Identification method for bad data of vehicle power system

Abstract

The invention belongs to the technical field of vehicle power systems, and particularly relates to a method for identifying bad data of a vehicle power system, which comprises the following steps of: 1, pre-screening acquired original measurement data; 2, constructing a multi-physical domain panoramic perception model; step 3, based on state estimation solution, realizing panoramic perception; 4, identifying bad data by adopting a maximum standardized residual error method; and 5, carrying out iterative cleaning and intelligent completion on the bad data. According to the method, a system physical model is constructed and a state estimation algorithm is utilized to realize accurate and rapid identification of bad data in vehicle power system monitoring data.

Description

Technical Field

[0001] This invention belongs to the field of vehicle electrical system technology, and specifically relates to a method for identifying bad data in a vehicle electrical system. Background Technology

[0002] Vehicle electrical systems are evolving towards integration, high voltage, and intelligence, supporting core functions such as power drive and autonomous driving. These systems rely on the collaborative work of numerous sensors and controllers. Monitoring data such as voltage, current, and temperature are the cornerstone of state perception and decision-making; their accuracy directly affects vehicle performance, range, and driving safety. In actual vehicle operation, due to drastic changes in operating conditions, mechanical vibration, extreme temperatures, and complex electromagnetic interference, sensor data is highly susceptible to distortion, drift, or sudden errors, resulting in "bad data." The presence of bad data can seriously mislead the vehicle's control system. For example, bad battery temperature data may falsely trigger charging protection, while bad current data may cause power imbalance, threatening safety. Therefore, how to quickly and accurately identify and eliminate this bad data from massive amounts of monitoring data, ensuring the data purity of the core control system, has become a key technical challenge for improving the reliability and safety of modern vehicle electrical systems.

[0003] Currently, bad data identification in power systems is typically transformed into a verification problem based on state estimation. By establishing a physical model of the system, state estimation algorithms are used to calculate theoretical values, which are then compared with actual measured values ​​to identify abnormal data with significant deviations. Due to their theoretical completeness and high accuracy, this type of method is widely used in online monitoring of traditional power systems and other industrial sectors.

[0004] Chinese Patent CN110750575A (Applicant: Hohai University, Publication Date: February 4, 2020) is designed to effectively identify various types of bad data in an integrated energy system with interconnected electricity and heat, solving the problems of traditional methods failing to identify bad data in heating network temperature measurements and the failure to identify bad data at coupling nodes due to neglecting system constraints. The implementation process is as follows: First, acquire the grid information (including grid topology, branch parameters, generator parameters, and grid measurement information) and heat network information (including pipe length, diameter, roughness, impedance coefficient, thermoelectric ratio of coupling elements, and heat network measurement information) of the integrated energy system for electrothermal interconnection. Then, based on this information, establish and solve the Lagrange state estimation model considering equality constraints. By constructing the Lagrange function, use the Gauss-Newton method to solve the nonlinear equations to obtain the system state estimate. Then, calculate the residual sensitivity matrix and the regularized residuals of each measurement. Next, set the maximum threshold for the regularized residuals and determine whether the maximum value of the regularized residuals exceeds the limit. If it does, determine that the corresponding measurement is bad data and remove it. Finally, return and repeat the above steps of establishing the model, calculating parameters, and determining removal until the maximum value of the regularized residuals is less than the set threshold, thus completing the bad data identification.

[0005] Chinese Patent CN114626959A (Applicant: Tianjin University, Publication Date: June 14, 2022) is designed to comprehensively utilize multi-source measurement data (PMU measurement, SCADA measurement, AMI measurement) in the power distribution network to perform robust state estimation, while also identifying bad measurement data to improve the accuracy and speed of state estimation. The implementation process is as follows: First, for an active distribution network containing multi-source measurement data, the topological connection relationship of the distribution network, line impedance parameters, load and distributed power source access location information, and multi-source measurement data configuration information are input. Based on this information, the state variables (including node voltage state variables, line state variables, and node injected current amplitude state variables) for robust state estimation using second-order cone programming are determined. Then, the calculated measurement values ​​and calculation weights of the multi-source measurement data at the estimation time are obtained, and measurement constraints between the multi-source measurement data and state variables, as well as zero-injection constraints between zero-injection nodes and state variables, are constructed. The objective function is the weighted minimum absolute value of the residuals between the calculated and estimated values ​​of the multi-source measurement data. The robust state estimation problem at the estimation time is solved using the second-order cone programming algorithm. Finally, the estimation results of the state variables are converted into node voltage amplitude and phase angle estimation results. During this process, the identification of bad measurement data is completed.

[0006] Defects of the Background Technology While existing patents in my country have addressed the problem of bad data identification in power systems to some extent, their application in vehicles, especially in vehicle power systems operating under harsh conditions, exhibits significant performance deficiencies and limitations. Traditional identification methods in power systems are largely based on static or quasi-static model assumptions. When directly applied to vehicles, these methods fail to fully consider the severe challenges posed by the inherent characteristics of onboard power systems to bad data identification. Compared to large-scale ground-based power grids, vehicle power systems face the following problems in bad data identification: (1) The system operating conditions are highly dynamic and nonlinear. During operation, the vehicle experiences frequent acceleration and deceleration, sudden load changes, and energy recovery, and the system parameters are in a state of continuous and drastic change. This strong dynamic characteristic undermines the static or quasi-static model assumptions on which traditional identification methods rely, resulting in model mismatch and a significant decrease in identification accuracy and convergence speed.

[0007] (2) The real-time requirements for computation are extremely stringent. Vehicle safety control and state estimation need to be decided within milliseconds. However, many existing methods rely on complex iterative calculations, and the computational load increases sharply with the number of measurement points, making it difficult to meet the real-time requirements in the vehicle environment. This may lead to delayed response and missing the best processing opportunity.

[0008] (3) Harsh operating conditions can easily lead to multi-point, coupled bad data. The strong vibration, wide temperature range and complex electromagnetic interference environment of vehicles can easily cause multiple sensors to malfunction at the same time, and there may be correlation between these abnormal data. This can cause the phenomenon of "residual contamination", that is, a serious bad data will affect the residuals of other normal measurements, causing misjudgment and omission of identification algorithms, and limiting processing capabilities.

[0009] Chinese patent (CN110750575A: Bad Data Identification Method for Integrated Electrothermal Energy Systems Based on Lagrange State Estimator) describes a method that constructs a Lagrange state estimation model, iteratively solves for, and identifies bad data. However, its drawbacks include: the model contains a large number of nonlinear constraint equations (such as the heating network temperature drop equation and the loop pressure drop equation), requiring repeated iterations to calculate the residual sensitivity matrix and regularized residuals during the solution process. In large-scale electrothermal interconnected systems, the computational load increases exponentially with the number of nodes and measurements, making real-time performance difficult to guarantee; furthermore, this method is highly dependent on the initial state estimate. If the initial value deviates significantly from the true value, it will lead to slow iteration convergence and even local optima, affecting the accuracy of bad data identification.

[0010] Chinese patent (CN116305764A A Method for Identifying Bad Data and Estimating State) describes a method that identifies leverage measurements and bad data through measurement equations. However, its drawbacks are as follows: the identification of leverage measurements relies on the projection statistics method, which requires traversing the row vectors of the Jacobian matrix. When the system topology is complex and the measurement dimension is high, the Jacobian matrix becomes enormous, significantly increasing the computation time. At the same time, the bad data identification model is sensitive to the linearization approximation of the measurement equations. When the power system operates under highly nonlinear conditions (such as high load and large disturbances), the linearization error accumulates, leading to inaccurate calculation of the normalized residuals and affecting the accuracy of bad data identification.

[0011] Analysis of the shortcomings of the Chinese patent (CN105224361A, "A Method for Identifying Bad Data in Power Systems Based on an Improved Lagrange Multiplier Method"): The core of this method is to use the Lagrange multiplier method to locate bad data. However, this method relies on transforming bad data identification into a constrained optimization problem, and its effectiveness is highly dependent on the accuracy of the model and the stationarity of the data. In vehicle power systems, operating conditions (such as acceleration / deceleration and sudden load changes) are highly dynamic, leading to continuous changes in system parameters and unstable data distribution. This strong dynamic characteristic undermines the static or quasi-static model assumptions upon which the Lagrange multiplier method relies, resulting in a decrease in its identification accuracy and convergence speed. Furthermore, this method has limited ability to handle multiple simultaneous and coupled bad data, making it prone to misjudgment and missed judgment. Summary of the Invention

[0012] (a) Technical problems to be solved The technical problem to be solved by this invention is: how to provide an online monitoring, panoramic perception, and bad data estimation and cleaning technology for vehicle power information under harsh working conditions, so as to overcome the defects of the existing technology: existing online vehicle monitoring technologies are mostly for single types of data, making it difficult to achieve multi-dimensional data fusion perception, and are easily affected by interference under harsh working conditions, leading to measurement distortion; traditional bad data processing methods (such as deletion method, mean interpolation method, etc.) will lead to large deviations in analysis results when the data missing rate is high or there are a large number of outliers, and cannot effectively identify residual pollution problems caused by strong coupling relationships.

[0013] (II) Technical Solution To address the aforementioned technical problems, this invention provides a method for identifying bad data in a vehicle's electrical system, such as... Figure 1 As shown, the identification method treats the vehicle's electrical system as an observable physical system and uses panoramic data perception to verify the accuracy of the data. The identification method includes the following steps: Step 1: Pre-screen the collected raw measurement data; Step 2: Construct a multi-physics domain panoramic perception model; Step 3: Solve based on state estimation to achieve panoramic perception; Step 4: Identify bad data using the maximum standardized residual method; Step 5: Iteratively clean and intelligently complete the bad data.

[0014] In step one, the collected raw measurement data is pre-screened. "Explicitly bad data" such as invalid timestamps, values ​​outside the reasonable range, or abnormal spatial correlations are removed to form a valid dataset.

[0015] In step two, a multi-physical domain panoramic perception model is constructed. Based on real-time acquisition of multi-dimensional monitoring data of the vehicle's electrical system from onboard terminals and sensor networks, a measurement-state mapping model integrating electromagnetic and thermodynamic principles is constructed. This model not only includes directly measurable measurements but also introduces implicit state variables. The core of multi-physical domain panoramic perception is to combine the laws of multi-physical domains with virtual measurements, which can ensure accuracy while significantly improving solution efficiency. The specific construction method is as follows: Step 21: Determine the system's measured quantities and state quantities; Measured quantities z These are parameters that the sensor can directly collect, state variables. x It cannot be directly measured but can reflect the inherent characteristics of the system; Step 22: Construct a basic measurement-state mapping physical model; the reference model is... z = h ( x ) + e ,in z The representative measurement vector represents the raw data that the sensor can directly collect. h (·) represents a measurement function, which is a mathematical model that encapsulates the physical laws of the system and describes the state variables. x How to determine the theoretical quantity to be measured z ; e Represents the measurement error vector; Step 23: Introduce pseudo-quantity measurement to simplify the calculation structure, and convert the dq-axis voltage... u d , u q and magnetic flux ψ d ψ q Using these as pseudo-quantities for measurement, and simultaneously adding their corresponding dynamic constraint equations, this step significantly reduces the complexity of the Jacobian matrix in subsequent solutions, and is key to achieving millisecond-level state estimation.

[0016] In step three, panoramic perception is achieved by solving based on state estimation. The weighted least squares (WLS) state estimation algorithm is used to iteratively solve the model constructed in step two to obtain the optimal estimate of the system's global state, thereby achieving accurate perception of unmeasurable parameters, including rotor angle and battery health status. The specific process of the weighted least squares state estimation algorithm is as follows: Figure 2 As shown; In step four, the maximum standardized residual method is used to identify bad data. The state estimate obtained in step three is compared with the actual measurement value to calculate the standardized residual reflecting the degree of deviation. The maximum standardized residual value is located and it is determined whether it exceeds the preset threshold. Based on this, the current bad data is identified. In step four, in order to identify bad data and obtain the measurement residual r, the state estimate obtained in step three is compared with the actual measurement value to calculate the residual calculation part of the standardized residual reflecting the degree of deviation.

[0017] In step five, bad data is iteratively cleaned and intelligently completed. An integrated mechanism of "identification-removal-completion" is adopted to remove the identified bad data from the dataset, and to complete the data based on the historical state estimation results. The process is repeated until all residuals are less than the threshold. Step 5 performs removal, completion, return to re-estimation and termination judgment.

[0018] The "bad data identification and removal process based on state estimation" formed in steps four and five is as follows: Figure 3 As shown; The specific operational procedures for maximum residual iteration removal in steps four and five are as follows: To address the "residual contamination" problem caused by the coupling of multiple abnormal data under harsh operating conditions, an iterative elimination mechanism based on the maximum standardized residual is proposed to ensure that the error sources with the greatest impact are located and eliminated first. Suppose that the standardized residual set calculated at a certain moment is R. N : The judgment and execution process of this mechanism can be represented as follows: (1) Locating the maximum residual: from set R N Find the standardized residual with the largest absolute value in the middle. and its corresponding first i Individual measurements; (2) Threshold judgment: Compare with a preset confidence threshold; (3) Identification and Removal: If If the threshold is exceeded, then the corresponding [number] [item] is determined. iIf a measurement is found to be bad data at the current moment, it will be immediately removed from the measurement dataset. (4) Data completion: The mean of the first 3 valid state estimation results is used to complete the data points that were removed, so as to ensure the continuity of the data sequence; (5) Iterative loop: Return to step three, and re-perform state estimation and residual analysis based on the updated dataset. Repeat this process until set R is reached. N The absolute value of all residuals is less than the threshold.

[0019] The method described above achieves accurate and rapid identification of bad data in vehicle electrical system monitoring data by constructing a system physical model and using a state estimation algorithm. (III) Beneficial Effects This invention achieves panoramic perception by integrating multi-dimensional monitoring data and constructing measurement equations, and accurately identifies and cleans bad data based on a standardized residual mechanism for state estimation, thereby improving the accuracy and reliability of vehicle power information monitoring under harsh operating conditions.

[0020] Compared with existing technologies, this invention introduces a closed-loop mechanism of "pre-selection-panoramic perception-iterative cleaning" in the identification of bad data in vehicle power systems. It is the first to transplant the pre-screening concept for complex topologies to the dynamic environment of vehicles: first, lightweight physical rules are used to quickly eliminate obviously abnormal measurements, reducing the solution space; then, an electromagnetic-thermodynamic multi-physics domain state estimation model is constructed, explicitly incorporating implicit states such as unmeasurable rotor angles and state of harm (SOH) into the residual calculation; finally, using the maximum standardized residual as a pointer, bad data is iteratively eliminated and completed, preventing residual contamination. These three steps work together to achieve high-precision bad data location and repair within milliseconds, balancing real-time performance and robustness.

[0021] Compared with existing technologies, this invention significantly improves the overall performance of bad data identification in vehicle power systems. Through multi-physical domain panoramic perception modeling, integrating multi-source monitoring data and introducing pseudo-measurements and dynamic constraints, computational complexity is greatly reduced, achieving millisecond-level state estimation and effectively adapting to dynamically changing vehicle operating conditions. Furthermore, a standardized residual mechanism based on weighted least squares estimation is employed, combined with a maximum residual iterative elimination and intelligent historical data completion strategy, to accurately identify bad data and prevent residual contamination. This significantly improves identification accuracy and system robustness, while ensuring data continuity and control reliability, providing solid support for the safe and stable operation of vehicles in environments with strong interference and high dynamics. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the overall technical solution of the present invention.

[0023] Figure 2 This is a flowchart of the WLS state estimation algorithm used in this invention.

[0024] Figure 3 This is a flowchart for identifying bad data based on state estimation. Detailed Implementation

[0025] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0026] To address the aforementioned technical problems, this invention provides a method for identifying bad data in a vehicle's electrical system, such as... Figure 1 As shown, the identification method treats the vehicle's electrical system as an observable physical system and uses panoramic data perception to verify the accuracy of the data. The identification method includes the following steps: Step 1: Pre-screen the collected raw measurement data; Step 2: Construct a multi-physics domain panoramic perception model; Step 3: Solve based on state estimation to achieve panoramic perception; Step 4: Identify bad data using the maximum standardized residual method; Step 5: Iteratively clean and intelligently complete the bad data.

[0027] In step one, the collected raw measurement data is pre-screened. "Explicitly bad data" such as invalid timestamps, values ​​outside the reasonable range, or abnormal spatial correlations are removed to form a valid dataset.

[0028] In step two, a multi-physical domain panoramic perception model is constructed. Based on real-time acquisition of multi-dimensional monitoring data of the vehicle's electrical system from onboard terminals and sensor networks, a measurement-state mapping model integrating electromagnetic and thermodynamic principles is constructed. This model not only includes directly measurable measurements but also introduces implicit state variables. The core of multi-physical domain panoramic perception is to combine the laws of multi-physical domains with virtual measurements, which can ensure accuracy while significantly improving solution efficiency. The specific construction method is as follows: Step 21: Determine the system's measured quantities and state quantities; Measured quantities z These are parameters that the sensor can directly collect, state variables. x It cannot be directly measured but can reflect the inherent characteristics of the system; Step 22: Construct a basic measurement-state mapping physical model; the reference model is... z = h ( x ) + e ,in z The representative measurement vector represents the raw data that the sensor can directly collect. h(·) represents a measurement function, which is a mathematical model that encapsulates the physical laws of the system and describes the state variables. x How to determine the theoretical quantity to be measured z ; e Represents the measurement error vector; Step 23: Introduce pseudo-quantity measurement to simplify the calculation structure, and convert the dq-axis voltage... u d , u q and magnetic flux ψ d ψ q Using these as pseudo-quantities for measurement, and simultaneously adding their corresponding dynamic constraint equations, this step significantly reduces the complexity of the Jacobian matrix in subsequent solutions, and is key to achieving millisecond-level state estimation.

[0029] In step three, panoramic perception is achieved by solving based on state estimation. The weighted least squares (WLS) state estimation algorithm is used to iteratively solve the model constructed in step two to obtain the optimal estimate of the system's global state, thereby achieving accurate perception of unmeasurable parameters, including rotor angle and battery health status. The specific process of the weighted least squares state estimation algorithm is as follows: Figure 2 As shown; In step four, the maximum standardized residual method is used to identify bad data. The state estimate obtained in step three is compared with the actual measurement value to calculate the standardized residual reflecting the degree of deviation. The maximum standardized residual value is located and it is determined whether it exceeds the preset threshold. Based on this, the current bad data is identified. In step four, in order to identify bad data and obtain the measurement residual r, the state estimate obtained in step three is compared with the actual measurement value to calculate the residual calculation part of the standardized residual reflecting the degree of deviation.

[0030] In step five, bad data is iteratively cleaned and intelligently completed. An integrated mechanism of "identification-removal-completion" is adopted to remove the identified bad data from the dataset, and to complete the data based on the historical state estimation results. The process is repeated until all residuals are less than the threshold. Step 5 performs removal, completion, return to re-estimation and termination judgment.

[0031] The "bad data identification and removal process based on state estimation" formed in steps four and five is as follows: Figure 3 As shown; The specific operational procedures for maximum residual iteration removal in steps four and five are as follows: To address the "residual contamination" problem caused by the coupling of multiple abnormal data under harsh operating conditions, an iterative elimination mechanism based on the maximum standardized residual is proposed to ensure that the error sources with the greatest impact are located and eliminated first. Suppose that the standardized residual set calculated at a certain moment is R. N : The judgment and execution process of this mechanism can be represented as follows: (1) Locating the maximum residual: from set R N Find the standardized residual with the largest absolute value in the middle. and its corresponding first i Individual measurements; (2) Threshold judgment: Compare with a preset confidence threshold; (3) Identification and Removal: If If the threshold is exceeded, then the corresponding [number] [item] is determined. i If a measurement is found to be bad data at the current moment, it will be immediately removed from the measurement dataset. (4) Data completion: The mean of the first 3 valid state estimation results is used to complete the data points that were removed, so as to ensure the continuity of the data sequence; (5) Iterative loop: Return to step three, and re-perform state estimation and residual analysis based on the updated dataset. Repeat this process until set R is reached. N The absolute value of all residuals is less than the threshold.

[0032] The method described above achieves accurate and rapid identification of bad data in vehicle electrical system monitoring data by constructing a system physical model and using a state estimation algorithm.

[0033] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying bad data in a vehicle electrical system, characterized in that, The identification method treats the vehicle's electrical system as an observable physical system and uses panoramic data perception to verify the accuracy of the data. The identification method includes the following steps: Step 1: Pre-screen the collected raw measurement data; Step 2: Construct a multi-physics domain panoramic perception model; Step 3: Solve based on state estimation to achieve panoramic perception; Step 4: Identify bad data using the maximum standardized residual method; Step 5: Iteratively clean and intelligently complete the bad data.

2. The method for identifying bad data in a vehicle electrical system as described in claim 1, characterized in that, In step one, the collected raw measurement data is pre-screened; "Explicitly bad data" such as invalid timestamps, values ​​outside the reasonable range, or abnormal spatial correlations are removed to form a valid dataset.

3. The method for identifying bad data in a vehicle electrical system as described in claim 2, characterized in that, In step two, a multi-physical domain panoramic perception model is constructed. Based on real-time acquisition of multi-dimensional monitoring data of the vehicle's electrical system from onboard terminals and sensor networks, a measurement-state mapping model integrating electromagnetic and thermodynamic principles is constructed. This model not only includes directly measurable measurements but also introduces implicit state variables. The core of multi-physical domain panoramic perception is to combine the laws of multi-physical domains with virtual measurements, which can ensure accuracy while significantly improving solution efficiency. The specific construction method is as follows: Step 21: Determine the system's measured quantities and state quantities; Measured quantities z These are parameters that the sensor can directly collect, state variables. x It cannot be directly measured but can reflect the inherent characteristics of the system; Step 22: Construct a basic measurement-state mapping physical model; the reference model is... z = h ( x ) + e ,in z The representative measurement vector represents the raw data that the sensor can directly collect. h (·) represents a measurement function, which is a mathematical model that encapsulates the physical laws of the system and describes the state variables. x How to determine the theoretical quantity to be measured z ; e Represents the measurement error vector; Step 23: Introduce pseudo-quantity measurement to simplify the calculation structure, and convert the dq-axis voltage... u d , u q and magnetic flux ψ d ψ q Using these as pseudo-quantities for measurement, and simultaneously adding their corresponding dynamic constraint equations, this step significantly reduces the complexity of the Jacobian matrix in subsequent solutions, and is key to achieving millisecond-level state estimation.

4. The method for identifying bad data in a vehicle electrical system as described in claim 3, characterized in that, In step three, panoramic perception is achieved by solving based on state estimation. The model constructed in step two is iteratively solved using a weighted least squares state estimation algorithm to obtain the optimal estimation result of the global state of the system, thereby achieving accurate perception of unmeasurable parameters, including rotor angle and battery health status.

5. The method for identifying bad data in a vehicle electrical system as described in claim 4, characterized in that, In step four, the maximum standardized residual method is used to identify bad data; The state estimate obtained in step three is compared with the actual measurement value to calculate the standardized residual reflecting the degree of deviation. The maximum standardized residual value is located and it is determined whether it exceeds the preset threshold. Based on this, the current bad data is identified. In step four, in order to identify bad data and obtain the measurement residual r, the state estimate obtained in step three is compared with the actual measurement value to calculate the residual calculation part of the standardized residual reflecting the degree of deviation.

6. The method for identifying bad data in a vehicle electrical system as described in claim 5, characterized in that, In step five, bad data is iteratively cleaned and intelligently completed. An integrated "identification-removal-completion" mechanism is adopted to remove the identified bad data from the dataset, and to complete the data based on the historical state estimation results. The process returns to step three to re-estimate the state, and repeats until all residuals are less than the threshold. Step five executes removal, completion, return to re-estimation, and termination judgment.

7. The method for identifying bad data in a vehicle electrical system as described in claim 6, characterized in that, The specific operational procedures for maximum residual iteration removal in steps four and five are as follows: To address the "residual contamination" problem caused by the coupling of multiple abnormal data under harsh operating conditions, an iterative elimination mechanism based on the maximum standardized residual is proposed to ensure that the error sources with the greatest impact are located and eliminated first. Suppose that the standardized residual set calculated at a certain moment is R. N : The judgment and execution process of this mechanism can be represented as follows: (1) Locating the maximum residual: from set R N Find the standardized residual with the largest absolute value in the middle. and its corresponding first i Individual measurements; (2) Threshold judgment: Compare with a preset confidence threshold; (3) Identification and Removal: If If the threshold is exceeded, then the corresponding [number] [item] is determined. i If a measurement is found to be bad data at the current moment, it will be immediately removed from the measurement dataset. (4) Data completion: The mean of the first 3 valid state estimation results is used to complete the data points that were removed, so as to ensure the continuity of the data sequence; (5) Iterative loop: Return to step three, and re-perform state estimation and residual analysis based on the updated dataset. Repeat this process until set R is reached. N The absolute value of all residuals is less than the threshold.

8. The method for identifying bad data in a vehicle electrical system as described in claim 7, characterized in that, The method constructs a system physical model and utilizes a state estimation algorithm to achieve accurate and rapid identification of bad data in vehicle electrical system monitoring data.

9. The method for identifying bad data in a vehicle electrical system as described in claim 7, characterized in that, The method belongs to the field of vehicle electrical system technology.

10. The method for identifying bad data in a vehicle electrical system as described in claim 7, characterized in that, The method achieves panoramic perception by integrating multi-dimensional monitoring data and constructing measurement equations, and accurately identifies and cleans bad data based on the standardized residual mechanism of state estimation, thereby improving the accuracy and reliability of vehicle power information monitoring under harsh working conditions.

Images