate the MSE loss between the predicted and actual values. Simultaneously, the gradient of the loss function with respect to the parameters of each layer (LSTM weights, CNN convolutional kernels, and fully connected layer weights) can be calculated based on the SGD algorithm.

[0140] Furthermore, the parameters can be updated based on the gradient direction and learning rate, and the update direction can be adjusted in conjunction with the momentum term. Further, the above process is repeated, calculating the MSE of the validation set after each iteration. When the MSE of the validation set does not decrease for 10 consecutive iterations or reaches the maximum number of iterations (e.g., an additional 50 iterations), optimization is stopped, and the optimized model parameters are saved. This yields a target health assessment model with higher prediction accuracy and stronger generalization ability, whose MSE is reduced by more than 30% compared to the initial health assessment model.

[0141] In some optional implementations, the target fault diagnosis model based on support vector machines in step S205 above is obtained through the following steps: Step c1: Obtain multiple historical health assessment results for the electrochemical energy storage power station to be diagnosed under multiple fault types.

[0142] For the specific process, please refer to the detailed acquisition process of multiple real-time health assessment results data in step S204 above, which will not be repeated here.

[0143] Step c2: Using multiple fault types as labels, the support vector machine is trained using multiple historical health assessment results to obtain an initial fault diagnosis model based on the support vector machine.

[0144] Specifically, by using historical health assessment data reflecting the quantitative characteristics of battery health status as input samples and corresponding fault types (such as battery cell short circuit, electrolyte leakage, sensor failure, etc.) as classification labels, the support vector machine (SVM) is trained to learn the mapping relationship between health features and fault types, thereby generating an initial model with basic fault classification capabilities.

[0145] For example, the acquired historical health assessment results can be organized into a feature vector, which may include historical health values, quantitative indicators of health status (healthy / sub-healthy / failed), and statistical characteristics of key parameters (such as voltage fluctuation amplitude and temperature anomaly frequency). Simultaneously, multiple fault types (such as battery cell faults, cluster-level equalization faults, and grid interaction faults) can be mapped to discrete labels.

[0146] Furthermore, the sample set is divided into a training set (for model learning) and a test set (for preliminary validation) in a 7:3 ratio, and the feature vectors are standardized (e.g., mapped to the [0, 1] interval) to avoid the difference in feature dimensions affecting the kernel function calculation of SVM.

[0147] Furthermore, since the fault types are multi-category (such as 3 or more types of faults), a one-to-one or one-to-many strategy can be used to transform the multi-classification problem into a binary classification problem.

[0148] Furthermore, the training set features and corresponding labels are input into the SVM. The SVM can then maximize the sample margins for different fault types by finding the optimal hyperplane (a linear boundary separating different fault types in a high-dimensional space). Further, the hyperplane parameters can be determined by optimizing the objective function. After training, the model parameters (hyperplane coefficients, support vectors) are saved, resulting in an initial fault diagnosis model capable of preliminary classification of fault types.

[0149] Step c3: Using the radial basis function as the kernel function, the initial fault diagnosis model based on support vector machine is optimized using the cross-validation method to obtain the target fault diagnosis model based on support vector machine.

[0150] The radial basis function (RBF) is a real-valued function that takes the distance from a sample point to a center point as its independent variable. Its function value depends only on the Euclidean distance between the input vector and the center point, and does not depend on the direction of the input vector. It has the property of radial symmetry.

[0151] Cross-validation is a model validation method used to evaluate the performance of machine learning models and avoid overfitting. Its core logic is to divide the limited training dataset into multiple mutually exclusive subsets and fully utilize the data for model training and performance testing through a "training-validation" loop, so as to obtain more reliable model evaluation results and help select the optimal model parameters.

[0152] Specifically, by introducing radial basis function (RBF) as the kernel function to solve nonlinear classification problems, and using cross-validation to select the optimal parameters (penalty coefficient and kernel function parameters), the risk of model overfitting can be reduced, the generalization ability to unknown faults can be improved, and ultimately a high-precision target fault diagnosis model can be generated.

[0153] For example, a radial basis function (RBF) can be used as the kernel function, and the search range of the parameters to be optimized can be set, which may include the range of penalty coefficients and the range of kernel parameters.

[0154] Furthermore, a 5-fold cross-validation method is adopted, in which the training set is randomly divided into 5 subsets. Each time, 4 subsets are used as training data and 1 subset is used as validation data, and this is repeated 5 times to ensure that each subset is used as a validation set once.

[0155] Furthermore, within the preset range of penalty coefficients and kernel parameters, the SVM model is trained for each set of parameters by combining parameters through grid traversal (e.g., adjusting parameter values ​​by powers of 2 each time), and the average accuracy of 5-fold cross-validation is calculated.

[0156] Furthermore, the parameter combination with the highest cross-validation accuracy is selected as the optimal parameter, and then the support vector machine is retrained using the optimal parameter. All training set data is input to determine the final hyperplane parameters and support vectors.

[0157] Furthermore, the model performance is verified using a test set. If the accuracy of the test set is improved by more than 10% compared with the initial fault diagnosis model based on support vector machines, the model is saved to obtain the target fault diagnosis model based on support vector machines.

[0158] This embodiment also provides a fault diagnosis system for electrochemical energy storage power stations, such as... Figure 3 As shown, the electrochemical energy storage power station fault diagnosis system 1 includes an electrochemical energy storage power station 11 and a fault diagnosis module 12.

[0159] The electrochemical energy storage power station 11 includes a battery management system 111, an ambient temperature and humidity sensor 112, and a power change system 113. The battery compartment of the electrochemical energy storage power station 11 is equipped with multiple sensors, including voltage sensors, current sensors, temperature sensors, gas sensors, sound sensors, etc.

[0160] Furthermore, the fault diagnosis module 12 includes a storage module 121 and a display module 122.

[0161] Specifically, multiple sensors are used to acquire multiple sensor data and send the multiple sensor data to the fault diagnosis module 12. Among them, the multiple sensor data include the battery voltage, current and temperature.

[0162] Furthermore, the battery management system 111 is used to acquire the battery operation history dataset and send the battery operation history dataset to the fault diagnosis module 12.

[0163] The battery operation history dataset can include the number of charge and discharge cycles, historical state of charge data, and historical health status data.

[0164] Furthermore, the ambient temperature and humidity sensor 112 is used to acquire multiple ambient temperature and humidity data and send the multiple ambient temperature and humidity data to the fault diagnosis module 12.

[0165] Furthermore, the power change system 113 is used to acquire multiple power performance data and send the multiple power performance data to the fault diagnosis module 12.

[0166] The process of obtaining the different data mentioned above can be referred to the relevant description of step S201 above, and will not be repeated here.

[0167] Furthermore, the fault diagnosis module 12 is used to execute the electrochemical energy storage power station fault diagnosis method provided in the above embodiments of the present invention.

[0168] Furthermore, the storage module 121 is used to store the fault diagnosis results of the electrochemical energy storage power station 11, multiple sensor data, multiple battery operation history data, multiple ambient temperature and humidity data, and multiple power performance data.

[0169] Furthermore, the storage module solves the problems of data loss and difficulty in traceability caused by the scattered storage of traditional energy storage power stations, enabling centralized management of multi-dimensional data. This provides complete data support for subsequent review of fault causes and analysis of power station health trends. Moreover, the stored data can serve as a sample basis for model iteration and optimization, helping to continuously improve the accuracy of target health assessment and target fault diagnosis models. Simultaneously, it meets the compliance requirements for data retention, thereby preventing data loss from affecting the entire lifecycle health management of the power station.

[0170] Display module 122 is used to display multiple real-time health assessment results and fault diagnosis results of electrochemical energy storage power station 11.

[0171] Furthermore, the display module breaks through the limitations of traditional technologies that require complex data reports to interpret health and fault information, significantly improving information acquisition efficiency. Simultaneously, the visualization of real-time data and diagnostic results helps maintenance personnel promptly identify power plant health anomalies and faults, shortening fault response time and providing an intuitive basis for accurately formulating maintenance strategies, further ensuring the safe and stable operation of the power plant.

[0172] The electrochemical energy storage power station fault diagnosis system provided in this embodiment, by combining the electrochemical energy storage power station and the fault diagnosis module that executes the electrochemical energy storage power station fault diagnosis method of the first aspect or any corresponding embodiment described above, realizes the systematic and automated diagnosis of power station faults, improves the intelligent level of power station operation and maintenance, and ensures the safe and stable operation of the power station.

[0173] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0174] The following is a detailed reference. Figure 4This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0175] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0176] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the electrochemical energy storage power station fault diagnosis method of the embodiments of the present invention.

[0177] Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0178] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the electrochemical energy storage power station fault diagnosis method shown in the above embodiments is implemented.

[0179] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0180] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A fault diagnosis method for an electrochemical energy storage power station, characterized in that, The method includes: Obtain the target multi-source dataset of the electrochemical energy storage power station to be diagnosed. The target multi-source dataset includes multiple sensor data, multiple battery operation history data, multiple environmental temperature and humidity data, and multiple power performance data. The target multi-source dataset is fused using an improved DS evidence theory to obtain a real-time fused dataset; Based on the real-time fusion dataset, multiple real-time health assessment indicators are determined; The multiple real-time health assessment index data are input into the target health assessment model for processing to obtain multiple real-time health assessment result data of the electrochemical energy storage power station to be diagnosed. The multiple real-time health assessment results are input into the target fault diagnosis model based on support vector machine to obtain the fault diagnosis results of the electrochemical energy storage power station to be diagnosed.

2. The method according to claim 1, characterized in that, Obtain the target multi-source dataset of the chemical energy storage power station to be diagnosed, including: Obtain the initial multi-source dataset of the chemical energy storage power station to be diagnosed; Anomaly detection and processing are performed on the initial multi-source dataset to obtain the first multi-source dataset; The first multi-source dataset is normalized to obtain the target multi-source dataset.

3. The method according to claim 2, characterized in that, Outlier detection and processing are performed on the initial multi-source dataset to obtain a first multi-source dataset, including: Based on the initial multi-source dataset, multiple local outliers are obtained through outlier detection model based on the local outlier algorithm. Based on the multiple local outlier factors, an anomalous multi-source dataset is identified in the initial multi-source dataset; The abnormal multi-source dataset is repaired using cubic spline interpolation, and the first multi-source dataset is determined.

4. The method according to claim 1, characterized in that, The target multi-source dataset is fused using an improved DS evidence theory to obtain a fused dataset, including: Obtain multiple battery health metrics; Based on the aforementioned multiple battery health indicators, and utilizing historical data analysis results and expert experience, multiple basic probability allocation values ​​for the target multi-source dataset are determined. Based on the aforementioned multiple basic probability assignment values, the target multi-source dataset is fused using the synthesis rules of the improved DS evidence theory to obtain the fused dataset.

5. The method according to claim 1, characterized in that, The method further includes: Obtain the historical fusion dataset of the electrochemical energy storage power station to be diagnosed; Based on the aforementioned historical fusion dataset, multiple historical health assessment indicators were determined; Using the historical fusion dataset and the multiple historical health assessment index data, a hybrid neural network is trained to obtain an initial health assessment model. The hybrid neural network includes a long short-term memory network, a convolutional neural network, a fully connected layer, and an output layer. Based on the mean squared error, the initial health assessment model is optimized using the stochastic gradient descent algorithm to obtain the target health assessment model.

6. The method according to claim 1, characterized in that, The method further includes: Obtain multiple historical health assessment results for the electrochemical energy storage power station to be diagnosed under multiple fault types; Using the multiple fault types as labels, the support vector machine is trained using the multiple historical health assessment results data to obtain an initial fault diagnosis model based on the support vector machine. Using the radial basis function as the kernel function, the initial fault diagnosis model based on support vector machine is optimized using the cross-validation method to obtain the target fault diagnosis model based on support vector machine.

7. A fault diagnosis system for an electrochemical energy storage power station, characterized in that, The system includes an electrochemical energy storage power station and a fault diagnosis module. The electrochemical energy storage power station includes a battery management system, an ambient temperature and humidity sensor, and a power change system. Multiple sensors are installed in the battery compartment of the electrochemical energy storage power station. The plurality of sensors are used to acquire multiple sensor data and send the multiple sensor data to the fault diagnosis module. The multiple sensor data includes the battery voltage, current and temperature. The battery management system is used to acquire battery operation history dataset and send the battery operation history dataset to the fault diagnosis module. The battery operation history dataset includes the number of charge and discharge cycles, historical state of charge data, and historical health data of the battery. The ambient temperature and humidity sensor is used to acquire multiple ambient temperature and humidity data and send the multiple ambient temperature and humidity data to the fault diagnosis module; The power change system is used to acquire multiple power performance data and send the multiple power performance data to the fault diagnosis module; The fault diagnosis module is used to execute the fault diagnosis method for electrochemical energy storage power stations as described in any one of claims 1 to 6.

8. The system according to claim 7, characterized in that, The fault diagnosis module includes: a storage module and a display module; The storage module is used to store the fault diagnosis results of the electrochemical energy storage power station, the data of the multiple sensors, the historical operating data of the multiple batteries, the environmental temperature and humidity data, and the power performance data. The display module is used to display multiple real-time health assessment results and fault diagnosis results of the electrochemical energy storage power station.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the fault diagnosis method for an electrochemical energy storage power station as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions, which are used to cause a computer to execute the fault diagnosis method for an electrochemical energy storage power station as described in any one of claims 1 to 6.

Images