A data fusion method for multi-sensor control systems based on iterative correction optimization

Abstract

The present invention discloses a data fusion method for a multi-sensor control system based on iterative correction optimization, comprising the following steps: 1. Correcting data collected by multiple sensors; 2. Adaptive weight allocation; 3. Multi-sensor control system data fusion output. The present invention is aimed at the real-time measurement values ​​given by each sensor in the control system, can detect abnormal values ​​and correct them without affecting the normal values, and then quickly allocate weights based on the trust between the measurement values ​​and output the fusion result. Compared with the fixed weighted fusion algorithm, the adaptive weighted fusion function with iterative correction optimization function can judge the degree of deviation of each measurement value according to the measured data, and then detect and correct the abnormal values ​​in the measured data group, make full use of each output information, and finally determine the appropriate weight allocation strategy. It has strong real-time performance, fault tolerance and robustness, and has high fusion accuracy and low system calculation amount.

Description

[0001] Technology Neighborhood

[0002] The present invention belongs to the field of data fusion technology, and in particular relates to a multi-sensor control system data fusion method based on iterative correction optimization. Background Art

[0003] A control system composed of multiple heterogeneous sensors is an information control system. It is sensitive to a wider range of environmental information, thereby improving the control system's anti-interference capability and operational reliability. Each sensor operates at a different mechanism and is sensitive to different environmental information. Through its own data processing, it converts this environmental information into system control information. The control information provided by different sensors possesses the redundancy, correlation, and complementarity required for data fusion. Therefore, a multi-sensor control system is a data fusion system with multiple information sources. The system control device comprehensively processes the control information from each sensor and generates a final action signal. In summary, the decision-making process in a multi-sensor control system can be summarized as how to utilize the local decision values ​​and prior information of each sensor to make a decision on the system's overall action signal—in other words, the question of which algorithm to adopt for data fusion. In a multi-sensor control system, each sensor has different sensitive information, operating mechanisms, anti-interference capabilities, and measurement accuracy. Using different types of sensors to detect a target and obtain comprehensive information yields a more accurate target state estimate than a single sensor alone. The two main fusion decision-making models commonly used in information control systems are simple arithmetic averaging and minimum mean square error. The minimum mean square error method utilizes prior information about the accuracy of each sensor, resulting in a fusion decision with statistically minimal variance. In summary, both methods assign fixed weights to each sensor before the system operates, resulting in low computational complexity. However, if individual sensors, particularly high-precision sensors, experience interference and produce abnormal measurements, the fusion results can exhibit significant deviations. Therefore, improving system robustness without significantly increasing the algorithm's computational complexity has become a hot topic in the development of multi-sensor control systems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the deficiencies in the above-mentioned prior art and provide a multi-sensor control system data fusion method based on iterative correction optimization. For the real-time measurement values ​​given by each sensor in the control system, the method can detect abnormal values ​​and correct them without affecting the normal values. Then, the weights are quickly allocated based on the trust between the measurement values ​​and the fusion results are output. Compared with the fixed weighted fusion algorithm, the adaptive weighted fusion function with iterative correction optimization function can judge the degree of deviation of each measurement value according to the measured data, and then detect and correct the abnormal values ​​in the measured data group, make full use of each output information, and finally determine the appropriate weight allocation strategy. It has strong real-time performance, fault tolerance and robustness, high fusion accuracy, small system calculation amount, and is easy to promote and use.

[0005] To solve the above technical problems, the present invention adopts a technical solution: a multi-sensor control system data fusion method based on iterative correction optimization, characterized in that the method comprises the following steps:

[0006] Step 1: Correct the data collected by multiple sensors. The process is as follows:

[0007] Step 101: Obtain the measurement value H collected by each sensor in the control system m , where m is the number of each sensor and m=1,2,...,N, and N is the total number of sensors;

[0008] Step 102: According to the formula Calculate the difference d between each measurement m,n 、H m The average distance to each measurement The average distance between all measurements and the average of all measurements Among them, the value range of n is the same as that of m;

[0009] Step 103: Check whether the measured value is the average value of all measured values. A neighborhood of Within, determine whether the measured value is an abnormal value;

[0010] The average of all measurements A neighborhood of Range

[0011] When the measured value is in a neighborhood of the mean value H of all measured values When the measured value is within the range of A neighborhood of When the value is outside the range, the measured value is an outlier. Record the numbers i of all outliers and build a set S. Then, the abnormal value is corrected and step 104 is executed;

[0012] Step 104: According to the formula Correct abnormal measurement value H i , get the abnormal measurement value H i Correction value H i ', using H i 'Replace the corresponding abnormal measurement value H i ;

[0013] Step 105: Repeat steps 102 to 104. During the repetition process, only H needs to be checked. i Whether it is within the updated neighborhood range, the measurement value judged as normal in the previous round does not need to participate in the next round of inspection. If the abnormal measurement value H i If the test is passed, it is considered that the output value of sensor i has been corrected to the normal value, and its serial number element in the set S is removed, and the above process is repeated until the set

[0014] Each time step 102 to step 104 is cycled, the differences between all measured values ​​and the new H m Update the neighborhood range by the average distance between each measurement value, the average distance between all measurements, and the average of all measurements;

[0015] Step 2: Adaptive weight allocation. The process is as follows:

[0016] Step 201: According to the formula Calculate the corrected measurement value H n The corrected measured value H m Trust level L n (H m ), where Δ n is the threshold value;

[0017] Step 202: According to the formula Calculate the corrected measurement value H m The total trust m(H m );

[0018] Step 203: According to the formula For m(H m ) to obtain the transformed trust

[0019] Step 204: According to the formula Get the corrected measurement value H m The weight ωm , where ∑ω m =1;

[0020] Step 3: Multi-sensor control system data fusion output: According to the formula Obtain the data fusion output H0 of the multi-sensor control system.

[0021] The above-mentioned multi-sensor control system data fusion method based on iterative correction optimization is characterized in that: the threshold value H n The average distance to each measurement value, and The acquisition process is the same.

[0022] The beneficial effect of the present invention is that, for the real-time measurement values ​​given by each sensor in the control system, it can detect abnormal values ​​and correct them without affecting the normal values, and then quickly allocate weights based on the trust between the measurement values ​​and output the fusion results. Compared with the fixed weighted fusion algorithm, the adaptive weighted fusion method with iterative correction and optimization function can judge the degree of deviation of each measurement value according to the measured data, and then detect and correct the abnormal values ​​in the measured data group, make full use of each output information, and finally determine the appropriate weight allocation strategy. It has strong real-time performance, fault tolerance and robustness, high fusion accuracy, small system calculation amount, and is easy to promote and use.

[0023] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flowchart of the method of the present invention.

[0025] Figure 2 Schematic diagram of dense data and outlier data in the present invention. DETAILED DESCRIPTION

[0026] like Figure 1 and Figure 2 As shown, a multi-sensor control system data fusion method based on iterative correction optimization of the present invention includes the following steps:

[0027] Step 1: Correct the data collected by multiple sensors. The process is as follows:

[0028] Step 101: Obtain the measurement value H collected by each sensor in the control system m , where m is the number of each sensor and m=1,2,...,N, and N is the total number of sensors;

[0029] Step 102: According to the formula Calculate the difference d between each measurement value m,n 、H m The average distance to each measurement The average distance between all measurements and the average of all measurements Among them, the value range of n is the same as that of m;

[0030] Step 103: Check whether the measured value is the average value of all measured values. A neighborhood of Within, determine whether the measured value is an abnormal value;

[0031] The average of all measurements A neighborhood of Range

[0032] When the measured value is the average of all measured values A neighborhood of When the measured value is within the range of A neighborhood of When the value is outside the range, the measured value is an outlier. Record the numbers i of all outliers and build a set S. Then, the abnormal value is corrected and step 104 is executed;

[0033] Step 104: According to the formula Correct abnormal measurement value H i , get the abnormal measurement value H i Correction value H i ', using H i 'Replace the corresponding abnormal measurement value H i ;

[0034] Step 105: Repeat steps 102 to 104. During the repetition process, only H needs to be checked. i Whether it is within the updated neighborhood range, the measurement value judged as normal in the previous round does not need to participate in the next round of inspection. If the abnormal measurement value H i If the test is passed, it is considered that the output value of sensor i has been corrected to the normal value, and its serial number element in the set S is removed, and the above process is repeated until the set

[0035] Each time step 102 to step 104 is cycled, the differences between all measured values ​​and the new H m Update the neighborhood range by the average distance between each measurement value, the average distance between all measurements, and the average of all measurements;

[0036] It should be noted that the corrected value will replace the original measurement value to participate in the fusion decision of the control system. Other measurement values ​​that are not identified as abnormal values ​​should remain unchanged. In actual measurement, after an abnormal value is corrected, it may still be an abnormal value. Therefore, it is necessary to loop steps 102 to 104. Each time steps 102 to 104 are looped, the sequence number of each measurement value remains unchanged. Based on the corrected sequence value, the differences between all measurement values ​​and the new H are recalculated. m Update the neighborhood range to the average distance between each measurement, the average distance between all measurements, and the average of all measurements.

[0037] Step 2: Adaptive weight allocation. The process is as follows:

[0038] Step 201: According to the formula Calculate the corrected measurement value H n The corrected measured value H m Trust level L n (H m ), where Δ n is the threshold value;

[0039] Step 202: According to the formula Calculate the corrected measurement value H m The total trust m(H m );

[0040] Step 203: According to the formula For m(H m ) to obtain the transformed trust

[0041] It should be noted that in order to make the weight distribution of each sensor more reasonable, it is necessary to use mathematical methods to transform the trust. Considering that if there are fewer information sources involved in decision-making in the control system, it is necessary to further weaken the impact of measurement values ​​with relatively large deviations and the high real-time requirements of the system, this transformation method amplifies the role of normal measurement values ​​without significantly increasing the computational complexity, while also weakening the impact of measurement values ​​with relatively large deviations.

[0042] Step 204: According to the formula Get the corrected measurement value H m The weight ω m , where ∑ω m =1;

[0043] It should be noted that after the preprocessing of the correction link, although all the measured values ​​are in an appropriate range, the measured values ​​can still be divided into two categories: the part that is relatively close to the true value and the part that is relatively far from the true value. The former is defined as "dense data" and the latter is defined as "outlier data". The distribution of these data is as follows: Figure 2 According to the above definition, "dense data" is distributed near the measured true value, while "outlier data" is relatively far away from the true value. Therefore, the algorithm that calculates weights based entirely on real-time measurement data must meet the following two requirements:

[0044] First, the measurement data that belong to “dense data” should support each other because they are relatively close to the true value;

[0045] Second, the measurement data belonging to "outlier data" is relatively far away from the true value, so it is not supported by "dense data" and needs to support "dense data" as a penalty.

[0046] Step 3: Multi-sensor control system data fusion output: According to the formula Obtain the data fusion output H0 of the multi-sensor control system.

[0047] In this embodiment, the threshold value H n The average distance to each measurement value, and The acquisition process is the same.

[0048] when When Δ n =max{d m,n}, it means that the measurement value belonging to “outlier data” needs to be supported by the measurement values ​​given by all sensors, where and max{d m,n} are all obtained by calculation from the measurement values ​​given by the sensor itself, and will change with the changes in the measurement data output at every moment. They are highly adaptable and avoid the subjective factors brought about by reliance on prior information.

[0049] When the present invention is used, no prior information of the sensor measurement value is required, each measured data is fully utilized, and the threshold and weight allocation strategy are determined according to the mutual relationship between the measured data, thereby avoiding the subjective factors brought by setting the empirical threshold, and being able to reconstruct the outliers through iterative correction, thereby greatly reducing the negative impact of the outliers in the data fusion process, being able to fully utilize the measurement values ​​contributed by each sensor, and being able to transform the outliers to a normal level through iterative correction, thereby making up for the defects of other adaptive algorithms, making the fusion structure of the system at different times relatively stable, having higher fusion accuracy and robustness than other adaptive algorithms, and having small computational complexity, thus meeting the high real-time requirements of the system.

[0050] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A data fusion method for a multi-sensor control system based on iterative correction optimization, characterized in that: The method comprises the following steps: Step 1: Correct the data collected by multiple sensors. The process is as follows: Step 101: Obtain the measurement value H collected by each sensor in the control system m , where m is the number of each sensor and m=1,2,...,N, and N is the total number of sensors; Step 102: According to the formula Calculate the difference d between each measurement m,n 、H m The average distance to each measurement The average distance between all measurements and the average of all measurements Among them, the value range of n is the same as that of m; Step 103: Check whether the measured value is the average value of all measured values. A neighborhood of Within, determine whether the measured value is an abnormal value; The average of all measurements A neighborhood of Range When the measured value is the average of all measured values A neighborhood of When the measured value is within the range of A neighborhood of When the value is outside the range, the measured value is an outlier. Record the numbers i of all outliers and build a set S. Then, the abnormal value is corrected and step 104 is executed; Step 104: According to the formula Correct abnormal measurement value H i , get the abnormal measurement value H i Correction value H′ i , using H′ i Replace the corresponding abnormal measurement value H i ; Step 105: Repeat steps 102 to 104. During the repetition process, only H needs to be checked. i Whether it is within the updated neighborhood range, the measurement value judged as normal in the previous round does not need to participate in the next round of inspection. If the abnormal measurement value H i If the test is passed, it is considered that the output value of sensor i has been corrected to the normal value, and its serial number element in the set S is removed, and the above process is repeated until the set Each time step 102 to step 104 is cycled, the differences between all measured values ​​and the new H m Update the neighborhood range by the average distance between each measurement value, the average distance between all measurements, and the average of all measurements; Step 2: Adaptive weight allocation. The process is as follows: Step 201: According to the formula Calculate the corrected measurement value H n The corrected measured value H m Trust level L n (H m ), where Δ n is the threshold value; Step 202: According to the formula Calculate the corrected measurement value H m The total trust m(H m ); Step 203: According to the formula For m(H m ) to obtain the transformed trust Step 204: According to the formula Get the corrected measurement value H m The weight ω m , where ∑ω m =1; Step 3: Multi-sensor control system data fusion output: According to the formula Obtain the data fusion output H0 of the multi-sensor control system.

2. The data fusion method for a multi-sensor control system based on iterative correction optimization according to claim 1, characterized in that: The threshold value H n The average distance to each measurement value, and The acquisition process is the same.

Images