Abnormality diagnosis method and system for automatic monitoring data of surface water

Abstract

The invention discloses a surface water automatic monitoring data anomaly diagnosis method and system, and relates to the technical field of water environment monitoring, and the surface water automatic monitoring data anomaly diagnosis method comprises the following steps: S1, receiving original data, and preprocessing the original data to obtain monitoring data; s2, inputting the monitoring data into a pre-constructed data prediction model, and generating prediction data by the data prediction model; s3, receiving the collected data, and preprocessing the collected data to obtain actual data; and S4, inputting the prediction data and the actual data into the abnormality diagnosis model, generating a data diagnosis result by the abnormality diagnosis model, and sending the data diagnosis result. According to the invention, the efficiency of data anomaly diagnosis and the accuracy of data anomaly diagnosis analysis can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of water environment monitoring, and in particular to an abnormality diagnosis method and system for automatic surface water monitoring data. Background Art

[0002] Automatic surface water monitoring data is obtained by real-time or regular testing of various water quality indicators of surface water bodies (such as rivers, lakes and reservoirs) through automated monitoring equipment and systems. These data usually include but are not limited to the following aspects:

[0003] 1. Physical indicators: such as water temperature, turbidity and conductivity.

[0004] 2. Chemical indicators: such as dissolved oxygen, pH value, chemical oxygen demand (COD), ammonia nitrogen, total phosphorus and total nitrogen.

[0005] 3. Heavy metal content: such as lead, mercury, cadmium and chromium.

[0006] 4. Multiple indicators of rainfall data.

[0007] As the basic data for evaluating water environment conditions, surface water automatic monitoring data can timely reflect the water quality conditions and changing trends of surface water bodies, thus providing important basis and support for water resource protection, water pollution prevention and control, and water environment management. Therefore, the accuracy of surface water automatic monitoring data is particularly important.

[0008] However, the existing abnormality diagnosis methods for automatic surface water monitoring data have the following problems:

[0009] (1) Before performing abnormal diagnosis on the surface water automatic monitoring data, the abnormal data needs to be marked first, and then the marked data is diagnosed with abnormalities through the existing model, such as the existing supervised abnormal data classification model. However, the amount of data from the automatic station is large, the marking is time-consuming and labor-intensive, and there is a problem of low efficiency in abnormal diagnosis.

[0010] (2) The existing anomaly diagnosis methods for automatic surface water monitoring data cannot perform anomaly diagnosis on both single-point data and continuous sequence data, and the accuracy of data anomaly diagnosis analysis is low. Summary of the invention

[0011] The purpose of the present application is to provide a method and system for diagnosing anomalies in automatic surface water monitoring data, which can improve the efficiency of data anomaly diagnosis and the accuracy of data anomaly diagnosis analysis.

[0012] To achieve the above-mentioned purpose, the present application provides an abnormal diagnosis method for automatic surface water monitoring data, comprising the following steps: S1: receiving raw data, and preprocessing the raw data to obtain monitoring data; wherein the raw data includes: at least one section data, each section data includes: a section coordinate and N groups of indicator data, each group of indicator data includes: a previous time node and multiple indicators; the monitoring data includes: at least one noise reduction section data, each noise reduction section data includes: a section coordinate and N groups of noise reduction indicator data, each group of noise reduction indicator data includes: a previous time node and multiple noise reduction indicators; one section data corresponds to one noise reduction section data, and the section coordinates of the section data are the same as the section coordinates of the corresponding noise reduction section data; S2: inputting the monitoring data into a pre-constructed data prediction model, and generating predicted data by the data prediction model; wherein the predicted data includes: at least one predicted section data, each predicted section data includes: a Section coordinates and M groups of prediction index data, each group of prediction index data includes: a prediction time node and multiple prediction indexes; one denoised section data corresponds to one predicted section data, and the section coordinates of the denoised section data are the same as the section coordinates of the corresponding predicted section data; S3: receiving the collected data, and preprocessing the collected data to obtain actual data; wherein, the actual data includes: at least one actual section data, each actual section data includes: a section coordinate and M groups of actual index data, each group of actual index data includes: a later time node and multiple actual indexes; one predicted section data corresponds to one actual section data, and the section coordinates of the predicted section data are the same as the section coordinates of the corresponding actual section data; S4: inputting both the predicted data and the actual data into the abnormal diagnosis model, generating a data diagnosis result by the abnormal diagnosis model, and sending it; wherein, the data diagnosis result includes: at least one sub-diagnosis result, and the sub-diagnosis result is data abnormality or data normal.

[0013] As above, the sub-steps of receiving the original data and preprocessing the original data to obtain the monitoring data are as follows: S11: Read the original data. If there is only one section data in the original data, set the preprocessing sequence number of the section data to 1, and use the section data as the current section data, and execute S13; if there are multiple section data in the original data, generate a preprocessing sequence number for each section data in a random order, and the preprocessing sequence number of the section data with an earlier random order is smaller than the preprocessing sequence number of the section data with a later random order, and execute S12; S12: Use the section data with the smallest preprocessing sequence number as the current section data, and execute S13; S13: Sort the N groups of indicator data in the current section data in the order from front to back of the previous time node, and generate a noise reduction sequence number for each group of indicator data. No., the denoising sequence number of a group of indicator data at the previous time node is smaller than the denoising sequence number of a group of indicator data at the previous time node, execute S14; S14: construct the original data sequence according to the denoising sequence number from small to large for N groups of indicator data, perform denoising on the original data sequence, obtain denoised section data after the denoising, and execute S15; S15: use the total number of section data in the original data to judge the preprocessing sequence number of the current section data. If the total number of section data in the original data is greater than the preprocessing sequence number of the current section data, then remove the preprocessing sequence number of the current section data and execute S12; if the total number of section data in the original data is equal to the preprocessing sequence number of the current section data, then remove the preprocessing sequence number of the current section data and use all denoised section data as monitoring data.

[0014] As above, wherein the monitoring data is input into a pre-built data prediction model, and the sub-steps of generating prediction data by the data prediction model are as follows: S21: Read the monitoring data. If there is only one noise reduction section data in the monitoring data, set the prediction task number of the noise reduction section data to 1, and use the noise reduction section data as the current data to be predicted, and execute S23; if there are multiple noise reduction section data in the monitoring data, generate a prediction task number for each noise reduction section data in a random order, and the prediction task number of the noise reduction section data with a first random order is smaller than the prediction task number of the noise reduction section data with a later random order, and execute S22; S22: Use the noise reduction section data with the smallest prediction task number as the current data to be predicted, and execute S23; S23: Construct M prediction time nodes for each group of noise reduction index data according to the total number N of previous time nodes of each group of noise reduction index data of the current data to be predicted, and execute S24; wherein, or S24: In the order from front to back of the previous time node, the N groups of noise reduction index data in the current prediction data and the M prediction time nodes of each group of noise reduction index data are input into the pre-built data prediction model in sequence, and the data prediction model generates multiple prediction indicators for each prediction time node of each group of noise reduction index data respectively, and constitutes M groups of prediction index data, and the M groups of prediction index data and the corresponding section coordinates are used as a prediction section data, and S25 is executed, wherein a group of prediction index data includes: a prediction time node and multiple prediction indicators corresponding to the prediction time node; S25: The prediction task number of the current data to be predicted is judged by the total number of noise reduction section data in the monitoring data. If the total number of noise reduction section data in the monitoring data is greater than the prediction task number of the current data to be predicted, the prediction task number of the current data to be predicted is eliminated, and S22 is executed; if the total number of noise reduction section data in the monitoring data is equal to the prediction task number of the current data to be predicted, the prediction task number of the previous data to be predicted is eliminated, and all the prediction section data are used as the prediction data.

[0015] As above, the sub-steps of receiving the collected data and preprocessing the collected data to obtain the actual data are as follows: S31: Read the collected data. If there is only one collected section data in the collected data, set the preprocessing sequence number of the collected section data to 1, and use the collected section data as the current collected section data, and execute S33; if there are multiple collected section data in the collected data, generate a preprocessing sequence number for each collected section data in a random order, and the preprocessing sequence number of the collected section data with an earlier random order is smaller than the preprocessing sequence number of the collected section data with a later random order, and execute S32; S32: Use the collected section data with the smallest preprocessing sequence number as the current collected section data, and execute S33; S33: Sort the M groups of collected index data in the current collected section data in order from front to back at the later time nodes, and generate a noise reduction sequence number for each group of collected index data No., the denoising sequence number of a group of acquisition index data before the subsequent time node is less than the denoising sequence number of a group of acquisition index data after the subsequent time node, execute S34; S34: construct the original acquisition data sequence according to the denoising sequence number from small to large for the M groups of acquisition index data, perform denoising on the original acquisition data sequence, obtain the actual section data after the denoising, and execute S35; S35: use the total number of acquisition section data in the acquisition data to judge the preprocessing sequence number of the current acquisition section data, if the total number of acquisition section data in the acquisition data is greater than the preprocessing sequence number of the current acquisition section data, then remove the preprocessing sequence number of the current acquisition section data, and execute S32; if the total number of acquisition section data in the acquisition data is equal to the preprocessing sequence number of the current acquisition section data, then remove the preprocessing sequence number of the current acquisition section data, and use all the actual section data as the actual data.

[0016] As above, the predicted data and the actual data are both input into the abnormal diagnosis model, and the sub-steps of generating the data diagnosis result by the abnormal diagnosis model are as follows: S41: Read the predicted data and the actual data. If there is one predicted section data in the predicted data and one actual section data in the actual data, then use the predicted section data and the actual section data as the current diagnosis data, and set the sequence number of the current diagnosis data to be diagnosed to 1, and execute S43; if there are multiple predicted section data in the predicted data and multiple actual section data in the actual data, then use the predicted section data and the actual section data with the same section coordinates as a group of data to be diagnosed, and randomly assign each group of data to be diagnosed A diagnosis sequence number is generated according to the random sequence, and the diagnosis sequence number of a group of diagnosis data in the front of the random sequence is less than the diagnosis sequence number of a group of diagnosis data in the back of the random sequence, and S42 is executed; S42: the group of diagnosis data with the smallest diagnosis sequence number is used as the current diagnosis data, and S43 is executed; S43: the M groups of prediction index data and the M groups of actual index data in the current diagnosis data are grouped, and the prediction index data and the actual index data with the same prediction time node and the same subsequent time node are used as a group of analysis data, and an analysis sequence number is generated for each group of analysis data in the order from the front to the back of the subsequent time node, and the analysis sequence number of the group of analysis data in the front of the subsequent time node is less than At the analysis sequence number of a group of analysis data at the later time node, execute S44; S44: perform error trend analysis on the M group of analysis data by the first analysis model in the abnormal diagnosis model to obtain a first diagnosis result, wherein the first diagnosis result is: data abnormality or data normal; if the first diagnosis result is that the data is normal, execute S45; if the first diagnosis result is that the data is abnormal, execute S46; S45: analyze the Mth group of analysis data by the second analysis model in the abnormal diagnosis model to obtain a second diagnosis result, wherein the second diagnosis result is: data abnormality or data normal; S46: generate a sub-diagnosis result according to the first diagnosis result or the second diagnosis result, execute S47; wherein, if the first diagnostic result is data abnormality, then the sub-diagnostic result is data abnormality; if the second diagnostic result is data normal, then the sub-diagnostic result is data normal; if the second diagnostic result is data abnormality, then the sub-diagnostic result is data abnormality; S47: use the total number of current diagnostic data to judge the sequence number of the current diagnostic data to be diagnosed. If the total number of current diagnostic data is greater than the sequence number of the current diagnostic data to be diagnosed, then the sequence number of the current diagnostic data to be diagnosed is eliminated and S42 is executed; if the total number of current diagnostic data is equal to the sequence number of the current diagnostic data to be diagnosed, then the sequence number of the current diagnostic data to be diagnosed is eliminated and all sub-diagnostic results are used as data diagnostic results.

[0017] As above, the first analysis model in the abnormal diagnosis model performs error trend analysis on the M group analysis data, and the sub-steps of obtaining the first diagnosis result are as follows: S441: Obtain the error sequence of each prediction indicator in the M group analysis data by the first analysis model in the abnormal diagnosis model, wherein the error sequence in, The kth prediction indicator in the analysis data with analysis number 1 and the kth actual index The prediction error between The kth prediction indicator in the analysis data with analysis number 2 and the kth actual index The prediction error between For the kth prediction indicator in the analysis data with analysis number M and the kth actual index The prediction error value between ; k∈[1,K], K is the total number of prediction indicators; S442: The first analysis model in the abnormal diagnosis model performs error trend analysis on the error sequence of each prediction indicator respectively to obtain K first sub-results. If the trend of the error sequence is increasing, the first sub-result is abnormal; if the trend of the error sequence is not increasing, the first sub-result is normal; S443: Analyze the K first sub-results and generate a first diagnosis result. If the K first sub-results are all normal, the first diagnosis result is that the data is normal; if there is at least one abnormality among the K first sub-results, the first diagnosis result is that the data is abnormal.

[0018] As above, the total number of types of prediction indicators K = 9; the 9 prediction indicators are: rainfall, rainfall intensity, rainfall duration, rainfall frequency, rainfall distribution uniformity, maximum hourly rainfall, maximum daily rainfall, return period and previous rainfall.

[0019] As above, the second analysis model in the abnormal diagnosis model analyzes the Mth group of analysis data, and the sub-steps for obtaining the second diagnosis result are as follows: S451: Take the cross-sectional coordinates of the Mth group of analysis data as the center of the sphere, and construct a spherical area with a preset area radius as the radius, and use the spherical area as the target sample area; S452: Traverse the sample database according to the target sample area, and use the sample data with overlapping parts of the area range and the target sample area as the target sample data, wherein each target sample data includes: input data, output data and area range; input data includes: multiple input sample indicators; output data includes: multiple sample standard indicators; multiple input sample indicators correspond to multiple sample standard indicators one by one; multiple sample standard indicators correspond to multiple prediction indicators one by one; S453: Input the input data of the target sample data into the data prediction model, and generate prediction sample data by the data prediction model, and each prediction sample data includes: Multiple prediction sample indicators; S454: Use the output data of the target sample data to analyze the prediction sample data to obtain the benchmark error of each prediction indicator; S455: Use the benchmark error of each prediction indicator to judge the current error value of each prediction indicator in the M-th group of analysis data, and generate K second sub-results. If the current error value of the k-th prediction indicator is less than or equal to the benchmark error, the second sub-result of the k-th prediction indicator is normal; if the current error value of the k-th prediction indicator is greater than the benchmark error, the second sub-result of the k-th prediction indicator is abnormal; wherein, the current error value of the k-th prediction indicator = the k-th prediction indicator in the M-th group of analysis data - the k-th actual indicator in the M-th group of analysis data; S456: Analyze the K second sub-results and generate a second diagnosis result. If the K second sub-results are all normal, the second diagnosis result is that the data is normal; if there is at least one abnormality in the K second sub-results, the second diagnosis result is that the data is abnormal.

[0020] As above, the benchmark errors of different types of prediction indicators are different, and the expression of the benchmark error of the k-th prediction indicator is: Among them, Bwc k is the benchmark error of the k-th prediction indicator; is the kth prediction sample index in the output data of the vth prediction sample data; is the kth sample standard index in the output data of the vth target sample data, v∈[1,V], V is the total number of target sample data; Δbzc k is the standard deviation of the k-th prediction indicator.

[0021] The present application also provides an abnormality diagnosis system for automatic surface water monitoring data, including: multiple monitoring nodes and a data abnormality diagnosis center; wherein the monitoring nodes are used to send original data; send collected data; receive data diagnosis results; and the data abnormality diagnosis center is used to execute the above-mentioned abnormality diagnosis method for automatic surface water monitoring data.

[0022] The beneficial effects achieved by this application are as follows:

[0023] (1) The anomaly diagnosis method and system for automatic surface water monitoring data of the present application can directly perform data anomaly diagnosis on the original data without pre-marking the abnormal data, which can effectively improve the anomaly diagnosis efficiency of a large amount of original data.

[0024] (2) The anomaly diagnosis method and system of surface water automatic monitoring data of the present application can effectively improve the accuracy of data anomaly diagnosis and analysis.

[0025] (3) The abnormality diagnosis method and system of the surface water automatic monitoring data of the present application can perform abnormality diagnosis on single-point data, and can also perform abnormality diagnosis on continuous sequence data.

[0026] (4) The abnormality diagnosis method and system of the surface water automatic monitoring data of the present application can fully consider the correlation between the upstream and downstream section data, and at the same time judge whether the data of multiple sections are abnormal.

[0027] (5) The abnormality diagnosis method and system of the surface water automatic monitoring data of the present application can automatically identify the original data with abnormalities. Timely and accurate detection of abnormal data can provide support for the operation and maintenance of the automatic station and the early detection of pollution incidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0029] Figure 1 A schematic diagram of the structure of an embodiment of an abnormality diagnosis system for automatic surface water monitoring data;

[0030] Figure 2 The present invention is a flow chart of an embodiment of a method for diagnosing abnormalities in automatic surface water monitoring data. DETAILED DESCRIPTION

[0031] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0032] like Figure 1 As shown, the present application provides an abnormality diagnosis system for automatic surface water monitoring data, including: multiple monitoring nodes 110 and a data abnormality diagnosis center 120.

[0033] The monitoring node 110 is used to send original data, send collected data, and receive data diagnosis results.

[0034] The data anomaly diagnosis center 120 is used to execute the following anomaly diagnosis method for automatic surface water monitoring data.

[0035] Furthermore, the monitoring node 110 is an automatic station, but is not limited to an automatic station. The present application preferably uses an automatic station.

[0036] Furthermore, the data anomaly diagnosis center 120 includes: a preprocessing unit, a prediction unit, an anomaly diagnosis unit and a storage unit.

[0037] Among them, the preprocessing unit is used to receive raw data and preprocess the raw data to obtain monitoring data; it is used to receive collected data and preprocess the collected data to obtain actual data.

[0038] Prediction unit: used to input monitoring data into a pre-built data prediction model, and the data prediction model generates prediction data.

[0039] Abnormal diagnosis unit: used to input both predicted data and actual data into the abnormal diagnosis model, and the abnormal diagnosis model generates data diagnosis results and sends them.

[0040] Storage unit: used to store a sample database, the sample database includes: a plurality of sample data, each sample data includes: input data, output data and area range; the area range of each sample data is different.

[0041] like Figure 2 As shown, the present application provides a method for diagnosing abnormalities in surface water automatic monitoring data, comprising the following steps:

[0042] S1: Receive raw data and preprocess the raw data to obtain monitoring data; wherein the raw data includes: at least one section data, each section data includes: a section coordinate and N groups of indicator data, each group of indicator data includes: a previous time node and multiple indicators; the monitoring data includes: at least one noise reduction section data, each noise reduction section data includes: a section coordinate and N groups of noise reduction indicator data, each group of noise reduction indicator data includes: a previous time node and multiple noise reduction indicators; one section data corresponds to one noise reduction section data, and the section coordinates of the section data are the same as the section coordinates of the corresponding noise reduction section data.

[0043] Specifically, the previous time node is the time node before the time node at which the monitoring node sends the original data this time. The total number of previous time nodes in each section data is consistent with the number of groups of index data in the section data. The total number of previous time nodes in each noise reduction section data is consistent with the number of groups of noise reduction index data in the noise reduction section data.

[0044] Among them, a variety of indicators are used to represent the rainfall data in the data collection area at previous time nodes.

[0045] The specific number of types of multiple indicators is set according to actual conditions, and this application preferably uses 9 indicators.

[0046] The specific contents of various indicators are set according to the actual situation. The various indicators at least include: rainfall, rainfall intensity, rainfall duration, rainfall frequency, rainfall distribution uniformity, maximum hourly rainfall, maximum daily rainfall, recurrence period and previous rainfall.

[0047] Among them, rainfall refers to the total amount of rain that falls within a certain period of time.

[0048] Rainfall intensity is the amount of rainfall per unit time.

[0049] Rainfall duration is the time that rainfall lasts.

[0050] Rainfall frequency is the number of times a specific amount of rainfall occurs within a certain period of time.

[0051] Rainfall distribution uniformity is used to describe the degree of spatial balance of rainfall distribution.

[0052] The maximum hourly rainfall and maximum daily rainfall refer to the maximum rainfall in one hour and one day respectively.

[0053] The return period refers to the average interval between the recurrence of rainfall of a certain magnitude over a long period of time.

[0054] The previous rainfall is the cumulative rainfall in the period before this rainfall.

[0055] The multiple indicators may also include: physical indicators, chemical indicators and heavy metal content.

[0056] The section coordinates are the coordinates of the location where the section data is collected. One section data corresponds to one section coordinate.

[0057] Furthermore, the sub-steps of receiving the raw data and preprocessing the raw data to obtain the monitoring data are as follows:

[0058] S11: Read the original data. If there is only one cross-sectional data in the original data, set the preprocessing number of the cross-sectional data to 1, and use the cross-sectional data as the current cross-sectional data, and execute S13; if there are multiple cross-sectional data in the original data, generate a preprocessing number for each cross-sectional data in a random order, and the preprocessing number of the cross-sectional data with an earlier random order is smaller than the preprocessing number of the cross-sectional data with a later random order, and execute S12.

[0059] Specifically, if there is only one cross-sectional data in the original data, it means that the original data is single-point data.

[0060] If there are multiple cross-section data in the original data, it means that the original data is multi-point data. Multi-point data includes: multiple single-point data with continuous time series (ie: continuous sequence data) or multiple single-point data with discontinuous time series.

[0061] S12: Take the section data with the smallest preprocessing sequence number as the current section data and execute S13.

[0062] S13: Sort the N groups of indicator data in the current section data in order from the front to the back of the previous time node, and generate a denoising sequence number for each group of indicator data. The denoising sequence number of a group of indicator data at the front of the previous time node is smaller than the denoising sequence number of a group of indicator data at the back of the previous time node, and execute S14.

[0063] Furthermore, the time interval between two adjacent prior time nodes may be equal or unequal. In the present application, the time interval between two adjacent prior time nodes is preferably equal.

[0064] The specific value of the time interval between two adjacent prior time nodes is set according to actual conditions, and can be a second, a minute, a hour, a day, a week, a month, a quarter or a year. In this application, the time interval between two adjacent prior time nodes is preferably 1 hour.

[0065] S14: N groups of index data are arranged in ascending order of denoising numbers to form an original data sequence, and denoising is performed on the original data sequence. After the denoising is completed, denoised cross-sectional data is obtained, and S15 is executed.

[0066] Specifically, the original data collected by the monitoring node has noise, so it is necessary to perform noise reduction processing on the original data to improve the accuracy of the monitoring data.

[0067] The denoising of the original data sequence can be achieved by using existing denoising methods, such as mean filtering denoising method, median filtering denoising method and Gaussian filtering denoising method.

[0068] S15: Use the total number of section data in the original data to judge the preprocessing number of the current section data. If the total number of section data in the original data is greater than the preprocessing number of the current section data, the preprocessing number of the current section data is eliminated and S12 is executed; if the total number of section data in the original data is equal to the preprocessing number of the current section data, the preprocessing number of the current section data is eliminated and all the noise-reduced section data are used as monitoring data.

[0069] Specifically, one denoised section data corresponds to one section data, the N previous time nodes of the section data correspond one-to-one with the N previous time nodes of the corresponding denoised section data, and the section coordinates of the section data are the same as the section coordinates of the corresponding denoised section data.

[0070] Furthermore, the specific number of groups of indicator data is set according to actual conditions, and the specific number of groups of indicator data may be: N∈[2, 16], N∈[3, 24] or N∈[12, 24].

[0071] S2: Input the monitoring data into a pre-built data prediction model, and generate prediction data by the data prediction model; wherein the prediction data includes: at least one prediction section data, each prediction section data includes: a section coordinate and M groups of prediction index data, each group of prediction index data includes: a prediction time node and multiple prediction indicators; one noise reduction section data corresponds to one prediction section data, and the section coordinates of the noise reduction section data are the same as the section coordinates of the corresponding prediction section data.

[0072] Specifically, the specific value of the time interval between two adjacent prediction time nodes is set according to actual conditions, and can be a second, a minute, a hour, a day, a week, a month, a quarter or a year. The present application preferably sets the time interval between two adjacent prediction time nodes to 1 hour.

[0073] Furthermore, the monitoring data is input into a pre-built data prediction model, and the sub-steps of generating prediction data from the data prediction model are as follows:

[0074] S21: Read the monitoring data. If there is only one denoising section data in the monitoring data, set the prediction task number of the denoising section data to 1, and use the denoising section data as the current data to be predicted, and execute S23; if there are multiple denoising section data in the monitoring data, generate a prediction task number for each denoising section data in a random order, and the prediction task number of the denoising section data with an earlier random order is smaller than the prediction task number of the denoising section data with a later random order, and execute S22.

[0075] S22: Take the noise reduction section data with the smallest prediction task number as the current data to be predicted, and execute S23.

[0076] S23: construct M prediction time nodes for each group of noise reduction index data according to the total number N of previous time nodes of each group of noise reduction index data to be predicted, and execute S24; wherein, or

[0077] Specifically, or But not limited to or

[0078] Among them, the specific number of prediction time nodes is set according to the actual situation, and the preferred number in this application is: M∈[1,8].

[0079] A set of noise reduction index data corresponds to M sets of prediction index data, and a set of prediction index data corresponds to a prediction time node and multiple prediction indicators.

[0080] S24: In order from front to back of the previous time node, the N groups of noise reduction index data in the current prediction data and the M prediction time nodes of each group of noise reduction index data are input into the pre-built data prediction model in sequence, and the data prediction model generates multiple prediction indicators for each prediction time node of each group of noise reduction index data, and constitutes M groups of prediction index data, and the M groups of prediction index data and the corresponding section coordinates are taken as a prediction section data, and S25 is executed, wherein a group of prediction index data includes: a prediction time node and multiple prediction indicators corresponding to the prediction time node.

[0081] Furthermore, the data prediction model is a pre-trained neural network model.

[0082] S25: Use the total number of denoised section data in the monitoring data to judge the prediction task number of the current data to be predicted. If the total number of denoised section data in the monitoring data is greater than the prediction task number of the current data to be predicted, the prediction task number of the current data to be predicted is eliminated and S22 is executed; if the total number of denoised section data in the monitoring data is equal to the prediction task number of the current data to be predicted, the prediction task number of the previous data to be predicted is eliminated, and all the prediction section data are used as prediction data.

[0083] Specifically, one denoised section data corresponds to one predicted section data, and the section coordinates of the denoised section data are the same as the section coordinates of the corresponding predicted section data.

[0084] S3: Receive collected data, and pre-process the collected data to obtain actual data; wherein the actual data includes: at least one actual section data, each actual section data includes: a section coordinate and M groups of actual indicator data, each group of actual indicator data includes: a subsequent time node and multiple actual indicators; one predicted section data corresponds to one actual section data, and the section coordinates of the predicted section data are the same as the section coordinates of the corresponding actual section data.

[0085] Furthermore, the sub-steps of receiving the collected data and preprocessing the collected data to obtain the actual data are as follows:

[0086] S31: Read the collected data. If there is only one collected section data in the collected data, set the preprocessing sequence number of the collected section data to 1, and use the collected section data as the current collected section data, and execute S33; if there are multiple collected section data in the collected data, generate a preprocessing sequence number for each collected section data in a random order, and the preprocessing sequence number of the collected section data with an earlier random order is smaller than the preprocessing sequence number of the collected section data with a later random order, and execute S32.

[0087] Specifically, the collected data includes: at least one collected section data, each collected section data includes: a section coordinate and M groups of collected index data, each group of collected index data includes: a subsequent time node and multiple indicators.

[0088] One acquisition section data corresponds to one prediction section data, the subsequent time node of the acquisition section data is consistent with the prediction time node of the prediction section data, and the section coordinates of the acquisition section data are consistent with the section coordinates of the prediction section data.

[0089] S32: Use the acquisition section data with the smallest preprocessing sequence number as the current acquisition section data, and execute S33.

[0090] S33: Sort the M groups of acquisition index data in the current acquisition section data in order from front to back at the subsequent time node, and generate a denoising sequence number for each group of acquisition index data. The denoising sequence number of a group of acquisition index data in front of the subsequent time node is smaller than the denoising sequence number of a group of acquisition index data in the back of the subsequent time node, and execute S34.

[0091] Furthermore, the time interval between two adjacent subsequent time nodes may be equal or unequal. In the present application, it is preferred that the time interval between two adjacent subsequent time nodes is equal.

[0092] Specifically, the specific value of the time interval between two adjacent subsequent time nodes is set according to actual conditions, and can be a second, a minute, a hour, a day, a week, a month, a quarter or a year. The present application preferably sets the time interval between two adjacent subsequent time nodes to 1 hour.

[0093] S34: The M groups of acquisition index data are arranged in ascending order of the noise reduction sequence number to form an original acquisition data sequence, and the original acquisition data sequence is subjected to noise reduction processing. After the noise reduction processing is completed, the actual cross-sectional data is obtained, and S35 is executed.

[0094] Specifically, the noise reduction processing of the original collected data sequence can be achieved by adopting existing noise reduction processing methods, such as: mean filter noise reduction method, median filter noise reduction method and Gaussian filter noise reduction method.

[0095] S35: Use the total number of collected section data in the collected data to judge the preprocessing number of the current collected section data. If the total number of collected section data in the collected data is greater than the preprocessing number of the current collected section data, then the preprocessing number of the current collected section data is eliminated and S32 is executed; if the total number of collected section data in the collected data is equal to the preprocessing number of the current collected section data, then the preprocessing number of the current collected section data is eliminated and all actual section data are used as actual data.

[0096] Specifically, one predicted section data corresponds to one actual section data, the M predicted time nodes of the predicted section data correspond one-to-one to the M subsequent time nodes of the corresponding actual section data, and the section coordinates of the predicted section data are the same as the section coordinates of the corresponding actual section data.

[0097] S4: Inputting both the predicted data and the actual data into the abnormal diagnosis model, generating a data diagnosis result by the abnormal diagnosis model, and sending it; wherein the data diagnosis result includes: at least one sub-diagnosis result, and the sub-diagnosis result is data abnormality or data normal.

[0098] Furthermore, the predicted data and the actual data are input into the abnormal diagnosis model, and the sub-steps of generating the data diagnosis result by the abnormal diagnosis model are as follows:

[0099] S41: Read the predicted data and the actual data. If there is one predicted section data in the predicted data and one actual section data in the actual data, then use the predicted section data and the actual section data as the current diagnosis data, and set the sequence number of the current diagnosis data to be diagnosed to 1, and execute S43; if there are multiple predicted section data in the predicted data and multiple actual section data in the actual data, then use the predicted section data and the actual section data with the same section coordinates as a group of data to be diagnosed, and randomly generate a sequence number to be diagnosed for each group of data to be diagnosed, and the sequence number to be diagnosed of the group of data to be diagnosed in the front of the random order is less than the sequence number to be diagnosed of the group of data to be diagnosed in the back of the random order, and execute S42.

[0100] S42: Take the group of to-be-diagnosed data with the smallest to-be-diagnosed sequence number as the current diagnosis data, and execute S43.

[0101] S43: Group the M groups of predicted indicator data and the M groups of actual indicator data in the current diagnostic data, take the predicted indicator data and the actual indicator data with the same predicted time node and the same subsequent time node as a group of analysis data, and generate an analysis sequence number for each group of analysis data in the order from front to back at the subsequent time node, the analysis sequence number of the group of analysis data before the subsequent time node is smaller than the analysis sequence number of the group of analysis data after the subsequent time node, and execute S44.

[0102] S44: Perform error trend analysis on the M groups of analysis data using the first analysis model in the abnormal diagnosis model to obtain a first diagnosis result, wherein the first diagnosis result is: data abnormality or data normal; if the first diagnosis result is that the data is normal, execute S45; if the first diagnosis result is that the data is abnormal, execute S46.

[0103] Furthermore, the first analysis model in the abnormal diagnosis model performs error trend analysis on the M groups of analysis data, and the sub-steps of obtaining the first diagnosis result are as follows:

[0104] S441: Obtain the error sequence of each prediction indicator in the M groups of analysis data by the first analysis model in the abnormal diagnosis model, wherein the error sequence in, The kth prediction indicator in the analysis data with analysis number 1 and the kth actual index The prediction error between The kth prediction indicator in the analysis data with analysis number 2 and the kth actual index The prediction error between For the kth prediction indicator in the analysis data with analysis number M and the kth actual index The prediction error value between them; k∈[1,K], K is the total number of prediction indicators.

[0105] Furthermore, the total number of types K of prediction indicators is set according to actual conditions, and 9 types are preferred in this application.

[0106] Furthermore, the nine prediction indicators are: rainfall amount, rainfall intensity, rainfall duration, rainfall frequency, rainfall distribution uniformity, maximum hourly rainfall, maximum daily rainfall, return period and antecedent rainfall.

[0107] S442: The first analysis model in the abnormal diagnosis model performs error trend analysis on the error sequence of each prediction indicator to obtain K first sub-results. If the trend of the error sequence is increasing, the first sub-result is abnormal; if the trend of the error sequence is not increasing, the first sub-result is normal.

[0108] S443: Analyze the K first sub-results and generate a first diagnosis result. If all the K first sub-results are normal, the first diagnosis result is that the data is normal; if at least one of the K first sub-results is abnormal, the first diagnosis result is that the data is abnormal.

[0109] S45: Analyze the Mth group of analysis data using the second analysis model in the abnormal diagnosis model to obtain a second diagnosis result, wherein the second diagnosis result is: the data is abnormal or the data is normal.

[0110] Furthermore, the second analysis model in the abnormal diagnosis model analyzes the Mth group of analysis data to obtain the second diagnosis result in the following sub-steps:

[0111] S451: construct a spherical region with the cross-sectional coordinates of the Mth group of analysis data as the sphere center and a preset region radius as the radius, and use the spherical region as the target sample region.

[0112] S452: Traverse the sample database according to the target sample area, and take the sample data with overlapping parts in the area range and the target sample area as the target sample data, wherein each target sample data includes: input data, output data and area range; the input data includes: multiple input sample indicators; the output data includes: multiple sample standard indicators; the multiple input sample indicators correspond to the multiple sample standard indicators one by one; the multiple sample standard indicators correspond to the multiple prediction indicators one by one.

[0113] Specifically, the sample data having overlapping parts between the area range and the target sample area are all taken as the target sample data, which can fully consider the correlation between the upstream and downstream section data.

[0114] S453: Inputting the input data of the target sample data into the data prediction model, and generating predicted sample data by the data prediction model, each predicted sample data includes: a plurality of predicted sample indicators.

[0115] S454: Analyze the predicted sample data using the output data of the target sample data to obtain a benchmark error for each prediction indicator.

[0116] Further, as an embodiment, different types of prediction indicators have different reference errors, wherein the expression of the reference error of the k-th prediction indicator is:

[0117]

[0118] Among them, Bwc k is the benchmark error of the k-th prediction indicator; is the kth prediction sample index in the output data of the vth prediction sample data; is the kth sample standard index in the output data of the vth target sample data, v∈[1,V], V is the total number of target sample data; Δbzc k is the standard deviation of the k-th prediction indicator.

[0119] Specifically, based on different benchmark errors, the accuracy of abnormal diagnosis of surface water automatic monitoring data can be improved. The standard deviation of each prediction indicator is set according to the characteristics of the indicator and the actual situation.

[0120] Furthermore, as another embodiment, the baseline errors of different types of prediction indicators are the same, and a pre-built autoencoder model based on a deep spatiotemporal graph convolutional neural network statistically predicts the error distribution in the Mth hour in the future, and the error greater than the mean + 3 times the standard deviation is used as the baseline error for the Mth hour, and S455 is executed.

[0121] Specifically, an autoencoder model based on a deep spatiotemporal graph convolutional neural network is established, wherein the autoencoder model includes: a deep spatiotemporal graph convolutional neural network model and a simple convolutional neural network; the deep spatiotemporal graph convolutional neural network model is used as the encoder model, and the simple convolutional neural network is used as the decoder model. After the autoencoder model training is completed, the error distribution of the future Mth hour is statistically predicted, and the error greater than the mean + 3 times the standard deviation is used as the benchmark error of the Mth hour, and S455 is executed.

[0122] S455: Use the benchmark error of each prediction indicator to judge the current error value of each prediction indicator in the Mth group of analysis data, and generate K second sub-results. If the current error value of the kth prediction indicator is less than or equal to the benchmark error, the second sub-result of the kth prediction indicator is normal; if the current error value of the kth prediction indicator is greater than the benchmark error, the second sub-result of the kth prediction indicator is abnormal.

[0123] Furthermore, the current error value of the k-th prediction indicator = the k-th prediction indicator in the M-th group of analysis data - the k-th actual indicator in the M-th group of analysis data.

[0124] S456: Analyze the K second sub-results and generate a second diagnosis result. If all the K second sub-results are normal, the second diagnosis result is that the data is normal; if at least one of the K second sub-results is abnormal, the second diagnosis result is that the data is abnormal.

[0125] S46: Generate a sub-diagnosis result based on the first diagnosis result or the second diagnosis result, and execute S47; wherein, if the first diagnosis result is data abnormality, the sub-diagnosis result is data abnormality; if the second diagnosis result is data normal, the sub-diagnosis result is data normal; if the second diagnosis result is data abnormality, the sub-diagnosis result is data abnormality.

[0126] S47: Use the total number of current diagnostic data to judge the sequence number of the current diagnostic data to be diagnosed. If the total number of current diagnostic data is greater than the sequence number of the current diagnostic data to be diagnosed, then the sequence number of the current diagnostic data to be diagnosed is eliminated and S42 is executed; if the total number of current diagnostic data is equal to the sequence number of the current diagnostic data to be diagnosed, then the sequence number of the current diagnostic data to be diagnosed is eliminated and all sub-diagnosis results are used as data diagnosis results.

[0127] Specifically, after obtaining the data diagnosis result, the data anomaly diagnosis center sends the data diagnosis result to the monitoring node.

[0128] The error between the predicted data generated by the data prediction model and the original data of normal data is within the normal range, while the error between the predicted data generated by the data prediction model and the original data of abnormal data deviates from the normal range. Therefore, the monitoring data with the sub-diagnosis result of data abnormality is abnormal data.

[0129] The beneficial effects achieved by this application are as follows:

[0130] (1) The anomaly diagnosis method and system for automatic surface water monitoring data of the present application can directly perform data anomaly diagnosis on the original data without pre-marking the abnormal data, which can effectively improve the anomaly diagnosis efficiency of a large amount of original data.

[0131] (2) The anomaly diagnosis method and system of surface water automatic monitoring data of the present application can effectively improve the accuracy of data anomaly diagnosis and analysis.

[0132] (3) The abnormality diagnosis method and system of the surface water automatic monitoring data of the present application can perform abnormality diagnosis on single-point data, and can also perform abnormality diagnosis on continuous sequence data.

[0133] (4) The abnormality diagnosis method and system of the surface water automatic monitoring data of the present application can fully consider the correlation between the upstream and downstream section data, and at the same time judge whether the data of multiple sections are abnormal.

[0134] (5) The abnormality diagnosis method and system of the surface water automatic monitoring data of the present application can automatically identify the original data with abnormalities. Timely and accurate detection of abnormal data can provide support for the operation and maintenance of the automatic station and the early detection of pollution incidents.

[0135] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic creative concepts. Therefore, the scope of protection of the present application is intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application. Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the protection scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A method for abnormal diagnosis of surface water automatic monitoring data, characterized in that: The steps include: S1: receiving raw data and preprocessing the raw data to obtain monitoring data; wherein the raw data includes: at least one section data, each section data includes: a section coordinate and N groups of index data, each group of index data includes: a previous time node and multiple indexes; the monitoring data includes: at least one noise reduction section data, each noise reduction section data includes: a section coordinate and N groups of noise reduction index data, each group of noise reduction index data includes: a previous time node and multiple noise reduction indexes; one section data corresponds to one noise reduction section data, and the section coordinates of the section data are the same as the section coordinates of the corresponding noise reduction section data; S2: input the monitoring data into a pre-built data prediction model, and generate prediction data by the data prediction model; wherein the prediction data includes: at least one prediction section data, each prediction section data includes: a section coordinate and M groups of prediction index data, each group of prediction index data includes: a prediction time node and multiple prediction indexes; one noise reduction section data corresponds to one prediction section data, and the section coordinates of the noise reduction section data are the same as the section coordinates of the corresponding prediction section data; S3: receiving collected data, and preprocessing the collected data to obtain actual data; wherein the actual data includes: at least one actual section data, each actual section data includes: a section coordinate and M groups of actual index data, each group of actual index data includes: a subsequent time node and multiple actual indexes; one predicted section data corresponds to one actual section data, and the section coordinates of the predicted section data are the same as the section coordinates of the corresponding actual section data; S4: Inputting both the predicted data and the actual data into the abnormal diagnosis model, generating a data diagnosis result by the abnormal diagnosis model, and sending it; wherein the data diagnosis result includes: at least one sub-diagnosis result, and the sub-diagnosis result is data abnormality or data normal.

2. The abnormality diagnosis method of surface water automatic monitoring data according to claim 1 is characterized in that: The sub-steps of receiving and preprocessing the raw data to obtain monitoring data are as follows: S11: Read the original data. If there is only one cross-sectional data in the original data, set the preprocessing sequence number of the cross-sectional data to 1, and use the cross-sectional data as the current cross-sectional data, and execute S13; if there are multiple cross-sectional data in the original data, generate a preprocessing sequence number for each cross-sectional data in a random order, and the preprocessing sequence number of the cross-sectional data with an earlier random order is smaller than the preprocessing sequence number of the cross-sectional data with a later random order, and execute S12; S12: taking the section data with the smallest preprocessing sequence number as the current section data, and executing S13; S13: sorting the N groups of index data in the current section data in order from the previous time node to the next, and generating a denoising sequence number for each group of index data, the denoising sequence number of the group of index data at the previous time node is smaller than the denoising sequence number of the group of index data at the next time node, and executing S14; S14: N groups of index data are arranged in ascending order of denoising numbers to form an original data sequence, and denoising is performed on the original data sequence. After the denoising is completed, denoised cross-sectional data is obtained, and S15 is executed; S15: using the total number of cross-sectional data in the original data to judge the pre-processing number of the current cross-sectional data, if the total number of cross-sectional data in the original data is greater than the pre-processing number of the current cross-sectional data, then the pre-processing number of the current cross-sectional data is removed, and S12 is executed; If the total number of section data in the original data is equal to the preprocessing sequence number of the current section data, the preprocessing sequence number of the current section data is removed, and all the noise-reduced section data are used as monitoring data.

3. The abnormality diagnosis method of surface water automatic monitoring data according to claim 2 is characterized in that: The monitoring data is input into the pre-built data prediction model. The sub-steps of generating prediction data from the data prediction model are as follows: S21: Read the monitoring data. If there is only one noise reduction section data in the monitoring data, set the prediction task sequence number of the noise reduction section data to 1, and use the noise reduction section data as the current data to be predicted, and execute S23; if there are multiple noise reduction section data in the monitoring data, generate a prediction task sequence number for each noise reduction section data in a random order, and the prediction task sequence number of the noise reduction section data with a first random order is smaller than the prediction task sequence number of the noise reduction section data with a later random order, and execute S22; S22: taking the noise reduction section data with the smallest prediction task number as the current data to be predicted, and executing S23; S23: construct M prediction time nodes for each group of noise reduction index data according to the total number N of previous time nodes of each group of noise reduction index data to be predicted, and execute S24; wherein, or S24: In the order of the previous time nodes from front to back, the N groups of noise reduction index data in the current prediction data and the M prediction time nodes of each group of noise reduction index data are input into the pre-built data prediction model in sequence, and the data prediction model generates multiple prediction indicators for each prediction time node of each group of noise reduction index data, and forms M groups of prediction index data, and the M groups of prediction index data and the corresponding section coordinates are used as a prediction section data, and S25 is executed, wherein a group of prediction index data includes: a prediction time node and multiple prediction indicators corresponding to the prediction time node; S25: Use the total number of denoised section data in the monitoring data to judge the prediction task number of the current data to be predicted. If the total number of denoised section data in the monitoring data is greater than the prediction task number of the current data to be predicted, the prediction task number of the previous data to be predicted is eliminated and S22 is executed; if the total number of denoised section data in the monitoring data is equal to the prediction task number of the current data to be predicted, the prediction task number of the current data to be predicted is eliminated and all the prediction section data are used as prediction data.

4. The abnormality diagnosis method of surface water automatic monitoring data according to claim 3 is characterized in that: The sub-steps of receiving collected data and preprocessing the collected data to obtain actual data are as follows: S31: Read the collected data. If there is only one collected section data in the collected data, set the preprocessing sequence number of the collected section data to 1, and use the collected section data as the current collected section data, and execute S33; If there are multiple collected section data in the collected data, a preprocessing sequence number is generated for each collected section data in a random order, and the preprocessing sequence number of the collected section data with an earlier random order is smaller than the preprocessing sequence number of the collected section data with a later random order, and S32 is executed; S32: taking the acquisition section data with the smallest preprocessing sequence number as the current acquisition section data, and executing S33; S33: sorting the M groups of acquisition index data in the current acquisition section data in order from the front to the back at the back time node, and generating a noise reduction sequence number for each group of acquisition index data, the noise reduction sequence number of the group of acquisition index data in the front at the back time node is smaller than the noise reduction sequence number of the group of acquisition index data in the back at the back time node, and executing S34; S34: The M groups of collected index data are arranged in the order of the noise reduction sequence numbers from small to large to form an original collected data sequence, and the original collected data sequence is subjected to noise reduction processing. After the noise reduction processing is completed, actual cross-section data is obtained, and S35 is executed; S35: using the total number of collected section data in the collected data to judge the preprocessing number of the current collected section data, if the total number of collected section data in the collected data is greater than the preprocessing number of the current collected section data, then the preprocessing number of the current collected section data is removed, and S32 is executed; If the total number of collected section data in the collected data is equal to the preprocessing sequence number of the current collected section data, the preprocessing sequence number of the current collected section data is removed, and all actual section data are used as actual data.

5. The abnormality diagnosis method of surface water automatic monitoring data according to claim 4 is characterized in that: The predicted data and actual data are input into the anomaly diagnosis model. The sub-steps of generating data diagnosis results by the anomaly diagnosis model are as follows: S41: Read the predicted data and the actual data. If there is one predicted section data in the predicted data and one actual section data in the actual data, use the predicted section data and the actual section data as the current diagnosis data, and set the sequence number of the current diagnosis data to be diagnosed to 1, and execute S43; if there are multiple predicted section data in the predicted data and multiple actual section data in the actual data, use the predicted section data and the actual section data with the same section coordinates as a group of data to be diagnosed, and randomly generate a sequence number to be diagnosed for each group of data to be diagnosed, and the sequence number to be diagnosed of the group of data to be diagnosed in the front of the random order is less than the sequence number to be diagnosed of the group of data to be diagnosed in the back of the random order, and execute S42; S42: taking the group of to-be-diagnosed data with the smallest to-be-diagnosed sequence number as the current diagnosis data, and executing S43; S43: Grouping the M groups of predicted index data and the M groups of actual index data in the current diagnosis data, taking the predicted index data and the actual index data with the same predicted time node and the same subsequent time node as a group of analysis data, and generating an analysis sequence number for each group of analysis data in the order from the front to the back of the subsequent time node, the analysis sequence number of the group of analysis data with the front subsequent time node is smaller than the analysis sequence number of the group of analysis data with the back subsequent time node, and executing S44; S44: performing error trend analysis on the M groups of analysis data using the first analysis model in the abnormal diagnosis model to obtain a first diagnosis result, wherein the first diagnosis result is: data abnormality or data normal; if the first diagnosis result is that the data is normal, executing S45; if the first diagnosis result is that the data is abnormal, executing S46; S45: Analyze the Mth group of analysis data by the second analysis model in the abnormal diagnosis model to obtain a second diagnosis result, wherein the second diagnosis result is: the data is abnormal or the data is normal; S46: Generate a sub-diagnosis result according to the first diagnosis result or the second diagnosis result, and execute S47; wherein, if the first diagnosis result is data abnormality, the sub-diagnosis result is data abnormality; if the second diagnosis result is data normal, the sub-diagnosis result is data normal; if the second diagnosis result is data abnormality, the sub-diagnosis result is data abnormality; S47: Use the total number of current diagnostic data to judge the sequence number of the current diagnostic data to be diagnosed. If the total number of current diagnostic data is greater than the sequence number of the current diagnostic data to be diagnosed, then the sequence number of the current diagnostic data to be diagnosed is eliminated and S42 is executed; if the total number of current diagnostic data is equal to the sequence number of the current diagnostic data to be diagnosed, then the sequence number of the current diagnostic data to be diagnosed is eliminated and all sub-diagnosis results are used as data diagnosis results.

6. The abnormality diagnosis method of surface water automatic monitoring data according to claim 5 is characterized in that: The first analysis model in the abnormal diagnosis model performs error trend analysis on the M groups of analysis data to obtain the first diagnosis result in the following sub-steps: S441: Obtain the error sequence of each prediction indicator in the M groups of analysis data by the first analysis model in the abnormal diagnosis model, wherein the error sequence in, The kth prediction indicator in the analysis data with analysis number 1 and the kth actual index The prediction error between The kth prediction indicator in the analysis data with analysis number 2 and the kth actual index The prediction error between For the kth prediction indicator in the analysis data with analysis number M and the kth actual index The prediction error value between them; k∈[1,K], K is the total number of types of prediction indicators; S442: using the first analysis model in the abnormal diagnosis model to perform error trend analysis on the error sequence of each prediction indicator, respectively, to obtain K first sub-results. If the trend of the error sequence is increasing, the first sub-result is abnormal; if the trend of the error sequence is not increasing, the first sub-result is normal. S443: Analyze the K first sub-results and generate a first diagnosis result. If all the K first sub-results are normal, the first diagnosis result is that the data is normal; if at least one of the K first sub-results is abnormal, the first diagnosis result is that the data is abnormal.

7. The abnormality diagnosis method of surface water automatic monitoring data according to claim 6 is characterized in that: The total number of types of prediction indicators K = 9; the 9 prediction indicators are: rainfall, rainfall intensity, rainfall duration, rainfall frequency, rainfall distribution uniformity, maximum hourly rainfall, maximum daily rainfall, return period and previous rainfall.

8. The abnormality diagnosis method of surface water automatic monitoring data according to claim 6 is characterized in that: The sub-steps of analyzing the M-th group of analysis data by the second analysis model in the abnormal diagnosis model to obtain the second diagnosis result are as follows: S451: construct a spherical region with the cross-section coordinates of the Mth group of analysis data as the sphere center and a preset region radius as the radius, and use the spherical region as the target sample region; S452: Traversing the sample database according to the target sample area, taking the sample data whose area range overlaps with the target sample area as the target sample data, wherein each target sample data includes: input data, output data and area range; the input data includes: multiple input sample indicators; the output data includes: multiple sample standard indicators; the multiple input sample indicators correspond to the multiple sample standard indicators one by one; the multiple sample standard indicators correspond to the multiple prediction indicators one by one; S453: Inputting the input data of the target sample data into the data prediction model, and generating prediction sample data by the data prediction model, each prediction sample data includes: a plurality of prediction sample indicators; S454: Analyze the predicted sample data using the output data of the target sample data to obtain a benchmark error of each prediction indicator; S455: using the benchmark error of each prediction indicator to judge the current error value of each prediction indicator in the Mth group of analysis data, respectively, to generate K second sub-results, if the current error value of the kth prediction indicator is less than or equal to the benchmark error, then the second sub-result of the kth prediction indicator is normal; if the current error value of the kth prediction indicator is greater than the benchmark error, then the second sub-result of the kth prediction indicator is abnormal; wherein, the current error value of the kth prediction indicator = the kth prediction indicator in the Mth group of analysis data - the kth actual indicator in the Mth group of analysis data; S456: Analyze the K second sub-results and generate a second diagnosis result. If all the K second sub-results are normal, the second diagnosis result is that the data is normal; if at least one of the K second sub-results is abnormal, the second diagnosis result is that the data is abnormal.

9. The abnormality diagnosis method of surface water automatic monitoring data according to claim 8, characterized in that: Different types of prediction indicators have different benchmark errors. The expression of the benchmark error of the kth prediction indicator is: Among them, Bwc k is the benchmark error of the k-th prediction indicator; is the kth prediction sample index in the output data of the vth prediction sample data; is the kth sample standard index in the output data of the vth target sample data, v∈[1,V], V is the total number of target sample data; Δbzc k is the standard deviation of the k-th prediction indicator.

10. An abnormality diagnosis system for automatic surface water monitoring data, characterized in that: include: Multiple monitoring nodes and data anomaly diagnosis centers; Among them, the monitoring node is used to send raw data, send collected data, and receive data diagnosis results. Data anomaly diagnosis center: used to execute the anomaly diagnosis method of surface water automatic monitoring data as described in any one of claims 1-9.

Images