Battery detection method based on multi-dimensional data analysis
Through multi-dimensional data analysis and timing attenuation model, the problem of battery detection data decay over time is solved, the accuracy and reliability of battery health status detection is improved, and high-quality battery health indicators are generated.
Patent Information
- Application Number
- CN202410956083.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-17
AI Technical Summary
The existing battery health status detection methods do not fully consider the attenuation of the detection data over time, resulting in low detection accuracy.
Using a multi-dimensional data analysis method, the multi-dimensional detection sample set is obtained and classified into active and passive detection sample sets, a time-sequence decay model is established for attenuation analysis, training the detection effectiveness evaluation model, and optimizing battery health indicators with real-time detection data .
By considering time factors, the accuracy and reliability of battery health status detection are improved, and high-quality battery health indicators are generated.
Smart Images

Figure CN118501721B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of battery detection, and in particular to a battery detection method based on multi-dimensional data analysis. Background Art
[0002] With the continuous development of battery technology and the continuous expansion of its application fields, it is becoming increasingly important to accurately detect the health status of batteries. The detection of battery health status is of great significance for ensuring the safe use of batteries, extending the service life of batteries, and optimizing battery energy management strategies. In the prior art, battery health status detection is mainly achieved by analyzing various test data of the battery, including data in multiple dimensions such as voltage, current, and temperature. However, the existing battery detection methods do not fully consider the attenuation of test data over time. Directly using test data affected by time to judge the health status of the battery will result in results that deviate from the actual situation, resulting in low accuracy of battery health status detection. Summary of the invention
[0003] The present application provides a battery detection method based on multi-dimensional data analysis, aiming to solve the technical problem in the prior art that the accuracy of battery health status detection is low due to insufficient consideration of the attenuation of battery detection data over time.
[0004] The battery detection method based on multi-dimensional data analysis disclosed in the present application includes: obtaining a multi-dimensional detection sample set of a target battery; classifying the multi-dimensional detection sample set to obtain an active detection sample set and a passive detection sample set; establishing a timing decay model, performing timing decay analysis on the active detection sample set and the passive detection sample set respectively according to the timing decay model, and obtaining a first decay rate sample and a second decay rate sample, wherein the timing decay model includes an active-timing decay layer and a passive-timing decay layer; training a detection effectiveness evaluation model according to the first decay rate sample and the second decay rate sample and identification information of the detection effectiveness, wherein the detection effectiveness evaluation model is communicatively connected with the timing decay model; obtaining a multi-dimensional real-time detection data set of the target battery, inputting the multi-dimensional real-time detection data set into the detection effectiveness evaluation model, and obtaining the detection effectiveness; performing data feedback optimization according to the detection effectiveness and the preset effectiveness, and generating a battery health indicator according to the optimized battery detection data.
[0005] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0006] By adopting the method of obtaining a multi-dimensional detection sample set of the target battery, multi-dimensional data reflecting the performance of the battery at different time points and under different working conditions is obtained, which provides a basis for subsequent data analysis; the multi-dimensional detection sample set is classified to obtain an active detection sample set and a passive detection sample set, in preparation for the subsequent time series attenuation analysis; a time series attenuation model is established, and the active detection sample set and the passive detection sample set are respectively subjected to time series attenuation analysis according to the time series attenuation model to obtain the first attenuation rate sample and the second attenuation rate sample, and an attenuation model considering the time factor is established, and the active detection data and the passive detection data are respectively subjected to attenuation analysis by the model, so as to obtain the first attenuation rate sample and the second attenuation rate sample reflecting the attenuation characteristics over time, which provides a basis for the subsequent effectiveness evaluation; according to the first attenuation rate sample and the second attenuation rate sample and the identification information of the identification detection effectiveness, the detection effectiveness training is carried out. The technical solution of the whole battery health status detection process is implemented by adopting a real-time battery health status detection evaluation model, which solves the technical problem of low accuracy of battery health status detection in the prior art due to insufficient consideration of the decay of battery detection data over time. The technical solution of the real-time battery health status detection process is implemented by adopting a real-time battery health status detection evaluation model, which solves the technical problem of low accuracy of battery health status detection in the prior art due to insufficient consideration of the decay of battery detection data over time. The technical solution of the real-time battery health status detection process is implemented by adopting a real-time battery health status detection evaluation model, which solves the technical problem of low accuracy of battery health status detection in the prior art due to insufficient consideration of the decay of battery detection data over time. The technical solution of the real-time battery health status detection process is implemented by adopting a time series decay model to analyze the effectiveness of active detection data and passive detection data, which is combined with the detection effectiveness evaluation model, so as to achieve the technical effect of improving the accuracy of battery health status detection.
[0007] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] Figure 1 A flowchart of a battery detection method based on multi-dimensional data analysis is provided for an embodiment of the present application;
[0009] Figure 2 A flow chart of establishing a time-series decay model in a battery detection method based on multi-dimensional data analysis is provided for an embodiment of the present application. DETAILED DESCRIPTION
[0010] The overall idea of the technical solution provided by this application is as follows:
[0011] An embodiment of the present application provides a battery detection method based on multi-dimensional data analysis. First, a multi-dimensional detection sample set of the target battery is obtained, and it is divided into two parts: an active detection sample set and a passive detection sample set. Next, a timing decay model is established, and timing decay analysis is performed on the active detection sample set and the passive detection sample set, respectively, to obtain corresponding decay rate samples. On this basis, a detection effectiveness evaluation model is trained using the decay rate samples and identification information of the identification detection effectiveness. When detecting the battery health status, the multi-dimensional detection data collected in real time is input into the detection effectiveness evaluation model to obtain the detection effectiveness. Afterwards, the detection effectiveness is compared with the preset effectiveness, and the original multi-dimensional real-time detection data set is corrected by data feedback optimization, and an accurate battery health indicator is generated based on the optimized battery detection data.
[0012] To sum up, the present application fully considers the time-varying characteristics and reliability of battery detection data, and by constructing a timing attenuation model and an effectiveness evaluation model, it identifies high-reliability detection data and can accurately judge the health status of the battery based on this data, thereby solving the problem of low accuracy in judging the health status of the battery in the prior art and significantly improving the reliability of battery detection and the accuracy of judging the health status of the battery.
[0013] After introducing the basic principles of the present application, the non-limiting implementation methods of the present application will be specifically introduced below in conjunction with the drawings in the specification.
[0014] like Figure 1 As shown, the embodiment of the present application provides a battery detection method based on multi-dimensional data analysis, the method comprising:
[0015] S1: Obtain a multi-dimensional detection sample set of the target battery.
[0016] Specifically, the target battery refers to the battery object that needs to be tested for battery health status. The multi-dimensional detection sample set refers to a set of data obtained by testing the target battery from different angles or aspects, including electrical parameters (such as voltage, current, internal resistance, capacity, etc.), environmental parameters (such as temperature, humidity, etc.), operating conditions (such as charge and discharge times, charge and discharge rates, standing time, etc.) and performance indicators (such as energy density, power density, cycle life, etc.). The various parameters of the target battery are recorded in real time through sensors and data acquisition systems to obtain a multi-dimensional detection sample set of the target battery, which provides a basis for subsequent analysis and model building.
[0017] S2: Classify the multi-dimensional detection sample set to obtain an active detection sample set and a passive detection sample set.
[0018] Specifically, after obtaining the multi-dimensional detection sample set of the target battery, it is classified and processed to divide the multi-dimensional detection sample set into two parts: an active detection sample set and a passive detection sample set, in preparation for the subsequent timing attenuation analysis.
[0019] Active detection sample set refers to data obtained by actively intervening or controlling the battery system. These data are collected under specific test conditions for specific performance indicators of the battery, such as capacity test, internal resistance test, charge and discharge cycle test, etc. Active detection data can reflect the health status of the battery, but the acquisition cost is high and may have a certain impact on the battery performance. Passive detection sample set refers to data that can be collected without additional intervention during the normal use of the battery. These data come from various sensors in the battery management system, such as voltage, current, temperature, etc. Passive detection data can reflect the performance of the battery in the actual use environment, and the acquisition cost is low, but it will be interfered by external factors, and the data quality and consistency are difficult to guarantee. According to the data collection method of the multi-dimensional detection sample set, the samples collected under human intervention or control in the multi-dimensional detection sample set are divided into active detection samples to obtain the active detection sample set; the samples automatically collected in the multi-dimensional detection sample set are divided into passive detection samples to obtain the passive detection sample set.
[0020] By further refining the data categories and attributes on the basis of multi-dimensional detection sample sets, and obtaining active detection sample sets and passive detection sample sets, data support is provided for targeted timing decay analysis, thereby improving the accuracy and reliability of battery health status detection.
[0021] S3: Establish a timing attenuation model, and perform timing attenuation analysis on the active detection sample set and the passive detection sample set respectively according to the timing attenuation model to obtain a first attenuation rate sample and a second attenuation rate sample, wherein the timing attenuation model includes an active-timing attenuation layer and a passive-timing attenuation layer.
[0022] Specifically, a time series attenuation model is established, which consists of two components: an active-time series attenuation layer and a passive-time series attenuation layer. The active-time series attenuation layer is used to analyze the attenuation characteristics of each sample data source in the active detection sample set, and the passive-time series attenuation layer is used to analyze the attenuation characteristics of each sample sensor source in the passive detection sample set. By constructing two attenuation layers, the time series attenuation model can comprehensively characterize the effectiveness change law of active detection data and passive detection data in the time dimension.
[0023] After establishing the time series attenuation model, the time series attenuation analysis is performed on the active detection sample set and the passive detection sample set respectively. For the active detection sample set, the timestamp information of each sample is first extracted, and then the attenuation coefficient of the sample data source at different time points is calculated according to the order of the timestamps. Among them, the attenuation coefficient is used to quantify the degree of decrease in the effectiveness of the data source, and its value range is between 0 and 1. The smaller the value, the more serious the attenuation. By analyzing the attenuation coefficients corresponding to a series of timestamps, the overall attenuation trend of the active detection sample set is fitted to obtain the first attenuation rate sample. Similarly, for the passive detection sample set, the timestamp information of each sample is extracted, the attenuation coefficient of the sample sensor source at different time points is calculated, and the overall attenuation trend is fitted to obtain the second attenuation rate sample. Among them, the time series attenuation characteristics of active detection data and passive detection data are different. The effectiveness of active detection data is affected by factors such as human intervention measures and aging of test equipment, and the attenuation rate is relatively fast; while the effectiveness of passive detection data mainly depends on the performance and stability of the sensor, and the attenuation rate is relatively slow.
[0024] Through time-series decay analysis, the effectiveness of active and passive detection data at different time points can be quantitatively evaluated, providing important data support and basis for subsequent detection effectiveness evaluation and battery health status detection.
[0025] S4: training a detection effectiveness evaluation model according to the first attenuation rate sample, the second attenuation rate sample and identification information identifying detection effectiveness, wherein the detection effectiveness evaluation model is communicatively connected with the timing attenuation model.
[0026] Specifically, based on the first decay rate sample and the second decay rate sample, a detection effectiveness evaluation model is constructed to provide a reliable judgment basis for subsequent battery health status detection.
[0027] First, a set of identification information for identifying the effectiveness of detection is established, which is determined by experts based on experience and is used to characterize the effectiveness level of battery detection data at different time points. For example, use labels such as "high", "medium", and "low" to describe the effectiveness of detection data, or use a value between 0 and 1 to quantify the effectiveness. After obtaining the identification information, match it with the first decay rate sample and the second decay rate sample to construct a training data set. Each training sample consists of three parts: the first decay rate, the second decay rate, and the corresponding effectiveness identifier. The first decay rate and the second decay rate respectively characterize the degree of decay of active detection data and passive detection data at that point in time, and the effectiveness identifier characterizes the overall effectiveness of the detection data at that point in time.
[0028] Next, the detection effectiveness evaluation model is trained using the training data set. The model can use various machine learning algorithms, such as logistic regression, decision tree, support vector machine, neural network, etc. The input of the model is the first decay rate and the second decay rate, and the output is the detection effectiveness. By continuously adjusting the parameters and structure of the model and minimizing the error between the output value and the identification value, a detection effectiveness evaluation model with excellent performance is obtained.
[0029] Among them, a communication connection is established between the detection effectiveness evaluation model and the time-series decay model, so as to realize real-time data exchange and collaborative work between the two models, so as to efficiently and accurately evaluate the effectiveness of battery detection data and provide a reliable basis for judgment for subsequent battery health status detection. The time-series decay model analyzes the effectiveness decay of active detection data and passive detection data over time, obtains the first decay rate and the second decay rate at different time points, and then transmits the obtained first decay rate and the second decay rate to the detection effectiveness evaluation model in real time through the communication connection, so as to obtain the effectiveness of the battery detection data at the current time point.
[0030] By training the effectiveness evaluation model, the effectiveness of the battery detection data can be judged in real time according to the timing attenuation characteristics of the active detection data and the passive detection data, providing a reliable judgment basis for battery health status detection.
[0031] S5: Obtain a multi-dimensional real-time detection data set of the target battery, input the multi-dimensional real-time detection data set into the detection effectiveness evaluation model, and obtain detection effectiveness.
[0032] Specifically, after constructing the detection effectiveness evaluation model, the detection data of the target battery is acquired in real time, and its effectiveness is evaluated through the detection effectiveness evaluation model to provide reliable data support for battery health status detection.
[0033] First, obtain a multi-dimensional real-time detection data set of the target battery, including measurement values of multiple dimensions such as voltage, current, temperature, and internal resistance, to reflect the current state of the target battery. After obtaining the multi-dimensional real-time detection data set, classify it and divide the multi-dimensional real-time detection data set into two categories: active detection data and passive detection data. After completing the data classification, the active detection data and the passive detection data are respectively subjected to a time series attenuation analysis through a time series attenuation model to obtain the first attenuation rate of the active detection data and the second attenuation rate of the passive detection data, reflecting the degree of effectiveness decrease of the multi-dimensional real-time detection data set over time. The obtained first attenuation rate and the second attenuation rate are input into the detection effectiveness evaluation model to obtain the overall effectiveness of the multi-dimensional real-time detection data set at the current time point, as the detection effectiveness, such as a value between 0 and 1, indicating the credibility of the multi-dimensional real-time detection data set.
[0034] Through real-time effectiveness evaluation, the effectiveness index of the current test data of the target battery is obtained, which can more accurately reflect the credibility of the test data and provide high-quality and highly reliable data support for subsequent battery health status testing, thereby ensuring the accuracy and reliability of the test results.
[0035] S6: Perform data feedback optimization according to the detection effectiveness and the preset effectiveness, and generate a battery health index according to the optimized battery detection data.
[0036] Specifically, first, the obtained detection effectiveness is compared with the preset effectiveness. The preset effectiveness threshold is determined based on factors such as battery type, application scenario, reliability requirements, etc., and indicates the minimum acceptable effectiveness level of the detection data. By comparing the detection effectiveness with the preset effectiveness, it is determined whether the quality of the current multi-dimensional real-time detection data set meets the requirements. When the detection effectiveness is greater than or equal to the preset effectiveness, it means that the current multi-dimensional real-time detection data set has a high quality and can be directly used for battery health status assessment. At this point, based on the multi-dimensional real-time detection data set, the data of multiple dimensions such as voltage, current, and temperature are comprehensively analyzed to evaluate the health status of the target battery, obtain health indicators that reflect the overall performance and remaining life of the battery, and generate corresponding battery health indicators.
[0037] When the detection effectiveness is lower than the preset effectiveness, it means that the current multi-dimensional real-time detection data set is of poor quality, and direct use for health status assessment will lead to result deviation and misjudgment. At this time, the data feedback optimization mechanism is started. Specifically, with the goal of minimizing data loss, the multi-dimensional real-time detection data set is corrected and optimized through data screening methods, and the data with excessive effectiveness attenuation is removed to generate a multi-dimensional optimized detection data set with high effectiveness as the optimized battery detection data. Then, reliable and accurate battery health indicators are obtained based on the optimized data set.
[0038] Through the effectiveness comparison and data feedback optimization mechanism, the quality of real-time battery detection data can be adaptively adjusted and improved, and the reliability and accuracy of battery health status assessment can be improved. Based on the optimized detection data, the generated battery health indicators can comprehensively and quantitatively reflect the performance status and remaining life of the battery, providing important decision support for battery management.
[0039] Furthermore, the embodiment of the present application also includes:
[0040] The active detection sample set is data representing the battery health status obtained through active intervention, and the passive detection sample set is data representing the battery health status collected without additional intervention; a data conversion processing template is established, the active detection sample set and the passive detection sample set are input into the data conversion processing template, and according to the data conversion processing template, the processed active detection sample set and the processed passive detection sample set are obtained.
[0041] In a feasible implementation, an active detection sample set refers to a data set obtained by actively intervening in or controlling the battery system, including capacity test data, internal resistance test data, pulse test data, and charge and discharge cycle test data. By applying specific test conditions and loads to the battery system, an active detection sample set reflecting the performance and status of the battery is obtained. In contrast, a passive detection sample set refers to a data set that can be collected during normal use of the battery without additional human intervention, and comes from various sensors and monitoring devices in the battery management system, such as voltage sensors, current sensors, temperature sensors, etc. Passive detection sample sets are obtained by continuously collecting and recording the operating data of the battery in the actual application environment, reflecting the dynamic characteristics and health status of the battery.
[0042] Since the original collected active detection sample set and passive detection sample set have inconsistent data formats, large data quality differences, data noise interference and other problems, directly using them for subsequent feature extraction and health status assessment will affect the accuracy and reliability of the results. Therefore, the original active detection sample set and passive detection sample set are preprocessed and converted through the data conversion processing template to improve the quality and consistency of the data. The data conversion processing template includes a series of data processing algorithms and processes, such as data cleaning, data normalization, data filtering, data interpolation, etc. By performing data conversion processing on the active detection sample set and the passive detection sample set through the data conversion processing template respectively, invalid data, outliers, noise interference, etc. are removed, and the reliability of the data is improved; at the same time, through data normalization and standardization, data from different sources and different dimensions are unified to the same scale and range, which is convenient for subsequent feature extraction and model training. After being processed by the data conversion processing template, the original active detection sample set and passive detection sample set are converted into processed active detection sample sets and processed passive detection sample sets with higher quality and more unified format, which provide a high-quality data basis for subsequent time series decay analysis and health status assessment, and improve the performance and reliability of battery detection.
[0043] Furthermore, the embodiment of the present application also includes:
[0044] Analyze the data source of the active detection sample set to obtain multiple categories of sample data sources; record the intervention method and intervention equipment of each category of sample data source; perform attenuation influence analysis on the multiple categories of sample data sources according to the intervention method and the intervention equipment to obtain multiple categories of first attenuation influence coefficients; generate a first attenuation influence layer according to the multiple categories of first attenuation influence coefficients, and optimize the active-timing attenuation layer according to the first attenuation influence layer.
[0045] In a preferred embodiment, the active-temporal decay layer in the temporal decay model is specifically used to analyze and quantify the decay characteristics of the data source in the active detection sample set. In order to further optimize the performance of the active-temporal decay layer, it is optimized and improved.
[0046] First, the data source of the active detection sample set is analyzed to obtain multiple types of sample data sources. Specifically, the active detection sample set comes from multiple different data sources, such as capacity test equipment, internal resistance test equipment, pulse test equipment, and charge and discharge cycle test equipment. Different data sources differ in test principles, test conditions, data characteristics, etc. Therefore, the data source analysis of the active detection sample set is performed to identify and divide different types of sample data sources, such as capacity test data source, internal resistance test data source, etc., to obtain multiple types of sample data sources. After obtaining multiple types of sample data sources, the specific intervention methods and intervention devices of each type of sample data source are further analyzed and recorded. Intervention methods refer to the human intervention or control measures taken to obtain this type of data source, such as applying specific charge and discharge currents to the battery, setting specific temperature conditions, etc.; intervention devices refer to specific instruments and devices used to implement intervention methods and collect data, such as charge and discharge testers, temperature control boxes, etc. By recording intervention methods and intervention devices, necessary background information is provided for subsequent data attenuation analysis.
[0047] After clarifying the intervention methods and intervention devices of each type of sample data source, the impact of the intervention methods and intervention devices on data attenuation is analyzed. Different intervention methods and intervention devices will have different degrees of impact on the timeliness and accuracy of the data, resulting in different rates and laws of data attenuation over time. Through attenuation impact analysis, the data attenuation characteristics of each type of sample data source are evaluated to obtain the corresponding first attenuation impact coefficient. The first attenuation impact coefficient indicates the degree of influence of the intervention methods and intervention devices on the attenuation rate of this type of data source. The larger the value, the more significant the impact. After obtaining the first attenuation impact coefficients of multiple types of sample data sources, the first attenuation impact layer is constructed accordingly. The first attenuation impact layer is an additional layer parallel to the active-sequential attenuation layer, which is used to supplement and correct the attenuation analysis results of the active-sequential attenuation layer. Specifically, the first attenuation impact layer adjusts and optimizes the active-sequential attenuation layer according to the first attenuation impact coefficient of each type of data source, so that it can more accurately describe the actual attenuation law of different data sources. By introducing the first attenuation impact layer, the analysis accuracy and adaptability of the active-sequential attenuation layer can be effectively improved, thereby better evaluating the temporal attenuation characteristics of the active detection sample set.
[0048] By analyzing multiple types of sample data sources of the active detection sample set, recording its intervention methods and intervention devices, evaluating its impact on data attenuation, and constructing the first attenuation influence layer, the temporal attenuation law of the active detection data can be more refined and accurately characterized, providing more reliable data support for subsequent battery health status assessment.
[0049] Furthermore, the embodiment of the present application also includes:
[0050] According to the intervention method corresponding to each type of sample data source, the data output type and the data output path are recorded; according to the data output type and the data output path, the method data loss is obtained; the equipment life index of the equipment intervened by each type of sample data source is obtained, and the equipment data loss is obtained by analyzing the equipment life index; a loss function is introduced to perform a loss analysis on the method data loss and the equipment data loss, and the sum of the data losses corresponding to each type of sample data source is obtained; according to the sum of the data losses corresponding to each type of sample data source, multiple types of first attenuation influence coefficients are obtained.
[0051] In a preferred embodiment, different intervention methods will produce data outputs of different types and formats, and transmit and store them through different paths. For example, the intervention method of capacity test may output data such as battery capacity and voltage, and transmit them to the database through the serial port or network for storage; while the intervention method of internal resistance test may output data such as battery internal resistance and temperature, and transmit them to the local file through the I2C bus for storage. In order to fully understand the data acquisition and transmission process, the data output type of each type of sample data source is recorded in detail, such as analog signal, digital signal, text format, etc., as well as the data output path of the data from the acquisition point to the storage location, including transmission protocol, intermediate node, storage medium, etc., to provide an important reference for subsequent data loss analysis. In the actual data acquisition and transmission process, due to the characteristics of the data output type and data output path, a certain degree of data loss or distortion will occur. For example, when an analog signal is converted to a digital signal, quantization noise will be introduced, and data packet loss will occur in long-distance network transmission. In order to estimate and calculate the data loss of each type of sample data source, the influence of various factors needs to be considered comprehensively. For analog signal output, analyze the data distortion introduced by factors such as quantization noise, sampling frequency, and amplifier nonlinearity; for digital signal output, consider factors such as ADC conversion error, data compression loss, and delay introduced by digital filtering; for network transmission paths, evaluate the data quality degradation caused by factors such as packet loss rate, transmission delay, and network congestion; for storage media, analyze the information loss caused by factors such as data write error, read noise, and data corruption. By combining measured data and expert experience, evaluate the data loss of each type of sample data source during data acquisition and transmission. At the same time, as the physical carrier of data acquisition, the performance and status of the intervention device will also significantly affect the quality of the data. Equipment life indicators, such as usage time, number of working cycles, and calibration cycle, are important parameters for evaluating the health status and reliability of intervention equipment. In order to obtain equipment data loss, it is first necessary to consult the technical specifications and user manuals of each intervention device, extract its key life indicators, and record the actual use of the equipment, such as the usage time and the number of completed working cycles. Then, analyze the health status and performance degradation of the equipment based on the equipment life indicators and actual use. By combining the equipment performance degradation curve and data quality impact analysis, the equipment data loss caused by equipment aging and performance degradation of each type of sample data source is obtained.
[0052] Afterwards, the loss function is introduced to analyze the mode data loss and device data loss, and the data loss and corresponding to each type of sample data source are obtained. Among them, the loss function is used to measure and evaluate the comprehensive impact of data loss. In order to comprehensively consider the comprehensive effects of mode data loss and device data loss, a suitable loss function is selected, such as mean square error loss, cross entropy loss, etc., and the mode data loss and device data loss of each type of sample data source are used as input variables. The comprehensive data loss and are obtained through the loss function. Afterwards, according to the data loss and corresponding to each type of sample data source, multiple types of first attenuation influence coefficients are obtained to characterize the degree of influence of data loss on the attenuation rate. In order to obtain the first attenuation influence coefficient, a suitable mapping function is designed according to the actual situation and expert experience, and the data loss and are mapped to the degree of influence of the attenuation rate, wherein the mapping function can be a linear function, an exponential function, a logarithmic function, etc. The corresponding first attenuation influence coefficient is obtained by mapping and calculating the data loss and of each type of sample data source.
[0053] By fully considering the combined impact of data loss in the intervention method and data loss in the intervention device on data attenuation, a loss function is introduced to evaluate the data attenuation characteristics of each type of sample data source, and the corresponding first attenuation influence coefficient is obtained, which provides an important reference for optimizing the active-timing attenuation layer and helps to improve the accuracy and reliability of battery health status assessment.
[0054] Furthermore, the embodiment of the present application also includes:
[0055] Analyze the sensor data source of the passive detection sample set to obtain multiple categories of sample sensor sources; obtain the sensor stability and sensor failure rate of the multiple categories of sample sensor sources; obtain multiple categories of second attenuation influence coefficients based on the sensor stability and the sensor failure rate; generate a second attenuation influence layer according to the multiple categories of second attenuation influence coefficients, and optimize the passive-timing attenuation layer according to the second attenuation influence layer.
[0056] In a preferred embodiment, the passive-timing attenuation layer in the timing attenuation model is different from the active-timing attenuation layer in performing attenuation analysis on the active detection sample set. The passive-timing attenuation layer is specifically used to analyze and quantify the attenuation characteristics of the sensor data source in the passive detection sample set.
[0057] The passive detection sample set comes from various sensors and monitoring devices in the battery management system, such as voltage sensors, current sensors, temperature sensors, etc. Different types of sensors have significant differences in measurement principles, performance indicators, data characteristics, etc. In order to accurately evaluate the attenuation law of passive detection data, the sensor data sources of the passive detection sample set are comprehensively analyzed and classified. By identifying and extracting the key features of each sensor data source, such as measurement range, resolution, sampling rate, etc., it is divided into multiple different sample sensor source categories, such as voltage sensor source, current sensor source, temperature sensor source, etc., which is helpful for subsequent targeted analysis of the attenuation characteristics of different types of sensor sources. Subsequently, the sensor stability and sensor failure rate of multiple types of sample sensor sources are obtained. Among them, sensor stability and sensor failure rate are key indicators for evaluating sensor performance and reliability. Sensor stability characterizes the ability of the sensor to maintain measurement accuracy and repeatability within a certain period of time. The higher the stability, the more reliable the sensor's measurement results and the smaller the fluctuation; the sensor failure rate reflects the probability of failure or failure of the sensor during use. The lower the failure rate, the higher the reliability and life of the sensor. In order to obtain the sensing stability and sensing failure rate, the historical working data, fault records, calibration reports, etc. of each sensor are comprehensively analyzed. Through statistical analysis and probability calculation, the sensing stability and sensing failure rate of each type of sample sensor source are evaluated, providing important parameters for the subsequent attenuation impact analysis.
[0058] The second attenuation influence coefficient is an indicator used to quantify the degree of influence of sensor performance on data attenuation rate. In order to obtain the second attenuation influence coefficient, a quantitative correlation model between sensor performance and data attenuation is established. The model comprehensively considers the influence mechanism of sensor stability and sensor failure rate on data quality, and maps these two indicators into the influence coefficient of attenuation rate. Among them, the lower the sensor stability and the higher the sensor failure rate, the larger the corresponding second attenuation influence coefficient, indicating that the data attenuation speed of this type of sensor source is faster. By analyzing the sensor stability and sensor failure rate of each type of sample sensor source, the corresponding multi-class second attenuation influence coefficients are obtained to provide parameter support for optimizing the passive-timing attenuation layer. Thereafter, the second attenuation influence layer is generated according to the multi-class second attenuation influence coefficients, and the passive-timing attenuation layer is optimized according to the second attenuation influence layer. The second attenuation influence layer is an auxiliary layer parallel to the passive-timing attenuation layer. It is to adaptively adjust the parameters of the passive-timing attenuation layer according to the attenuation characteristics of different types of sensor sources, and more accurately describe the actual attenuation law of different sensor source data to improve the accuracy and reliability of attenuation analysis.
[0059] By analyzing the attenuation of the sensor data source for the passive detection sample set, the stability and fault characteristics of different types of sensor sources are fully considered, the second attenuation influence coefficient and the second attenuation influence layer are introduced, and the adaptive optimization of the passive-timing attenuation layer is realized, thereby improving the accuracy and reliability of the pool health status assessment.
[0060] Further, such as Figure 2 As shown, the embodiment of the present application also includes:
[0061] Obtain the active-attenuation timing characteristics of each type of sample data source in the active-timing attenuation layer, and the passive-attenuation timing characteristics of each type of sample sensor source in the passive-timing attenuation layer; generate a corresponding timestamp mark sequence according to the active-attenuation timing characteristics and the passive-attenuation timing characteristics; perform function matching based on an attenuation function call library according to the active-attenuation timing characteristics and the passive-attenuation timing characteristics, and generate a matching attenuation function; perform training according to the timestamp mark sequence and the matching attenuation function, and establish a timing attenuation model.
[0062] In a preferred embodiment, the decay time series feature refers to a parameter that reflects the decay law of a data source or a sensor source in the time dimension. For the active-time series decay layer, the active-decay time series feature of each type of sample data source includes decay rate, decay acceleration, decay inflection point, etc.; for the passive-time series decay layer, the passive-decay time series feature of each type of sample sensor source includes stable period, fault period, decay period, etc. In order to obtain these decay time series features, the historical data of each type of sample data source or sample sensor source is analyzed by time series decomposition, time series clustering, etc., to identify its decay law in the time dimension. Then, according to the active-decay time series features and the passive-decay time series features, a corresponding timestamp mark sequence is generated. Among them, the timestamp mark sequence is a timeline index used to describe the data decay process. By discretizing and encoding the decay time series features, the continuous time axis is mapped into a series of discrete timestamp marks, and each timestamp mark corresponds to a key time node in the decay process, such as the decay starting point, decay turning point, decay end point, etc. By generating a timestamp mark sequence, the decay processes of different sample data sources or sample sensor sources are unified into the same time reference system, which facilitates the establishment of a time series decay model.
[0063] Then, according to the active-attenuation time series characteristics and passive-attenuation time series characteristics, function matching is performed based on the attenuation function call library to generate a matching attenuation function. The matching attenuation function is a mathematical model used to describe the law of data attenuation, such as exponential attenuation function, power law attenuation function, logarithmic attenuation function, etc. In order to accurately characterize the attenuation characteristics of different sample data sources or sample sensor sources, the optimal attenuation function form is selected from the attenuation function call library. The attenuation function call library is a predefined function set that contains a variety of attenuation function types and their parameter ranges. Through curve fitting, with the goal of minimizing the fitting error, the fit between the active-attenuation time series characteristics and the passive-attenuation time series characteristics and different attenuation functions is compared to find the best matching attenuation function form, that is, the matching attenuation function. After that, training is performed based on the timestamp mark sequence and the matching attenuation function to establish a time series attenuation model. The time series attenuation model is a comprehensive model that can simultaneously characterize the attenuation law of active detection data and passive detection data. By taking the timestamp mark sequence and the matching attenuation function as input, the machine learning algorithm is used for training and optimization to obtain a model that can dynamically judge the data attenuation trend. During the model training process, the timestamp mark sequence plays the role of time axis alignment and sample synchronization, ensuring the temporal correlation between different sample data sources or sample sensor sources; while the matching attenuation function provides a priori assumptions and constraints on the attenuation law, reducing the parameter search space of the model and improving training efficiency and generalization ability.
[0064] By making full use of the timing characteristics of active detection data and passive detection data, introducing timestamp marking sequence and matching decay function, we can accurately characterize and dynamically predict the decay law of battery health status, and improve the accuracy and reliability of battery health status assessment.
[0065] Furthermore, the embodiment of the present application also includes:
[0066] When the detection effectiveness is greater than or equal to the preset effectiveness, a battery health status assessment is performed based on the multi-dimensional real-time detection data set of the target battery to obtain the battery health index; when the detection effectiveness is less than the preset effectiveness, a data feedback optimizer is established with the goal of minimizing the effectiveness difference between the preset effectiveness and the detection effectiveness, and the multi-dimensional real-time detection data set of the target battery is predicted and updated based on the data feedback optimizer, a multi-dimensional optimized detection data set is output, and a battery health status assessment is performed based on the multi-dimensional optimized detection data set to obtain the battery health index.
[0067] In a preferred embodiment, when the detection effectiveness is greater than or equal to the preset effectiveness, the battery health status is evaluated directly based on the multi-dimensional real-time detection data set of the target battery to obtain the battery health index. The preset effectiveness is a pre-set effectiveness threshold based on the accuracy requirements and application scenarios of the battery health status evaluation. When the actual detection effectiveness obtained is higher than or equal to the threshold, it means that the quality and credibility of the detection data in the current multi-dimensional real-time detection data set have met the evaluation requirements. These data are directly used to evaluate the battery health status, and key indicators reflecting the battery health status, such as capacity, internal resistance, life, etc., are obtained as battery health indicators. The performance degradation and aging degree of the battery are characterized from multiple dimensions, providing a basis for battery safety management and maintenance decisions.
[0068] When the detection effectiveness is less than the preset effectiveness, the real-time detection data set is predicted, updated and optimized by the data feedback optimizer, and then the battery health status is evaluated based on the optimized data set to finally obtain the battery health index. Among them, the data feedback optimizer is an adaptive data processing module, which takes minimizing the difference between the preset effectiveness and the detection effectiveness as the optimization goal, corrects, compensates and predicts the detection data in the multi-dimensional real-time detection data set, and generates a multi-dimensional optimized detection data set with higher quality and stronger credibility. Specifically, the data feedback optimizer first identifies the data with low effectiveness in the multi-dimensional real-time detection data set. Then, combined with the physical mechanism and statistical laws of the battery health status, a data prediction and extrapolation model is constructed, and the marked data is estimated and compensated to generate a multi-dimensional optimized detection data set with high effectiveness. Afterwards, the multi-dimensional optimized detection data set is used as the optimized detection data set to evaluate the battery health status and obtain a more stable and reliable battery health index.
[0069] By setting preset validity and introducing data feedback optimizer, we can achieve quality control and optimization upgrade of the evaluation process, which not only improves the credibility and stability of battery health indicators, but also realizes quality management and optimization control of real-time detection data, ensuring that the input data of battery health status evaluation is always at the optimal quality level.
[0070] Furthermore, the embodiment of the present application also includes:
[0071] Acquire a health assessment task; send the health assessment task to the data feedback optimizer for task analysis to obtain an active-correlation coefficient and a passive-correlation coefficient; acquire an active real-time detection data set and a passive real-time detection data set according to the multi-dimensional real-time detection data set; when the active-correlation coefficient is greater than the passive-correlation coefficient, predict and update the active real-time detection data set.
[0072] In a feasible implementation, the battery health status assessment serves different decision optimization purposes, such as battery remaining life prediction, battery balancing management, battery fault diagnosis, etc. Different health assessment tasks have different requirements for the quality and characteristics of the detection data. In order to enable data feedback optimization to adapt to the needs of different health assessment tasks, the current health assessment task type and optimization goal are first clarified through human-computer interaction, task configuration file parsing, etc. After clarifying the health assessment task, the data feedback optimizer can adjust the optimization strategy and parameters accordingly to improve the pertinence and effectiveness of the optimization. After obtaining the health assessment task, the health assessment task is sent to the data feedback optimizer for task parsing to obtain the active-correlation coefficient and the passive-correlation coefficient. Among them, the active-correlation coefficient and the passive-correlation coefficient respectively characterize the degree of correlation between the active detection data and the passive detection data and the current health assessment task. Different types of detection data have different roles and contributions to different health assessment tasks. For example, for battery capacity attenuation assessment, the correlation of charge and discharge test data is higher; while for battery temperature abnormality diagnosis, the correlation of temperature sensor data may be higher. In order to quantify this difference, the data feedback optimizer extracts the specific data needed to complete the health assessment task through task analysis, and analyzes the correlation coefficient between active detection data and passive detection data and the health assessment task. The higher the correlation coefficient, the greater the impact of the data on the current health assessment task, and the higher the priority and weight should be given when updating the data prediction.
[0073] When generating battery health indicators based on a multi-dimensional real-time detection data set, the multi-dimensional real-time detection data set is first distinguished and extracted according to the attributes of active detection and passive detection to obtain an active real-time detection data set and a passive real-time detection data set. Then, the correlation between the active real-time detection data set and the passive real-time detection data set and the current health assessment task is determined to obtain an active-correlation coefficient and a passive-correlation coefficient. When the active-correlation coefficient is greater than the passive-correlation coefficient, it indicates that the health assessment task is more dependent on the active detection data. At this time, the active real-time detection data set should be predicted and updated to improve the reliability of the assessment results.
[0074] By fully considering the differentiated requirements of different health assessment tasks for detection data quality and characteristics, task parsing and correlation analysis are performed, and dynamic adjustment and optimization of data prediction and update strategies are achieved, thereby improving the accuracy of battery health status assessment.
[0075] In summary, the battery detection method based on multi-dimensional data analysis provided by the embodiment of the present application has the following technical effects:
[0076] Acquire a multi-dimensional detection sample set of the target battery, and obtain multi-dimensional detection data of the battery at different time points and different working conditions, providing comprehensive and rich samples for subsequent analysis. Classify the multi-dimensional detection sample set to obtain active detection sample sets and passive detection sample sets, laying the foundation for separate time series decay analysis. Establish a time series decay model, perform time series decay analysis on the active detection sample set and the passive detection sample set according to the time series decay model, and obtain the first decay rate sample and the second decay rate sample, wherein the time series decay model includes an active-time series decay layer and a passive-time series decay layer, and by analyzing the decay law of active detection data and passive detection data over time, the decay rate sample obtained provides an important basis for the subsequent evaluation of the effectiveness of the detection data. According to the first decay rate sample and the second decay rate sample and the identification information of the detection effectiveness, train the detection effectiveness evaluation model, wherein the detection effectiveness evaluation model is connected to the time series decay model in communication to provide support for the optimization of detection data. Acquire a multi-dimensional real-time detection data set of the target battery, input the multi-dimensional real-time detection data set into the detection effectiveness evaluation model, obtain the detection effectiveness, and obtain a quantitative indicator reflecting the credibility of the detection data, so as to prepare for the subsequent optimization of data and judgment of battery health status. Perform data feedback optimization based on the detection effectiveness and the preset effectiveness, generate battery health indicators based on the optimized battery detection data, select and correct the original detection data by comparing the detection effectiveness with the preset effectiveness, and generate battery health indicators based on the optimized battery detection data, completing the entire battery health status judgment process and improving the accuracy of battery health status detection.
[0077] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.
[0078] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these changes and variations.
Claims
1. A battery detection method based on multi-dimensional data analysis, characterized in that: The method comprises: Obtain a multi-dimensional detection sample set of the target battery; Classifying the multi-dimensional detection sample set to obtain an active detection sample set and a passive detection sample set; Establishing a timing attenuation model, performing timing attenuation analysis on the active detection sample set and the passive detection sample set respectively according to the timing attenuation model, and obtaining a first attenuation rate sample and a second attenuation rate sample, wherein the timing attenuation model includes an active-timing attenuation layer and a passive-timing attenuation layer; According to the first attenuation rate sample and the second attenuation rate sample and identification information identifying detection effectiveness, training a detection effectiveness evaluation model, wherein the detection effectiveness evaluation model is communicatively connected with the timing attenuation model; Acquire a multi-dimensional real-time detection data set of the target battery, input the multi-dimensional real-time detection data set into the detection effectiveness evaluation model, and acquire detection effectiveness; Data feedback optimization is performed according to the detection effectiveness and the preset effectiveness, and a battery health index is generated according to the optimized battery detection data.
2. The method according to claim 1, characterized in that Obtain active detection sample sets and passive detection sample sets; The active detection sample set is data representing the battery health status obtained through active intervention, and the passive detection sample set is data representing the battery health status collected without additional intervention; A data conversion processing template is established, the active detection sample set and the passive detection sample set are input into the data conversion processing template, and a processed active detection sample set and a processed passive detection sample set are obtained according to the data conversion processing template.
3. The method according to claim 1, characterized in that The timing decay model includes an active-timing decay layer, and further includes: Analyze the data source of the active detection sample set to obtain multiple types of sample data sources; Record the intervention method and intervention device for each type of sample data source; Performing attenuation impact analysis on the multiple types of sample data sources according to the intervention method and the intervention device to obtain multiple types of first attenuation impact coefficients; A first attenuation influence layer is generated according to the multiple types of first attenuation influence coefficients, and the active-sequential attenuation layer is optimized according to the first attenuation influence layer.
4. The method according to claim 3, characterized in that According to the intervention mode and the intervention device, attenuation influence analysis is performed on the multiple types of sample data sources to obtain multiple types of first attenuation influence coefficients, the method comprising: According to the intervention method corresponding to each type of sample data source, record the data output type and data output path; According to the data output type and the data output path, the data loss is obtained; Obtaining equipment life indicators of equipment involved in each type of sample data source, and performing analysis based on the equipment life indicators to obtain equipment data loss; Introducing a loss function to perform loss analysis on the data loss of the method and the data loss of the device, and obtaining the sum of data losses corresponding to each type of sample data source; According to the sum of data losses corresponding to each type of sample data source, multiple types of first attenuation influence coefficients are obtained.
5. The method according to claim 3, characterized in that The time-series attenuation model includes a passive-time-series attenuation layer, including: Analyzing the sensor data source of the passive detection sample set to obtain multiple types of sample sensor sources; Obtaining the sensing stability and sensing failure rate of the multiple types of sample sensor sources; Obtaining multiple types of second attenuation influence coefficients according to the sensor stability and the sensor failure rate; A second attenuation influence layer is generated according to the multiple types of second attenuation influence coefficients, and the passive-temporal attenuation layer is optimized according to the second attenuation influence layer.
6. The method according to claim 1, characterized in that Establishing a time-series decay model, the method further comprises: Acquire active-attenuation time series characteristics of each type of sample data source in the active-time series attenuation layer, and passive-attenuation time series characteristics of each type of sample sensor source in the passive-time series attenuation layer; Generate a corresponding timestamp mark sequence according to the active-decay timing characteristics and the passive-decay timing characteristics; According to the active-attenuation timing characteristics and the passive-attenuation timing characteristics, function matching is performed based on an attenuation function call library to generate a matching attenuation function; Training is performed based on the timestamp mark sequence and the matching decay function to establish a time series decay model.
7. The method according to claim 1, characterized in that The data feedback optimization is performed according to the detection effectiveness and the preset effectiveness, the method comprising: When the detection effectiveness is greater than or equal to the preset effectiveness, a battery health status assessment is performed according to the multi-dimensional real-time detection data set of the target battery to obtain the battery health index; When the detection effectiveness is less than the preset effectiveness, a data feedback optimizer is established with the goal of minimizing the effectiveness difference between the preset effectiveness and the detection effectiveness. The multi-dimensional real-time detection data set of the target battery is predicted and updated according to the data feedback optimizer, and a multi-dimensional optimized detection data set is output. The battery health status is evaluated according to the multi-dimensional optimized detection data set to obtain a battery health indicator.
8. The method according to claim 7, characterized in that According to the data feedback optimizer, a multi-dimensional real-time detection data set of the target battery is predicted and updated, and the method further includes: Get health assessment tasks; Sending the health assessment task to the data feedback optimizer for task analysis to obtain active-correlation coefficients and passive-correlation coefficients; According to the multi-dimensional real-time detection data set, an active real-time detection data set and a passive real-time detection data set are obtained. When the active-correlation coefficient is greater than the passive-correlation coefficient, the active real-time detection data set is predicted and updated.
