River flow full-scale monitoring method based on multi-source data fusion

Abstract

The invention discloses a river flow full-scale monitoring method based on multi-source data fusion, which comprises the following steps of: (1) collecting real-time data obtained by a plurality of automatic flow monitoring methods according to different water level amplitudes of a river; (2) when the real-time data obtained by any flow automatic monitoring method within a set time period is analyzed and found to have a data missing value, the weight of each flow automatic monitoring value is recalculated; (3) when judging that the real-time data obtained by any flow automatic monitoring method exceeds a historical maximum error threshold value, filtering out a flow result obtained by the monitoring method through automatic screening; and (4) on the basis of the real-time data obtained by the various flow automatic monitoring methods, through a Kalman filtering flow fusion algorithm based on dynamic variance, obtaining fusion result data of the various flow real-time data. The sudden data missing condition can be optimized and adjusted, and the continuity, accuracy and stability of automatic flow monitoring data are ensured.

Description

Technical Field

[0001] The present invention relates to the field of intelligent analysis of river flow measurement data, and in particular to a full-range monitoring method for river flow based on multi-source data fusion. Background Art

[0002] Existing methods for automatic river flow monitoring include radar, imaging, acoustic Doppler, ultrasonic time-difference, hydraulic structure analysis, and the unification of water level-discharge relationships. Research on automatic river flow monitoring has mostly focused on the application of a single method, but has not yet explored full-scale automatic flow monitoring using multi-source data fusion.

[0003] Due to the impacts of climate change, ship traffic, and other factors, as well as limitations in the development of hydrological monitoring technology, automated flow monitoring in rivers with low flow rates and high flooding conditions is challenging. Most hydrological stations struggle to accurately monitor the full range of river flow using a single method. Different flow measuring instruments perform differently in different river water environments. Because river water environments vary over time, the variance of flow measured by each instrument fluctuates, and the corresponding flow monitoring parameters are dynamically adjusted. This makes single flow measurement methods and instruments inaccurate. In actual field measurements, flow measurement results may occasionally have short-term missing values due to rare factors such as changes in the water environment, communication interruptions, and equipment failure. Conventional Kalman filtering algorithms are unable to handle missing data. Therefore, it is necessary to optimize and adjust conventional Kalman filtering algorithms to address these missing values, ensure algorithm stability and continuity, and further improve measurement accuracy. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, the present invention proposes a full-scale monitoring method for river flow based on multi-source data fusion. It adopts a multi-source data fusion algorithm to realize full-scale automatic monitoring of river flow, fuses the real-time data provided by multiple automatic flow monitoring methods, optimizes and adjusts occasional sudden data missing values, ensures the continuity and stability of automatic flow monitoring data, and further improves the accuracy of measurement results.

[0005] The present invention provides a full-range monitoring method for river flow based on multi-source data fusion, comprising the steps of:

[0006] (1) Collecting real-time data from various automatic flow monitoring methods for different water level fluctuations in rivers;

[0007] (2) When the analysis finds that any of the real-time data obtained by the automatic flow monitoring method has missing data within the set time period, the weight of each automatic flow monitoring value is recalculated, and in the Kalman filter algorithm, the calculation of the Kalman gain, the optimal estimate value and the state covariance matrix are all optimized and adjusted accordingly;

[0008] (3) When it is determined that the real-time data obtained by any automatic flow monitoring method exceeds the historical maximum error threshold, the flow results obtained by the monitoring method that exceeds the historical maximum error threshold are automatically filtered out and are not included in the flow fusion calculation in the next step;

[0009] (4) Based on the real-time data obtained by the above-mentioned multiple flow automatic monitoring methods, a fusion result data of multiple flow real-time data is obtained through a Kalman filter flow fusion algorithm based on dynamic variance.

[0010] Furthermore, the method step (ii) specifically includes:

[0011] When the analysis finds that there are missing values in the real-time data obtained by any of the automatic flow monitoring methods within the set time period, the weight of each automatic flow monitoring value will be recalculated, that is, the weight corresponding to the missing value will be removed, and the weights of other automatic flow monitoring values will be normalized:

[0012]

[0013] Where D is a set of non-missing values;

[0014] In the Kalman filter algorithm, the calculation of the Kalman gain, optimal estimate, and state covariance matrix are all adjusted accordingly. The measurement results of the same river water body by different flow measuring instruments are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R. The Kalman filter formula is transformed into the Kalman filter formula for the missing value case.

[0015]

[0016] Where D is a set of non-missing values.

[0017] Furthermore, the multiple automatic flow monitoring methods described in method step (1) include at least two of the following: radar method, imaging method, acoustic Doppler method, ultrasonic time difference method, water level flow relationship unification method, ADCP automatic flow measurement method, and drop index automatic flow prediction method.

[0018] Furthermore, step (ii) also includes: when the analysis finds that within the set time period, the real-time data obtained by multiple flow automatic monitoring methods do not have missing data values, the measurement results of the same river water body by different flow measuring instruments are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R, and the Kalman filter formula is converted into a Kalman filter formula for multiple observation values.

[0019] Furthermore, in step (three) of the method, when it is determined that the real-time data obtained by any automatic flow monitoring method exceeds the historical maximum error threshold, the current measurement results of the method are compared with the past historical data to determine whether a huge jump or abnormal value occurs, exceeding the historical maximum error threshold.

[0020] Furthermore, the method step (four) specifically includes the following sub-steps:

[0021] (1) Calculate the weight w of each observation i , or the default weight of each observation value is 1 / n, where n is the number of observation values;

[0022] (2) Initialize the parameters, state covariance P0, prediction variance Q0, and initialize the flow value x'0 at the previous moment to be equal to the mean of all observations Flow change trend dx0=0;

[0023] (3) Calculate the traffic forecast value:

[0024] x k =x′ k-1 +dx k-1

[0025] (4) Calculate the observed mean:

[0026]

[0027] (5) Calculate the prediction variance:

[0028]

[0029] (6) Calculate the observation variance:

[0030]

[0031] (7) According to the five simplified Kalman filter formulas, the optimal fusion flow x' is calculated. k and state variance P' k ; where each observation value z i , the predicted value is x;

[0032] (8) Update the flow change trend for the next calculation:

[0033] dx k =x′ k -x′ k-1

[0034] (9) Repeat sub-steps (3) to (8) to obtain the fusion result data of multiple traffic real-time data.

[0035] Further, in method step (iv),

[0036] In the process of Kalman filtering algorithm, the observation error R is a fixed value. Through a variance dynamic adjustment method, the observation variance R and prediction variance Q are dynamically adjusted according to the collected data in real-time calculation, so that they can adapt to the continuous changes in the water environment.

[0037] Two gradient coefficients are added to the adjustment of the prediction variance Q and the observation variance R, and the update formula becomes:

[0038] Q k =(1-α)Q k-1 +αQ′ k

[0039] R k,i =(1-β)R k-1,i +βR′ k,i

[0040] The two gradient coefficients are α and β.

[0041] The following describes the technical advantages of the full-range monitoring method of the present invention by comparing it with the existing technology:

[0042] 1) The method of the present invention overcomes the shortcomings of the existing single river flow monitoring method, and automatically analyzes and finds that when there are missing values in the real-time data collected within a set time period, the weights of the monitoring values of the relevant measurement methods are adjusted in a timely manner, and the real-time flow data with relatively good test accuracy is selected. Then, multiple monitoring data are integrated for calculation, and flow automatic monitoring data with better quality than that of a single method can be obtained, which greatly improves the continuity, accuracy and reliability of the flow automatic monitoring data.

[0043] 2) The method of the present invention has a wider scope of application. In actual field measurements, it comprehensively considers the influencing factors of traditional measurement algorithms and the fact that short-term flow measurement results may contain missing values. Data analysis is more continuous and comprehensive, integrating the advantages of multiple flow measurement methods in different water level ranges, such as high water, medium water, and low water, to achieve automatic monitoring of river flow over the entire range, providing real-time flow data support for flood and drought disaster prevention, water resource management, water environment protection, and water ecological restoration.

[0044] 3) The monitoring method of the present invention also has the function of screening out certain flow measurement methods that are severely distorted. The current measurement results of a certain automatic flow monitoring method are compared with past historical data to determine whether there is a huge jump or abnormal value that exceeds the historical maximum error threshold. If exceeded, all flow results obtained by the monitoring method that exceeds the historical maximum error threshold are automatically screened out and do not participate in the flow fusion calculation of the next step until its measurement effect returns to normal. This avoids the interference of the flow measurement data of a certain flow measurement method that is severely distorted on the subsequent overall flow fusion calculation, and further improves the accuracy of the final fusion calculation data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The figure is a flow chart of the steps of the method of the present invention.

[0046] Figure 2 Schematic diagram of the variance adjustment principle of the present invention.

[0047] Figure 3 Flow process diagrams obtained by various flow measurement methods.

[0048] Figure 4 This is a comparison chart of the results of various flow measurement methods and the cruise ADCP.

[0049] Figure 5 It is a fusion flow process line diagram of the present invention.

[0050] Figure 6 This is a comparison chart of the fused flow obtained by the method of the present invention and the cruise ADCP. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the invention will be described clearly and completely below with reference to the accompanying drawings in this embodiment.

[0052] The full-range monitoring method provided in the first embodiment of the present invention comprehensively considers the situation where real-time data can be continuously collected with a high probability, or the situation where data missing values may occasionally occur from a single flow measurement method. Finally, a Kalman filter flow fusion algorithm based on dynamic variance is used to obtain the fusion result data of multiple real-time flow data.

[0053] First, the Kalman filter method is introduced.

[0054] Traditional Kalman filtering is an algorithm that uses the linear system state equation to optimally estimate the system state through system input and output observation data.

[0055] Assuming that the state prediction value is x, the state transfer matrix is F, the control input matrix is B, and the control vector is u, the state prediction formula at the kth moment is:

[0056]

[0057] Assuming the state covariance matrix is P and the prediction variance is Q, the variance update formula is:

[0058] P k =F k P k-1 F k T +Q k ②

[0059] Assuming the observation value is z, the observation variance is R, and the measurement matrix is H, the Kalman gain K is:

[0060] K=H k P k H k T (H k P k H k T +R k ) -1 =H k K′ ③

[0061] The optimal estimate of the state at the kth moment is:

[0062]

[0063] The state covariance matrix at the kth moment is updated as:

[0064] P k ′=(IK′H k )P k ⑤

[0065] The Kalman filter continuously iteratively updates the error term through the observation value and the above five formulas to minimize the error of the fusion result.

[0066] Next, the application of Kalman filtering in the flow calculation algorithm of the present invention is introduced.

[0067] The measurement results of different flow measuring instruments on the same water body are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R. Since there is more than one observation value, the Kalman filter formula needs to be converted into a Kalman filter formula for multiple observation values:

[0068]

[0069] Where n is the number of observations and i is the i-th observation.

[0070] This research team selected the trend prediction method as the traffic prediction model, so the state transfer matrix F is the unit matrix, and the control term B ku k It is equal to the difference dx between the flow rate values at the previous moment and the moment before that. The control item generally represents the influence of external factors on the result and is a variable. However, in this patent solution, we predict that the flow rate change is a fixed change, so a constant dx is used to replace the entire control item.

[0071] Since both the observed and predicted values are flow values, the measurement matrix H is also a unit matrix. Therefore, the five formulas of the traditional Kalman filter (i.e., formulas numbered ① to ⑤) are simplified and transformed into:

[0072]

[0073] P k =P k-1 +Q k ⑦

[0074]

[0075] Next, the dynamic variance adjustment idea of the method of the present invention is introduced.

[0076] In the traditional Kalman filter algorithm, the observation error R is a constant value. The present invention introduces a method for dynamically adjusting the variance. In real-time calculation, the observation variance R and the prediction variance Q are dynamically adjusted according to the collected data, so that the algorithm can adapt to the continuous changes in the water environment and calculate the optimal fused flow.

[0077] Real-time calculation to obtain each observation value z i , whose mean is The variance is μ, and the predicted value is x, then theoretically, the observed mean The closer it is to the predicted value x and the smaller the observation variance μ is, the more accurate and stable the final fused traffic result will be.

[0078] See also Figure 2 , Schematic diagram of the principle of dynamic adjustment of variance. When the observed variance remains unchanged and the deviation between the observed mean and the predicted value becomes larger, we believe that the predicted value is inaccurate and we are more inclined to believe the observed value. The algorithm increases the value of the predicted variance Q; when the deviation between the observed mean and the predicted value remains unchanged and the observed variance becomes larger, we believe that the stability of the observed value is reduced and we are more inclined to believe the predicted value. The algorithm increases the value of the observed variance R.

[0079] To ensure the continuity of data changes, prevent abnormal data from causing variance mutations, and prevent large changes in Q and R that cause large fluctuations in the calculation results, the present invention adds two gradient coefficients α and β to the adjustment of the predicted variance Q and the observed variance R. The gradient coefficients range from 0 to 1. The update formula becomes:

[0080] Q k =(1-α)Qk-1 +αQ′ k

[0081] R k,i =(1-β)R k-1,i +βR′ k,i

[0082] The two gradient coefficients are α and β.

[0083] like Figure 1 The second embodiment of the present invention provides a full-range monitoring method for river flow based on multi-source data fusion, including:

[0084] Step (1) collects real-time data obtained by multiple automatic flow monitoring methods for different water level fluctuations of the river. The multiple automatic flow monitoring methods include at least two of the following:

[0085] Radar method, imaging method, acoustic Doppler method, ultrasonic time difference method, water level-discharge relationship single value method, ADCP automatic flow measurement method, drop index automatic flow inference method, etc.

[0086] The method of the present invention collects real-time data obtained by multiple automatic flow monitoring methods, which refers to real-time data obtained by two or more existing automatic flow monitoring methods.

[0087] In actual field measurements, flow measurement results may be temporarily missing values due to occasional changes in the water environment, communication interruptions, equipment failures, and other reasons. Conventional Kalman filtering algorithms cannot handle missing data, so it is necessary to optimize and adjust conventional Kalman filtering algorithms to address missing data.

[0088] Step (2) When the analysis finds that there are missing values in the real-time data obtained by any automatic flow monitoring method within the set time period, for example, the set time period is 3 minutes or 10 minutes, the weight of each automatic flow monitoring value is recalculated, and in the Kalman filter algorithm, the calculation of the Kalman gain, the optimal estimate and the state covariance matrix are all optimized and adjusted accordingly.

[0089] Specifically, in the first case, when the analysis finds that the real-time data obtained by any of the automatic flow monitoring methods within the set time period has missing values, for example, the set time period is 3 minutes or 10 minutes, the weight of each automatic flow monitoring value is recalculated, that is, the weight corresponding to the missing value is removed, and the other weights are normalized. The weight corresponding to the i-th non-missing value measurement result becomes:

[0090]

[0091] Where D is a set of non-missing flow measurement methods; j is each flow measurement method in D.

[0092] In the Kalman filter algorithm, the calculation of the Kalman gain, optimal estimate, and state covariance matrix are all adjusted accordingly. The measurement results of the same river water body by different flow measuring instruments are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R. The Kalman filter formula is transformed into the Kalman filter formula for the missing value case.

[0093]

[0094] Where D is the set of non-missing values, n is the number of observations, and i is the i-th observation.

[0095] In the second case, if no missing data is found in the real-time data obtained by any automatic flow monitoring method within the set time period, for example, the set time period is 3 minutes or 10 minutes, the measurement results of the same river water body by different flow measuring instruments are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R, and the Kalman filter formula is converted into a Kalman filter formula with multiple observation values.

[0096] The measurement results of the same water body by different flow measuring instruments are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R. Since there is more than one observation value, the Kalman filter formula is converted into the Kalman filter formula of the multi-observation value:

[0097]

[0098] Where n is the number of observations and i is the i-th observation.

[0099] In step (3), when it is determined that the real-time data obtained by any automatic flow monitoring method exceeds the historical maximum error threshold, the flow result obtained by the monitoring method that exceeds the historical maximum error threshold is automatically filtered out and does not participate in the flow fusion calculation in the next step until its measurement effect returns to normal.

[0100] Specifically, the current measurement results of a certain automatic flow monitoring method are compared with past historical data to determine whether there are large jumps or abnormal values that exceed the historical maximum error threshold. If so, all flow results obtained by the monitoring method that exceeds the historical maximum error threshold are automatically filtered out and do not participate in the flow fusion calculation in the next step until the measurement effect returns to normal. If not, the flow fusion calculation in the next step is participated in. This avoids the interference of severely distorted flow measurement data from a certain flow measurement method on the subsequent overall flow fusion calculation, greatly improving the accuracy of the fusion results of multiple real-time flow data.

[0101] Step (4) is to obtain fusion result data of multiple flow real-time data based on the real-time data obtained by the multiple flow automatic monitoring methods through the Kalman filter flow fusion algorithm based on dynamic variance.

[0102] The specific steps include the following:

[0103] (1) Calculate the weight w of each observation i , is calculated based on the jitter of the observed value change curve, or, if comparative data is available, based on the deviation between the observed value and the comparative value. If no weight is calculated, the default weight is 1 / n, where n is the number of observations.

[0104] For example, suppose that in the jth comparison test, the comparison data value is z j ', the observation value at the corresponding time is z i,j , then the root mean square error is:

[0105]

[0106] The weight is calculated as:

[0107]

[0108] (2) Initialize the parameters, state covariance P0, prediction variance Q0, and initialize the flow value x'0 at the previous moment to be equal to the mean of all observations Flow change trend dx0=0;

[0109] (3) Calculate the traffic forecast value:

[0110] x k =x′ k-1 +dx k-1

[0111] (4) Calculate the observed mean:

[0112]

[0113] (5) Calculate the prediction variance:

[0114]

[0115] (6) Calculate the observation variance:

[0116]

[0117] (7) According to the simplified Kalman filter above five formulas (i.e. formulas 11 to 12), the optimal fusion flow x' is calculated k and state variance P' k ; where each observation value zi , the predicted value is x;

[0118] (8) Update the flow change trend for the next calculation:

[0119] dx k =x′ k -x′ k-1

[0120] (9) Repeat sub-steps (3) to (8) to obtain the fusion result data of multiple real-time flow data. The method of the present invention filters the flow values obtained by different devices and different methods through comprehensive analysis, optimizes and adjusts the missing data, and finally fuses them to obtain more accurate, stable and continuous river flow results.

[0121] In the process of Kalman filtering algorithm, the observation error R is a fixed value. Through a variance dynamic adjustment method, the observation variance R and prediction variance Q are dynamically adjusted according to the collected data in real-time calculation, so that the full-range monitoring algorithm can adapt to the continuous changes in the water environment.

[0122] Two gradient coefficients are added to the adjustment of the prediction variance Q and the observation variance R, and the update formula becomes:

[0123] Q k =(1-α)Q k-1 +αQ′ k

[0124] R k,i =(1-β)R k-1,i +βR′ k,i

[0125] The two gradient coefficients are α and β, and their value ranges can be adjusted between 0 and 1 according to actual conditions.

[0126] The advantages of the method of the present invention are illustrated below by comparing the flow rate test results of a hydrological station.

[0127] The flow automatic monitoring data obtained by the side-scan radar automatic flow measurement method, the horizontal ADCP automatic flow measurement method and two drop index automatic flow prediction methods at a hydrological station were selected and compared with the measured flow by the underway ADCP. The results are as follows: Figure 3-Figure 6 shown.

[0128] People in this field use the flow measurement results of the cruise ADCP as the standard reference value, which is most consistent with the actual flow rate. Figure 3 The red dots in the middle represent the cruise ADCP measurement results.

[0129] Figure 4In the figure, the red dashed line represents y = x, indicating that the horizontal and vertical axes are equal. The horizontal axis represents the flow rate value obtained by each measurement method, and the vertical axis represents the standard reference value (i.e., the cruise ADCP measurement result). Ideally, the two are completely equal, that is, all flow measurement points are on the red dashed line, and the ideal correlation coefficient is 1. The blue points represent the measurement values of each method. The farther they are from the red dashed line, the lower the measurement accuracy and the smaller the correlation coefficient. Figure 4 It can be seen that the measurement results of the four flow measurement methods have many abnormal values and points that deviate from the red dotted line, resulting in low accuracy of the final flow measurement value and a small correlation coefficient.

[0130] The side-scan radar automatic flow measurement method often suffers from missing values, exhibits frequent jumps at low flow rates and low water levels, and exhibits certain measurement errors at high water levels. Because these jumps often produce large errors, erroneous data can be easily filtered out before fusion. The head index method, particularly in low water level and low flow rate ranges, exhibits relatively large measurement errors. The horizontal ADCP automatic flow measurement method is relatively accurate, but can exhibit jitter caused by noise.

[0131] Using the full-scale monitoring method of the present invention, the data obtained by the above four measurement methods are collected respectively. After calculation, the root mean square error of the flow value measured by the cruise ADCP flow and the flow value measured at the corresponding time by each flow measurement method is 1417.37, 558.61, 676.83 and 1297.21 respectively. After taking the inverse and normalizing, the weight of each flow measurement method is 0.15, 0.38, 0.31 and 0.16. After calculation by the full-scale monitoring algorithm of Example 2 of the present invention, the fusion flow result is shown in FIG. Figure 5 .

[0132] The comparison results of the fusion flow obtained by the method of the present invention and the cruise ADCP are shown in Figure 6 The blue dot represents the fused flow value, and its proximity to the red dashed line indicates high measurement accuracy and a large correlation coefficient. The analysis results of various measurement parameters are shown in Table 1. It can be seen that the measurement parameter results of the present invention are superior to those of a single method.

[0133] Random uncertainty is a commonly used indicator in hydrological regulations, indicating the degree of data dispersion. Larger values indicate more dispersed data and poorer results. The completeness rate refers to the percentage of points with values out of the total number of points. Due to accidental reasons, some measurement results may be missing. Since the algorithm of this invention is a fusion algorithm, even if one method is missing a value at a certain moment, as long as other methods have results, the fusion method will still have a result. Therefore, the completeness rate is the highest among all methods, reflecting the continuity of the data results.

[0134] It can be seen from the test data that the full-range monitoring algorithm of river flow based on multi-source data fusion of the present invention is better than the above four single flow measurement methods.

[0135] Table 1 Usage of automatic flow monitoring method at a hydrological station

[0136]

[0137] As can be seen, after integrating the data, a hydrological station achieved full-scale automatic flow monitoring, ensuring the continuity of the automatic monitoring data and meeting various verification requirements. The method of the present invention prevents significant fluctuations in the final calculation results, ensuring the accuracy and stability of the flow measurement results. This method can produce higher-quality automatic flow monitoring data than single methods, ensuring the continuity and reliability of the data.

Claims

1. A full-scale monitoring method for river flow based on multi-source data fusion, characterized in that: Including steps: (1) Collecting real-time data from various automatic flow monitoring methods for different water level fluctuations in rivers; (2) When the analysis finds that any of the real-time data obtained by the automatic flow monitoring method has missing data within the set time period, the weight of each automatic flow monitoring value is recalculated, and in the Kalman filter algorithm, the calculation of the Kalman gain, the optimal estimate value and the state covariance matrix are all optimized and adjusted accordingly; (3) When it is determined that the real-time data obtained by any automatic flow monitoring method exceeds the historical maximum error threshold, the flow results obtained by the monitoring method that exceeds the historical maximum error threshold are automatically filtered out and are not included in the flow fusion calculation in the next step; (4) Based on the real-time data obtained by the above-mentioned multiple flow automatic monitoring methods, a fusion result data of multiple flow real-time data is obtained through a Kalman filter flow fusion algorithm based on dynamic variance.

2. The full-range monitoring method for river flow based on multi-source data fusion according to claim 1 is characterized in that: Step (2) specifically includes: When the analysis finds that there are missing values in the real-time data obtained by any of the automatic flow monitoring methods within the set time period, the weight of each automatic flow monitoring value will be recalculated, that is, the weight corresponding to the missing value will be removed, and the weights of other automatic flow monitoring values will be normalized: Where D is a set of non-missing values; In the Kalman filter algorithm, the calculation of the Kalman gain, optimal estimate, and state covariance matrix are all adjusted accordingly. The measurement results of the same river water body by different flow measuring instruments are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R. The Kalman filter formula is transformed into the Kalman filter formula for the missing value case. Where D is a set of non-missing values.

3. The full-range monitoring method for river flow based on multi-source data fusion according to claim 1 is characterized in that: Step (2) also includes: When the analysis finds that within the set time period, the real-time data obtained by multiple automatic flow monitoring methods do not have missing values, the measurement results of the same river water body by different flow measuring instruments are used as the observation value z in the Kalman filter calculation, and the instrument measurement error is used as the observation variance R, and the Kalman filter formula is converted into a Kalman filter formula with multiple observation values.

4. The full-range monitoring method for river flow based on multi-source data fusion according to claim 1 is characterized in that: The multiple automatic flow monitoring methods described in step (1) include at least two of the following: radar method, imaging method, acoustic Doppler method, ultrasonic time difference method, water level flow relationship single value method, ADCP automatic flow measurement method, and drop index automatic flow prediction method.

5. The full-range monitoring method for river flow based on multi-source data fusion according to claim 1 is characterized in that: In step (3), when it is determined that the real-time data obtained by any automatic flow monitoring method exceeds the historical maximum error threshold, the current measurement results of the method are compared with the past historical data to determine whether a huge jump or abnormal value occurs, exceeding the historical maximum error threshold.

6. The full-range monitoring method for river flow based on multi-source data fusion according to claim 1 is characterized in that: Step (4) specifically includes the following sub-steps: (1) Calculate the weight w of each observation i , or the default weight of each observation value is 1 / n, where n is the number of observation values; (2) Initialize the parameters, state covariance P0, prediction variance Q0, and initialize the flow value x'0 at the previous moment to be equal to the mean of all observations Flow change trend dx0=0; (3) Calculate the traffic forecast value: x k =x′ k-1 +dx k-1 (4) Calculate the observed mean: (5) Calculate the prediction variance: (6) Calculate the observation variance: (7) According to the simplified Kalman filter five formulas, the fusion flow x' is calculated k and state variance P' k ; where each observation value z i , the predicted value is x; (8) Update the flow change trend for the next calculation: dx k =x′ k -x′ k-1 (9) Repeat sub-steps (3) to (8) to obtain the fusion result data of multiple traffic real-time data.

7. The full-range monitoring method for river flow based on multi-source data fusion according to claim 1 or 6, characterized in that: In step (4), In the process of Kalman filtering algorithm, the observation error R is a fixed value. Through a variance dynamic adjustment method, the observation variance R and prediction variance Q are dynamically adjusted according to the collected data in real-time calculation, so that the full-range monitoring algorithm can adapt to the continuous changes in the water environment. Two gradient coefficients are added to the adjustment of the prediction variance Q and the observation variance R, and the update formula becomes: Q k =(1-α)Q k-1 +αQ′ k R k,i =(1-β)R k-1,i +βR′ k,i The two gradient coefficients are α and β.

Images