Control method for dynamically adjusting kalman filter parameters

Abstract

This application proposes a control method for dynamically adjusting Kalman filter parameters, relating to the field of industrial control system design. Addressing the shortcomings of existing technologies, such as poor signal and noise filtering in the early stages of data processing and poor control system stability caused by signal distortion, this application creatively dynamically adjusts the Kalman filter parameters. This achieves maximum filtering of external noise while preserving the integrity of the original sensor information, continuously optimizing the estimated output of the system state in dynamically changing environments. The data acquisition module obtains information based on the sensor array type; the multi-sensor fusion algorithm module transmits the information acquired by the data acquisition module to the adaptive Kalman filter, which fuses the heterogeneous information. Based on a mathematical model, the adaptive Kalman filter dynamically adjusts the Kalman filter gain according to the front-end data, dynamically adjusting the estimated output state.

Description

Technical Field

[0001] This application relates to the field of industrial control system design, specifically proposing a method for multi-sensor control based on a Kalman filter.

[0002] A control method for adaptive adjustment through data fusion is applicable to data integration, noise reduction, and optimization in various industrial scenarios. Background Technology

[0003] With the widespread use of new energy technologies, UV printing has become the preferred solution for insulating coatings of power battery cells, based on the technology of zero pollution emissions and recyclability of batteries.

[0004] In existing industrial automation scenarios, it is common to encounter various stand-alone devices equipped with multiple sensors of different types. Different types of sensors upload sensor information data of different standards to the upper level, which then integrates them through the decision-making capabilities of the upper level. By filtering and classifying this information, real-time and comprehensive information support is provided to the control system.

[0005] However, the process of integrating various types of information requires a series of steps, including data acquisition, data preprocessing, data feature recognition, and fusion technology algorithms. If information deviation occurs during the data acquisition or transmission process, it will have a strong negative impact on the stability of the control system.

[0006] Kalman filters are a widely used noise reduction filtering tool. They utilize the state equations of a linear system and employ optimal estimation algorithms to remove noise and restore accurate data. During the system design review phase, engineers need to design different fusion algorithms based on the complexity of the sensing system and the application scenario to ensure the stability and efficiency of the control system.

[0007] However, existing technologies based on Kalman filters for noise reduction and filtering still have significant shortcomings. On the one hand, the data noise is complex and uncontrollable, which negatively burdens data processing and affects the accuracy of data analysis. On the other hand, the poor standardization of sensor information and problems such as missing or distorted data make it impossible to achieve the ideal fusion state.

[0008] In view of the above, this application is hereby submitted. Summary of the Invention

[0009] The control method for dynamically adjusting Kalman filter parameters described in this application aims to solve the defects of the prior art, such as poor signal and noise filtering effect in the early stage of data processing and poor stability of the control system caused by signal distortion. It proposes an optimized data noise processing method, creatively dynamically adjusting the Kalman filter parameters, thereby achieving maximum filtering of external noise while preserving the integrity of the original sensing information, and continuously optimizing the estimated output of the system state in a dynamically changing environment.

[0010] To achieve the above-mentioned objective, the control method for dynamically adjusting the parameters of the Kalman filter employs the following control system:

[0011] The data acquisition module, as the input receiving layer of the front-end sensing link, is responsible for receiving real-time information from various sensors and sending all known data to the data processing layer through various sensing interface protocols to provide information support for decision-making.

[0012] The state-space model connects to the data acquisition module and the multi-sensor fusion algorithm module, and provides a reasonable allocation mode for the decision-making layer by analyzing the changing state of the system.

[0013] The adaptive Kalman filter adjusts the predicted state based on information uploaded by the data acquisition module, adopts a correction detection window to monitor the residual between the actual observation state and the predicted observation state in real time, and corrects the observation noise covariance matrix by comparing it with the uncertain state of the observation space, thereby achieving dynamic optimization of the filter.

[0014] The multi-sensor fusion algorithm module, based on adaptive Kalman filter data fusion, enables synchronous modeling and processing of heterogeneous sensors;

[0015] Status output device, outputs the processing results of the control system;

[0016] Based on the aforementioned control system, the information collected from the front end is gradually decomposed and optimized, and the optimal information output by the system is used for closed-loop self-checking and evaluation. This includes the following steps: the data acquisition module performs state-space mathematical modeling based on the information obtained from the sensor array type; the multi-sensor fusion algorithm module transmits the information data collected by the data acquisition module to the adaptive Kalman filter, which then performs fusion processing on the heterogeneous information; based on the mathematical model, the adaptive Kalman filter dynamically adjusts the Kalman filter gain according to the front-end data, and dynamically adjusts the estimated output state.

[0017] Includes the following steps:

[0018] Step (1): Identify the sensor system array;

[0019] The data acquisition module determines the communication protocol and content of the sensor, and determines the output dimensions of the model to be built;

[0020] Step (2), state space modeling;

[0021] Define the following control system: ;

[0022] in, Let k be the system state at time k. The system state at time k-1 is... Here is the state transition matrix. To control the input matrix, For the control matrix of the external system;

[0023] Define the following observation system: ;

[0024] in, Let H be the measurement matrix at time k, and H be the parameter matrix of the measurement system. It is the observation noise matrix;

[0025] Step (3): Establish a prediction mechanism and a correction mechanism;

[0026] Based on the state space in step (2), the current state of the system is set as k. The current state can be predicted from the previous state based on the system model. That is, the state prediction process is as follows: ;

[0027] in, Let be the prior state estimate at time k. Let be the posterior state estimate at time k-1. The external input matrix at time k-1;

[0028] Step (4): Establish a prediction error evaluation algorithm and introduce Kalman filter gain calculation;

[0029] Covariance is used to evaluate the prediction error at time k;

[0030] In summary, the mathematical description of the Kalman filter gain is as follows:

[0031]

[0032] in, For Kalman filter gain, Let the prior estimate of the covariance at time k be . It is the transpose matrix. R is the sum of the uncertainty of the predicted state and the uncertainty of the measurement error, and R is the covariance matrix of the observation noise, representing the uncertainty of the measurement error.

[0033] Step (5): Establish the adaptive Kalman filter algorithm and update the state uncertainty;

[0034] Define T as the continuous time interval parameter, and set the correction time window to... From the above equation, we get: ;

[0035] Its covariance within the correction window is:

[0036]

[0037] The above formula describes the real-time fluctuations between the actual observed values ​​and the predicted values ​​within a preset rolling window;

[0038] According to the description of Kalman filter gain, in the limiting case, the total error should be infinitely close to the residual between the actual observation and the predicted value, then:

[0039]

[0040] Finally, the observation noise covariance matrix at the current time is obtained. as follows:

[0041]

[0042] Based on the updated observation noise covariance matrix as follows:

[0043]

[0044] The posterior state estimate at time k, i.e., the optimal estimate, is:

[0045]

[0046] Steps (2) to (5) are executed repeatedly. By combining real observation data and historical prediction data, the Kalman filter gain is dynamically adjusted through prediction and correction, and the estimated output of the system state is continuously optimized in a dynamically changing environment.

[0047] Step (3) introduces the following correction step:

[0048] Define a correction factor Based on the observation system, the posterior state estimate at time k is corrected using a correction coefficient. The correction process is as follows: ;

[0049] in, Let H be the posterior estimate at time k, where H represents the relationship between the state and the observation. For observed values, correction coefficients That is, the Kalman filter gain; This is described as the residual between the actual observed state and the predicted observed state. The Kalman filter gain is added to the residual, indicating the weight of the residual model in the correction stage.

[0050] In step (4), let's assume... Let be the prior estimate covariance at time k, representing the uncertainty of the predicted state; The posterior estimated covariance at time k-1 is... The covariance represents the final uncertainty after verification of the measurements at time k-1;

[0051] Based on the state transition matrix of the control system, the error evaluation algorithm is as follows:

[0052]

[0053] in, Let be the process noise covariance matrix, representing the error between the state transition and the actual process;

[0054] The error evaluation algorithm uses the known uncertainty of the state at the previous time step to predict the uncertainty of the state at the current time step.

[0055] According to step (3), the Kalman filter gain directly determines the weights of the predicted state and the measured state, describing the proportion of the prediction uncertainty in the total error system under the measurement space; let R be the observation noise covariance matrix used to represent the measurement error uncertainty; according to the error evaluation algorithm and the observation system, the sum of the prediction state uncertainty and the measurement error uncertainty at time k under the observation space is: At time k in the observation space, through the transpose matrix Mapping to the observation space, the prediction uncertainty is written as .

[0056] Step (5) involves predicting the uncertainty of the state at time k based on step (4). The posterior covariance at time k represents the uncertainty of the observed state. The error update equation is as follows:

[0057]

[0058] in, It is an identity matrix used to maintain the consistency of the dimensional space; state uncertainty. Used to record the uncertainty of the state at a new time k, which is then used for evaluation at the next time k+1.

[0059] In summary, this application has the following advantages and beneficial effects compared with the prior art:

[0060] 1. This application exhibits stronger robustness. By utilizing an adaptively adjusted Kalman filter, it can better adapt to complex operating conditions, dynamically adjusting the gain and parameters even for unknown model parameters. Compared to existing technologies, this application no longer focuses on the determinism of internal model parameters, making it more adaptable to complex and ever-changing scenarios.

[0061] 2. This application significantly improves the accuracy of noise reduction filtering. For uncontrollable observation noise and process noise, if the characteristics of the two types of noise change abruptly during the process, the existing technology will not change the gain calculation method, which will lead to output imbalance. This application maintains high-precision output by adjusting internal parameters in real time to ensure that the system no longer relies on fixed setting values. Attached Figure Description

[0062] The following figures will now be used to explain and illustrate the technical solution proposed in this application in detail.

[0063] Figure 1 This is a block diagram of the control system for dynamically adjusting the parameters of the Kalman filter as described in this application;

[0064] Figure 2 This is a schematic diagram of the data acquisition module;

[0065] Figure 3 This is a schematic diagram of the state-space model;

[0066] Figure 4 This is a schematic diagram of an adaptive Kalman filter;

[0067] Figure 5 This is a schematic diagram of a multi-sensor fusion algorithm module; Detailed Implementation

[0068] Example 1: This application proposes a control method for multi-sensor system fusion using a Kalman filter.

[0069] The method adopts the following control system framework:

[0070] The data acquisition module, as the input receiving layer of the front-end sensing link, is responsible for receiving real-time information from various sensors and sending all known data to the data processing layer through various sensing interface protocols to provide information support for decision-making.

[0071] The state-space model connects to the data acquisition module and the multi-sensor fusion algorithm module, and provides a reasonable allocation mode for the decision-making layer by analyzing the changing state of the system.

[0072] The adaptive Kalman filter adjusts the predicted state based on information uploaded by the data acquisition module, adopts a correction detection window to monitor the residual between the actual observation state and the predicted observation state in real time, and corrects the observation noise covariance matrix by comparing it with the uncertain state of the observation space, thereby achieving dynamic optimization of the filter.

[0073] The multi-sensor fusion algorithm module, based on adaptive Kalman filter data fusion, enables synchronous modeling and processing of heterogeneous sensors;

[0074] Status output device, outputs the processing results of the control system;

[0075] The control method for dynamically adjusting the parameters of the Kalman filter proposed in this application is based on the above control system, which gradually decomposes and optimizes the information collected at the front end, and performs closed-loop self-testing and evaluation with the optimal information output by the system.

[0076] Specifically, the data acquisition module performs the next step of state-space mathematical modeling based on the information obtained from the sensor array type (specifically, heterogeneous sensor array or homogeneous redundant sensor array).

[0077] The multi-sensor fusion algorithm module transmits the information data collected by the data acquisition module to the adaptive Kalman filter, which then fuses the heterogeneous information.

[0078] Based on the mathematical model, the adaptive Kalman filter dynamically adjusts the Kalman filter gain according to the front-end data, and dynamically adjusts the estimated output state, making it more adaptable to dynamic environments and with higher filtering accuracy.

[0079] Includes the following steps:

[0080] Step (1): Identify the sensor system array;

[0081] The data acquisition module determines the communication protocol and content of the sensor to ensure that the data input is authentic and valid;

[0082] If the sensor array contains heterogeneous sensors, it means that the system model has multidimensional input and output; if there are no heterogeneous sensors, it is a one-dimensional input and output; the control system and observation system are set up according to the sensor array type.

[0083] The purpose of this step is to determine the output dimensions of the model that needs to be built;

[0084] Step (2), state space modeling;

[0085] Define the following control system: ;

[0086] in, Let k be the system state at time k. The system state at time k-1 is... Here is the state transition matrix. To control the input matrix, For the control matrix of the external system;

[0087] Define the following observation system: ;

[0088] in, Let H be the measurement matrix at time k, and H be the parameter matrix of the measurement system. It is the observation noise matrix;

[0089] Step (3): Establish a prediction mechanism and a correction mechanism;

[0090] Based on the state space in step (2), the current state of the system is set as k. The current state can be predicted from the previous state based on the system model. That is, the state prediction process is as follows: ;

[0091] in, Let be the prior state estimate at time k. Let be the posterior state estimate at time k-1. The external input matrix at time k-1;

[0092] The above formula describes the current time data estimated by the system based on the data of the previous time. However, the prediction process is subject to external and system interference, and the measurement data itself is susceptible to multivariate noise. Therefore, the following correction process is introduced.

[0093] Define a correction factor Based on the observation system, the posterior state estimate at time k is corrected using a correction coefficient. The correction process is as follows: ;

[0094] in, Let H be the posterior estimate at time k, where H represents the relationship between the state and the observation. For observed values, correction coefficients That is, the Kalman filter gain; This is described as the residual between the actual observed state and the predicted observed state. Adding the Kalman filter gain to this residual indicates the weight of the residual model in the correction stage.

[0095] Based on the above correction steps, if the measurement status is more reliable, a larger Kalman filter gain should be set; if the measurement status is more susceptible to interference, it depends more on the predicted status and a smaller Kalman filter gain should be set.

[0096] Step (4): Establish a prediction error evaluation algorithm and introduce Kalman filter gain calculation;

[0097] There exists an "error state" between the predicted and observed states, which directly affects the Kalman filter gain and thus directly determines the system's stability. In multivariate statistical analysis, covariance describes the degree of correlation between two random variables;

[0098] Therefore, this application uses covariance to evaluate the prediction error at time k;

[0099] set up Let be the prior estimate covariance at time k, representing the uncertainty of the predicted state; The posterior estimated covariance at time k-1 is... The covariance represents the final uncertainty after verification of the measurements at time k-1;

[0100] Based on the state transition matrix of the control system, the error evaluation algorithm is as follows:

[0101]

[0102] in, Let be the process noise covariance matrix, representing the error between the state transition and the actual process;

[0103] As can be seen from the formula, the error assessment algorithm uses the known uncertainty of the state at the previous moment to predict the uncertainty of the state at the current moment; this step is also called the error propagation step, which is used to quantify the uncertainty of the predicted state, thereby providing theoretical support for the Kalman filter gain, and to decide on the appropriate weights for the predicted state and the observed state.

[0104] According to step (3), the Kalman filter gain directly determines the weights of the predicted state and the measured state, describing the proportion of the prediction uncertainty in the total error system under the measurement space;

[0105] Let R be the observation noise covariance matrix used to represent the uncertainty of measurement error;

[0106] Based on the error assessment algorithm and the observation system, the sum of the predicted state uncertainty and the measurement error uncertainty at time k in the observation space can be obtained as follows: R is the covariance matrix of the observation noise; at time k, in the observation space, through the transpose matrix... Mapping to the observation space, the prediction uncertainty is written as ;

[0107] In summary, the mathematical description of the Kalman filter gain is as follows:

[0108]

[0109] The purpose of this step is to obtain the error propagation model based on the conclusion of step (3) and introduce a model describing the proportion of the prediction uncertainty in the total error system under the measurement space, namely the Kalman filter gain.

[0110] Step (5): Establish the adaptive Kalman filter algorithm and update the state uncertainty;

[0111] In the Kalman filter algorithm, once the model training is complete, the gain will be locked immediately. If noise changes in the observation space, it will cause filter misalignment, resulting in instability of the control system.

[0112] To this end, T is defined as the continuous time interval parameter, and the correction time window is set to... From the above equation, we get: ;

[0113] Its covariance within the correction window is:

[0114]

[0115] The above formula describes the real-time fluctuations between the actual observed values ​​and the predicted values ​​within a preset rolling window;

[0116] According to the description of Kalman filter gain, in the limiting case, the total error should be infinitely close to the residual between the actual observation and the predicted value, then:

[0117]

[0118] Finally, the observation noise covariance matrix at the current time is obtained. as follows:

[0119]

[0120] Based on the updated observation noise covariance matrix as follows:

[0121]

[0122] The posterior state estimate at time k, i.e., the optimal estimate, is:

[0123]

[0124] The uncertainty of the state is predicted based on the k-time prediction in step (4). The posterior covariance at time k represents the uncertainty of the observed state. The error update equation is as follows:

[0125]

[0126] in, It is an identity matrix used to maintain the consistency of the dimensional space; state uncertainty. Used to record the uncertainty of the state at a new time k, for evaluation at the next time k+1;

[0127] The purpose of this step is to improve the Kalman filter gain based on its determinism, thereby introducing the adaptive Kalman filter algorithm and ultimately obtaining the state uncertainty at the new time as a reference index for the system evaluation at the next time.

[0128] Steps (2) to (5) are executed repeatedly. By combining real observation data and historical prediction data, the Kalman filter gain is dynamically adjusted through prediction and correction, and the estimated output of the system state is continuously optimized in a dynamically changing environment.

[0129] The foregoing provides an overall explanation of the control method for dynamically adjusting Kalman filter parameters proposed in this application. This method employs an improved Kalman filter multi-sensor fusion control system, capable of adapting to complex and changing operating conditions. To illustrate the feasibility of the control method in this application, the following specific implementation scenario is provided.

[0130] Suppose a multi-dimensional state tracking scenario exists with two states in a given operating condition, and a heterogeneous sensor array exists. The control architecture allocation is determined based on this scenario.

[0131] For tracking scenarios with two states, a two-dimensional control system needs to be established. Heterogeneous sensor arrays represent two types of observation models, requiring the establishment of a two-dimensional state observation model. Communication connections need to be established to ensure the data acquisition module can receive all information correctly.

[0132] Based on the model, a two-dimensional control system model can be established as follows:

[0133]

[0134] Two-dimensional observation systems include:

[0135]

[0136] The prediction process based on the model is as follows:

[0137]

[0138] The correction process is established as follows:

[0139]

[0140] The two-dimensional systematic error propagation algorithm is established as follows:

[0141]

[0142] Based on the model, establish an adaptive update strategy for the multidimensional filtering system:

[0143] Adaptive Kalman gain includes:

[0144]

[0145] The output of the multidimensional optimal state estimation is:

[0146]

[0147] The error update equations are:

[0148]

[0149] As described above, the embodiments given in conjunction with the accompanying drawings are merely preferred solutions for achieving the objectives of this invention. Those skilled in the art can draw inspiration from this and directly derive other alternative structures that conform to the design concept of this invention. Other structural features derived therefrom should also fall within the scope of the solutions described in this invention.

Claims

1. A control method for dynamically adjusting the parameters of a Kalman filter, characterized in that: The following control system is adopted. The data acquisition module, as the input receiving layer of the front-end sensing link, is responsible for receiving real-time information from various sensors and sending all known data to the data processing layer through various sensing interface protocols to provide information support for decision-making. The state-space model connects to the data acquisition module and the multi-sensor fusion algorithm module, and provides a reasonable allocation mode for the decision-making layer by analyzing the changing state of the system. The adaptive Kalman filter adjusts the predicted state based on information uploaded by the data acquisition module, adopts a correction detection window to monitor the residual between the actual observation state and the predicted observation state in real time, and corrects the observation noise covariance matrix by comparing it with the uncertain state of the observation space, thereby achieving dynamic optimization of the filter. The multi-sensor fusion algorithm module, based on adaptive Kalman filter data fusion, enables synchronous modeling and processing of heterogeneous sensors; Status output device, outputs the processing results of the control system; Based on the above control system, the information collected at the front end is gradually decomposed and optimized, and the optimal information output by the system is used for closed-loop self-testing and evaluation. This includes the next step of state-space mathematical modeling, which is performed by the data acquisition module based on the information obtained from the sensor array type. The multi-sensor fusion algorithm module transmits the information data collected by the data acquisition module to the adaptive Kalman filter, which then performs fusion processing on the heterogeneous information. Based on the mathematical model, the adaptive Kalman filter dynamically adjusts the Kalman filter gain according to the front-end data, and dynamically adjusts the estimated output state.

2. The control method for dynamically adjusting Kalman filter parameters according to claim 1, characterized in that: Includes the following steps, Step (1): Identify the sensor system array; The data acquisition module determines the communication protocol and content of the sensor, and determines the output dimensions of the model to be built; Step (2), state space modeling; Define the following control system: ; in, Let k be the system state at time k. The system state at time k-1, Here is the state transition matrix. To control the input matrix, For the control matrix of the external system; Define the following observation system: ; in, Let H be the measurement matrix at time k, and H be the parameter matrix of the measurement system. It is the observation noise matrix; Step (3): Establish a prediction mechanism and a correction mechanism; Based on the state space in step (2), the current state of the system is set as k. The current state can be predicted from the previous state based on the system model. That is, the state prediction process is as follows: ; in, Let k be the prior state estimate at time k. Let be the posterior state estimate at time k-1. The external input matrix at time k-1; Step (4): Establish a prediction error evaluation algorithm and introduce Kalman filter gain calculation; Covariance is used to evaluate the prediction error at time k; In summary, the mathematical description of the Kalman filter gain is as follows: in, For Kalman filter gain, Let the prior estimate of the covariance at time k be . It is the transpose matrix. R is the sum of the uncertainty of the predicted state and the uncertainty of the measurement error, and R is the covariance matrix of the observation noise, representing the uncertainty of the measurement error. Step (5): Establish the adaptive Kalman filter algorithm and update the state uncertainty; Define T as the continuous time interval parameter, and set the correction time window to... From the above equation, we get: ; Its covariance within the correction window is: The above formula describes the real-time fluctuations between the actual observed values ​​and the predicted values ​​within a preset rolling window; According to the description of Kalman filter gain, in the limiting case, the total error should be infinitely close to the residual between the actual observation and the predicted value, then: Finally, the observation noise covariance matrix at the current time is obtained. as follows: Based on the updated observation noise covariance matrix as follows: The posterior state estimate at time k, i.e., the optimal estimate, is: Steps (2) to (5) are executed repeatedly. By combining real observation data and historical prediction data, the Kalman filter gain is dynamically adjusted through prediction and correction, and the estimated output of the system state is continuously optimized in a dynamically changing environment.

3. The control method for dynamically adjusting Kalman filter parameters according to claim 2, characterized in that: Step (3) introduces the following correction step: Define a correction factor Based on the observation system, the posterior state estimate at time k is corrected using a correction coefficient. The correction process is as follows: ; in, Let H be the posterior estimate at time k, where H represents the relationship between the state and the observation. For observed values, correction coefficients That is, the Kalman filter gain; This is described as the residual between the actual observed state and the predicted observed state. The Kalman filter gain is added to the residual, indicating the weight of the residual model in the correction stage.

4. The control method for dynamically adjusting Kalman filter parameters according to claim 2, characterized in that: In step (4), let's assume... Let be the prior estimate covariance at time k, representing the uncertainty of the predicted state; The posterior estimated covariance at time k-1 is... The covariance represents the final uncertainty after verification of the measurements at time k-1; Based on the state transition matrix of the control system, the error evaluation algorithm is as follows: in, Let be the process noise covariance matrix, representing the error between the state transition and the actual process; The error assessment algorithm uses the known uncertainty of the state at the previous time step to predict the uncertainty of the state at the current time step; according to step (3), the Kalman filter gain directly determines the weights of the predicted state and the measured state, describing the measurement space. The proportion of prediction uncertainty in the total error system at time k; let R be the observation noise covariance matrix used to represent the measurement error uncertainty; based on the error assessment algorithm and the observation system, the sum of prediction state uncertainty and measurement error uncertainty at time k in the observation space can be obtained as follows: At time k in the observation space, through the transpose matrix Mapping to the observation space, the prediction uncertainty is written as .

5. The control method for dynamically adjusting Kalman filter parameters according to claim 2, characterized in that: Step (5) involves predicting the uncertainty of the state at time k based on step (4). The posterior covariance at time k represents the uncertainty of the observed state. The error update equation is as follows: in, It is an identity matrix used to maintain the consistency of the dimensional space; state uncertainty. Used to record the uncertainty of the state at a new time k, which is then used for evaluation at the next time k+1.

Images