Fusion algorithm based on improved weighted Kalman data

Abstract

The invention relates to the technical field of data filtering and data fusion, in particular to an improved weighted Kalman data fusion algorithm. The algorithm comprises the following steps: establishing a state equation, calculating a state prediction value at the current moment according to state estimation of current / voltage data at the previous moment to become priori, and calculating an error covariance matrix at the same time; obtaining a prediction covariance matrix at the current moment through matrix operation; calculating an observation noise covariance matrix; calculating to obtain a Kalman gain; calculating to obtain an updated optimal state estimation value; calculating a current optimal estimation value and an optimal estimation value covariance matrix; calculating a weighting factor, and compensating and updating the covariance matrix of the current / voltage data; calculating to obtain an updated covariance matrix; and then loop iteration is carried out on the steps, so that the estimation precision of the system state is gradually improved. The fusion algorithm provided by the invention can improve the data precision and accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of data filtering and data fusion, and in particular to an improved weighted Kalman data fusion algorithm. Background Art

[0002] With the development of science and technology, the acquisition of information has become more convenient, the amount of data transmission has continued to increase, and data fusion technology has also been gradually improved. At present, there have been a series of studies on data fusion. Among them, Zhang Pingchuan et al. designed a multi-sensor data fusion algorithm combining Kalman filtering, Euclidean distance formula and multivariate clustering statistics in the article Design and Application of Multi-Sensor Data Fusion Algorithm for Greenhouse Internet of Things. This method is highly dependent on the grouping method of sensors. If the grouping is different, the results of data fusion will be different, and the errors will also be different. Li Shaonian et al. proposed an improved Kalman data fusion algorithm in the article Research on Greenhouse Internet of Things Collection System Based on Improved Kalman Data Fusion Algorithm. After preprocessing the collected data using the Romanovsky criterion, the Kalman filter algorithm was used to statically fuse the data at the coordinator node, but this is only a simple elimination of the data. After filtering the deviated data, it will affect the authenticity of the data. Tan Baocheng et al. studied the weighted average data fusion algorithm based on the multi-sensor system of unmanned vehicles in the article "Research on Weighted Average Data Fusion Algorithm Based on Unmanned Vehicle Sensor System". The sensors can work together to make up for the shortcomings of a single sensor and better play the overall performance. However, this algorithm may still have a large deviation from the actual situation without compensation and update. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] The present invention provides a data fusion algorithm based on improved weighted Kalman to solve the problem that clutter noise is easily mixed into the data collected by the sensor, which increases the difficulty of data processing and reduces the true accuracy of the collected data.

[0005] (II) Technical solution

[0006] To solve the above problems, the present invention provides an improved weighted Kalman data fusion algorithm comprising:

[0007] (1) Establish the state equation, calculate the current state prediction value based on the state estimation of the current / voltage data at the previous moment, and calculate the error covariance matrix at the same time;

[0008] According to the state transfer equation and the covariance matrix of the previous moment, the prediction covariance matrix of the current moment is obtained through matrix operation; and the observation noise covariance matrix is ​​calculated through sensor observation data;

[0009] (2) The Kalman gain is calculated by multiplying the prediction covariance matrix by the observation noise covariance matrix and the inverse matrix of the sum of the observation noise covariance matrix and the prediction covariance matrix;

[0010] By predicting the state estimate, the observed data and the Kalman gain, the updated optimal state estimate is calculated; the current optimal estimate and the optimal estimate covariance matrix are calculated;

[0011] (3) Calculate weighting factors for each data sensor used to compensate and update the covariance matrix of current / voltage data;

[0012] The updated covariance matrix is ​​calculated by using the Kalman gain, the compensation updated prediction covariance matrix and the observation noise covariance matrix;

[0013] The above steps are then iterated to gradually improve the estimation accuracy of the system state.

[0014] Preferably, the state estimation formula based on the Kalman filter algorithm prediction value is:

[0015] X(k|k-1)=A×X(k-1|k-1)+B×U(k) (1-1)

[0016] In the formula, X(k|k-1) is the result obtained by using the previous state prediction; X(k-1|k-1) is the optimal result of the previous state;

[0017] A is the state control matrix; B is the input control item matrix; U(k) is the control quantity of the state at time K;

[0018] Preferably, the covariance matrix of the predicted value is calculated, and the covariance matrix estimation formula of the predicted value is P(k|k-1)=A×P(k-1|k-1)×AT+Q(1-2)

[0019] Where P(k|k-1) is the covariance corresponding to X(k|k-1); P(k-1|k-1) is the covariance corresponding to X(k-1|k-1); Q is the covariance of the system noise; and the estimated value is then corrected based on the prior estimated value and the measured value.

[0020] Preferably, the state estimation formula of the Kalman filter update value is

[0021] X(k|k)=X(k|k-1)+G(k)×[Z(k)-H×X(k|k-1)] (1-3)

[0022] Where H is the measurement transmission matrix, X(k|k) is the current optimal estimate, G is the Kalman gain,

[0023] The calculation is G(k) = P(k|k-1) × HTH × P(k|k-1) × HT + R (1-4)

[0024] Where R is the measurement noise covariance;

[0025] Preferably, the optimal state estimation formula for the covariance update value is

[0026] P(k|k)=[IG(k)×H]P(k|k-1) (1-5)

[0027] Where P(k|k) is the covariance of the current optimal estimate; I is the unit matrix.

[0028] Preferably, weighted fusion is performed on each sensor. When the sensor is a current sensor and current parameters are collected, assuming that the measured current is in an environment where the covariance is relatively stable and small, firstly, 10 data are measured by each sensor to calculate the respective covariances;

[0029] Sort the covariance values, cut off the last 10% of the data, and pre-set a weight based on this part of the data, that is,

[0030] According to the principle of minimum co-square error, the weighting factor of each current sensor is calculated:

[0031]

[0032] Fusion of current sensors:

[0033]

[0034]

[0035] Finally, the fusion value of the current is obtained for calibration;

[0036] Then by

[0037] After diagonalizing the matrix H k Get: Diagonal variance matrix

[0038] Abnormality

[0039] Compensate the posterior covariance matrix according to the degree of abnormality:

[0040] P k =P k|k-1 -K k H k P k|k-1 +∈ k βk K k H k P k|k-1 (1-12)

[0041] where ∈ k =1~2,

[0042] The voltage parameters are calculated and calibrated as above.

[0043] (III) Beneficial effects

[0044] The improved weighted Kalman data fusion algorithm provided by the present invention collects data through multiple sensors, which effectively reduces the large deviation of the data collected by a single sensor that is prone to failure. The measurement points involved in the actual measurement site are mostly distributed outdoors, and the environmental temperature and humidity have an impact on the measurement accuracy of the probe. At the same time, on the basis of the Kalman filter algorithm, multiple sensors are used to jointly detect the target, and the corresponding weighting factors are calculated according to the principle of minimum mean square error. The feedforward information weighted fusion algorithm is used to overcome the disadvantages of global optimal fusion information pollution of the main filter, and then the covariance matrix is ​​compensated and updated through the abnormality, so as to improve the data precision and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The figure is a flow chart of an improved weighted Kalman data fusion algorithm according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] like Figure 1 As shown, the present invention provides an improved weighted Kalman data fusion algorithm, which specifically includes:

[0048] (1) Establish the state equation, calculate the current state prediction value based on the state estimation of the current / voltage data at the previous moment, and calculate the error covariance matrix at the same time;

[0049] According to the state transfer equation and the covariance matrix of the previous moment, the prediction covariance matrix of the current moment is obtained through matrix operation. And through the sensor observation data, the observation noise covariance matrix is ​​calculated;

[0050] (2) The Kalman gain is calculated by multiplying the prediction covariance matrix by the observation noise covariance matrix and the inverse matrix of the sum of the observation noise covariance matrix and the prediction covariance matrix;

[0051] By predicting the state estimate, the observed data, and the Kalman gain, the updated optimal state estimate is calculated. The current optimal estimate and the optimal estimate covariance matrix are calculated;

[0052] (3) Calculate weighting factors for each data sensor used to compensate and update the covariance matrix of current / voltage data;

[0053] The updated covariance matrix is ​​calculated by using the Kalman gain, the compensation updated prediction covariance matrix and the observation noise covariance matrix;

[0054] These steps are then iterated to gradually improve the accuracy of the estimation of the system state.

[0055] The above steps specifically include:

[0056] Preferably, the state estimation formula based on the Kalman filter algorithm prediction value is:

[0057] X(k|k-1)=A×X(k-1|k-1)+B×U(k) (1-1)

[0058] In the formula, X(k|k-1) is the result obtained by using the previous state prediction; X(k-1|k-1) is the optimal result of the previous state;

[0059] A is the state control matrix; B is the input control item matrix; U(k) is the control quantity of the state at time K;

[0060] Preferably, the covariance matrix of the predicted value is calculated, and the covariance matrix estimation formula of the predicted value is P(k|k-1)=A×P(k-1|k-1)×AT+Q(1-2)

[0061] Where P(k|k-1) is the covariance corresponding to X(k|k-1); P(k-1|k-1) is the covariance corresponding to X(k-1|k-1); Q is the covariance of the system noise; and the estimated value is then corrected based on the prior estimated value and the measured value.

[0062] Preferably, the state estimation formula of the Kalman filter update value is

[0063] X(k|k)=X(k|k-1)+G(k)×[Z(k)-H×X(k|k-1)] (1-3)

[0064] Where H is the measurement transmission matrix, X(k|k) is the current optimal estimate, G is the Kalman gain,

[0065] The calculation is G(k) = P(k|k-1) × HTH × P(k|k-1) × HT + R (1-4)

[0066] Where R is the measurement noise covariance;

[0067] Preferably, the optimal state estimation formula for the covariance update value is

[0068] P(k|k)=[IG(k)×H]P(k|k-1) (1-5)

[0069] Where P(k|k) is the covariance of the current optimal estimate; I is the unit matrix.

[0070] Preferably, weighted fusion is performed on each sensor. When the sensor is a current sensor and current parameters are collected, assuming that the measured current is in an environment where the covariance is relatively stable and small, firstly, 10 data are measured by each sensor to calculate the respective covariances;

[0071] Sort the covariance values, cut off the last 10% of the data, and pre-set a weight based on this part of the data, that is,

[0072] According to the principle of minimum co-square error, the weighting factor of each current sensor is calculated:

[0073]

[0074] Fusion of current sensors:

[0075]

[0076]

[0077] Finally, the fusion value of the current is obtained for calibration;

[0078] Then by

[0079] After diagonalizing the matrix H k Get: Diagonal variance matrix

[0080] Abnormality

[0081] Compensate the posterior covariance matrix according to the degree of abnormality:

[0082] P k =P k|k-1 -K k H k P k|k-1+∈ k β k K k H k P k|k-1 (1-12)

[0083] Where ∈k=1~2,

[0084] The voltage parameters are calculated and calibrated as above.

[0085] Through the fusion of data collected by multiple sensors, it is effectively reduced that the data collected by a single sensor is prone to failure and is prone to large deviations. The actual measurement sites are mostly distributed outdoors, and the environmental temperature and humidity have an impact on the measurement accuracy of the probe. At the same time, based on the Kalman filter algorithm, multiple sensors are used to jointly detect the target, and the corresponding weighting factors are calculated according to the principle of minimum mean square error. Through the feedforward information weighted fusion algorithm, the disadvantages of global optimal fusion information pollution of the main filter are overcome, and the covariance matrix is ​​compensated and updated through the abnormality, so as to improve the data precision and accuracy.

[0086] The above implementation modes are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention. The patent protection scope of the present invention should be defined by the claims.

Claims

1. An improved weighted Kalman data fusion algorithm, characterized in that: include: (1) Establish the state equation, calculate the state prediction value at the current moment according to the state estimation of the current / voltage data at the previous moment, and use it as a priori, and calculate the error covariance matrix at the same time; According to the state transfer equation and the covariance matrix of the previous moment, the prediction covariance matrix of the current moment is obtained through matrix operation; And through the sensor observation data, the observation noise covariance matrix is ​​calculated; (2) The Kalman gain is calculated by multiplying the prediction covariance matrix by the observation noise covariance matrix and the inverse matrix of the sum of the observation noise covariance matrix and the prediction covariance matrix; By predicting the state estimate, the observed data and the Kalman gain, the updated optimal state estimate is calculated; the current optimal estimate and the optimal estimate covariance matrix are calculated; (3) Calculate weighting factors for each data sensor used to compensate and update the covariance matrix of current / voltage data; The updated covariance matrix is ​​calculated by using the Kalman gain, the compensation updated prediction covariance matrix and the observation noise covariance matrix; The above steps are then iterated to gradually improve the estimation accuracy of the system state.

2. The improved weighted Kalman data fusion algorithm according to claim 1, characterized in that: The state estimation formula based on the predicted value of the Kalman filter algorithm is: X(k|k-1)=A×X(k-1|k-1)+B×U(k) (1-1) In the formula, X(k|k-1) is the result obtained by using the previous state prediction; X(k-1|k-1) is the optimal result of the previous state; A is the state control matrix; B is the input control item matrix; U(k) is the control quantity of the state at time K.

3. The improved weighted Kalman data fusion algorithm according to claim 2, characterized in that: Calculate the covariance matrix of the predicted value. The covariance matrix estimation formula of the predicted value is P(k|k-1)=A×P(k-1|k-1)×AT+Q (1-2) Where P(k|k-1) is the covariance corresponding to X(k|k-1); P(k-1|k-1) is the covariance corresponding to X(k-1|k-1); Q is the covariance of the system noise; and the estimated value is then corrected based on the prior estimated value and the measured value.

4. The improved weighted Kalman data fusion algorithm according to claim 3, characterized in that: The state estimation formula of the Kalman filter update value is X(k|k)=X(k|k-1)+G(k)×[Z(k)-H×X(k|k-1)] (1-3) Where H is the measurement transmission matrix, X(k|k) is the current optimal estimate, G is the Kalman gain, The calculation is G(k) = P(k|k-1) × HTH × P(k|k-1) × HT + R(1-4) where R is the measurement noise covariance.

5. The improved weighted Kalman data fusion algorithm according to claim 4, characterized in that: The optimal state estimate of the covariance update value is P(k|k)=[IG(k)×H]P(k|k-1) (1-5) Where P(k|k) is the covariance of the current optimal estimate; I is the unit matrix.

6. The improved weighted Kalman data fusion algorithm according to claim 2, characterized in that: Perform weighted fusion on each sensor. When the sensor is a current sensor and collects current parameters, assume that the measured current is in an environment with a relatively stable and small covariance. First, measure 10 data from each sensor to calculate their respective covariances. Sort the covariance values, cut off the last 10% of the data, and pre-set a weight based on this part of the data, that is, According to the principle of minimum co-square error, the weighting factor of each current sensor is calculated: Fusion of current sensors: Finally, the fusion value of the current is obtained for calibration; Then by After diagonalizing the matrix H k Get: Diagonal variance matrix Abnormality Compensate the posterior covariance matrix according to the degree of abnormality: P k =P k|k-1 -K k H k P k|k-1 +∈ k β k K k H k P k|k-1 (1-12) in The voltage parameters are calculated and calibrated as above.

Images