Water conservancy facility internet of things real-time monitoring method, device, equipment and medium

Abstract

The invention relates to a water conservancy facility internet of things real-time monitoring method and device, equipment and a medium. The method comprises the following steps: acquiring water quality and quantity data, comparing the data with a preset safety threshold to generate a preliminary abnormal point location set, and processing through a sliding window algorithm and a historical time sequence model to generate a standardized abnormal report; generating an initial control command through the edge-deployed pruning LSTM model in combination with rainfall prediction data, and generating an equipment operation instruction after verification; executing the instruction to drive the equipment and collecting the response state of the equipment, and resolving conflicts through a Raft algorithm to generate a unified scheduling command; and after the unified scheduling command is executed, the model parameters and the preset safety threshold are updated through a Q-learning optimizer according to the flood discharge quantity feedback data. According to the method, real-time monitoring and intelligent regulation and control of the water conservancy facilities are achieved through data processing, intelligent decision making, collaborative scheduling, dynamic optimization and the like, and the safety, stability and intelligent level of operation of the water conservancy facilities are effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water conservancy operation and management, and particularly relates to a water conservancy facility Internet of Things real-time monitoring method, device, equipment and medium. BACKGROUND

[0002] The safe operation of water conservancy facilities and the efficient management of water resources are core links for ensuring national water security. With the development of the Internet of Things technology, the data collection of water conservancy facility monitoring has gradually realized automation. Based on the Internet of Things technology, a water conservancy monitoring system can collect data by deploying water level sensors, flow meters and other devices, and perform abnormal early warning based on a preset fixed threshold. However, due to instantaneous interference factors such as water flow fluctuations and device noise, false positives are easily generated, and often require manual secondary verification, which seriously affects the monitoring efficiency. Moreover, water conservancy facility control commands are usually set by a single node decision model, and in extreme weather conditions such as heavy rain, real-time rainfall prediction data cannot be combined for dynamic adjustment, resulting in insufficient control accuracy. In addition, although the system can realize collaborative monitoring of multiple edge nodes, in actual operation, the control commands of upstream and downstream water conservancy facilities such as gates and pumping stations often have hydraulic conflicts. For example, the flood discharge capacity of the upstream gate does not match the drainage capacity of the downstream pumping station, which not only reduces the efficiency of flood discharge, but also may cause the risk of river overload. SUMMARY

[0003] Therefore, it is necessary to provide a water conservancy facility Internet of Things real-time monitoring method, device, equipment and medium to improve the safety of water conservancy facility operation and the efficiency of water resource management, and realize real-time monitoring, abnormal processing and collaborative scheduling control of the operation state of water conservancy facilities.

[0004] In a first aspect, the application provides a water conservancy facility Internet of Things real-time monitoring method, comprising:

[0005] Obtaining water quality and quantity data, generating a preliminary abnormal point set by comparing the water quality and quantity data with a preset safety threshold in real time, wherein the preliminary abnormal point set contains an abnormal type code, a location coordinate and an over-standard amplitude data, and calculating a time change gradient by a sliding window algorithm according to the preliminary abnormal point set, and calling a pre-trained historical time series model to delete transient interference points to generate a standardized abnormal report;

[0006] Based on the standardized abnormal report, inferring by an edge-deployed pruning LSTM lightweight model, and combining rainfall prediction data to correct the output value to generate an initial control command, and converting the initial control command into a Modbus protocol control word after redundancy check and device register state verification to generate a device operation instruction;

[0007] The target water conservancy equipment is driven to run by executing the equipment running instruction, the actual response state of the equipment is collected, the actual response state of the equipment includes a response delay time and an execution deviation value, and based on the initial control command of the multiple edge nodes and the response state of the equipment, the instruction conflict is resolved through a Raft distributed consistency algorithm, and a unified scheduling command is generated;

[0008] After the unified scheduling command is executed, flood discharge feedback data is collected, the deviation between the flood discharge set value and the actual flood discharge is analyzed by a Q-learning optimizer, dynamic regulation parameters and safety threshold offset are generated, and the pruning LSTM lightweight model and the preset safety threshold are updated according to the dynamic regulation parameters and the safety threshold offset, respectively. The flood discharge feedback data includes instantaneous flow and cumulative flow.

[0009] In one of the embodiments, water quality and quantity data are acquired, and a preliminary abnormal point set is generated by comparing the water quality and quantity data with the preset safety threshold in real time, including:

[0010] Real-time water level data and real-time flow data are collected to obtain water quality and quantity data;

[0011] The real-time water level data and the real-time flow data are compared with the preset water level safety threshold and the preset flow safety threshold, respectively, to obtain a comparison result;

[0012] When the comparison result is that there are data points exceeding the preset water level safety threshold or the preset flow safety threshold, the data points are aggregated according to geographical location and time stamp to generate a preliminary abnormal point set.

[0013] In one of the embodiments, according to the preliminary abnormal point set, a time change gradient is calculated by a sliding window algorithm, and a pre-trained historical time series model is called to delete transient interference points to generate a standardized abnormal report, including:

[0014] The time change gradient of adjacent time points is calculated for the preliminary abnormal point set by using a sliding window algorithm;

[0015] The time change gradient is compared and interference screening is performed by using a pre-trained historical time series model, interference points with a time change gradient exceeding a preset fluctuation threshold in the preliminary abnormal point set are removed, and screened abnormal points are obtained;

[0016] The screened abnormal points are structured and encapsulated to obtain a standardized abnormal report, and the standardized abnormal report includes a time stamp, an abnormal level code, a confidence score and an abnormal type code.

[0017] In one of the embodiments, based on the standardized abnormal report, an inference is performed by using an edge-deployed pruning LSTM lightweight model, and an output value is corrected in combination with rainfall prediction data to generate an initial control command, including:

[0018] The abnormal type code and the abnormal level code in the standardized abnormal report are input into the pruning LSTM lightweight model for forward inference to generate a basic gate opening degree adjustment coefficient and a basic pump station power adjustment amount;

[0019] Obtain rainfall prediction data of the target area, the rainfall prediction data including rainfall intensity level and duration data;

[0020] And according to the rainfall prediction data, the basic gate opening degree adjustment coefficient and the basic pump station power adjustment amount are graded corrected to obtain the corrected adjustment parameters;

[0021] Based on the corrected adjustment parameters, an initial control command is generated, the initial control command including a gate opening adjustment value, a pump station power adjustment value and a time validity label.

[0022] In one of the embodiments, the initial control command is converted into a Modbus protocol control word after redundancy check and device register state verification to generate a device operation instruction, including:

[0023] According to the initial control command, a redundancy check calculation is performed and a check code difference is compared, and whether the difference between the timestamp of the initial control command and the current system time exceeds the preset retransmission threshold is verified, the instructions with the difference less than the preset retransmission threshold are screened out, and an effective control command set is generated;

[0024] Based on the effective control command set, a controllable mode flag bit in the state register of the target device is obtained, and a control command with a passed device state verification is generated based on the controllable mode flag bit;

[0025] The control command with the passed device state verification is encapsulated according to the Modbus-RTU communication specification to generate a device operation instruction.

[0026] In one of the embodiments, the target water conservancy device is driven to operate by executing the device operation instruction, and the actual response state of the device is collected, including:

[0027] According to the device address and the function code in the device operation instruction, the target water conservancy device is driven to perform operation, the target water conservancy device including a gate motor and a pump station frequency converter;

[0028] The real-time gate opening data and the real-time pump station current data are obtained through the gate opening sensor and the pump station current transformer, respectively;

[0029] Based on the real-time gate opening data and the gate opening adjustment value in the device operation instruction, an absolute difference value is calculated, and the response delay time from the sending time of the device operation instruction to the operation time of the target water conservancy device is recorded;

[0030] The absolute difference value is taken as an opening execution deviation value, combined with the error value of the real-time pump station current data and the power adjustment value in the device operation instruction, to generate a device actual response state.

[0031] In one of the embodiments, based on the initial control commands of multiple edge nodes and the device response state, the instruction conflicts are resolved by the Raft distributed consistency algorithm to generate unified scheduling commands, including:

[0032] According to the geographic coordinate information in the initial control commands of multiple edge nodes, data channels are divided by 5G network slices according to a preset transmission period, and data aggregation processing is performed according to the basin partition to generate a cross-node collaborative data set;

[0033] Based on a preset water conservancy facility topology relationship diagram, the gate opening adjustment value and the pump station power adjustment value in the cross-node collaborative data set are subjected to topology relationship analysis to identify the hydraulic dependence relationship of upstream and downstream devices, mark the gate opening adjustment value and the pump station power adjustment value that produce flood discharge conflict risks, generate a conflict instruction point set, and the conflict instruction point set includes effective reserved instructions and conflict invalid identifiers;

[0034] The conflict instruction point set is subjected to consistency voting processing by the Raft distributed consistency algorithm to obtain a conflict resolution instruction set;

[0035] The flood control priority rule library is called to perform weighted priority calculation on the effective reserved instructions in the conflict resolution instruction set according to the execution deviation value in the device actual response state and the real-time measured water level difference value to generate unified scheduling commands, and the unified scheduling commands include facility adjustment parameters and instruction effective time windows of each edge node.

[0036] In a second aspect, the application also provides a water conservancy facility Internet of Things real-time monitoring device, which comprises:

[0037] A data acquisition and anomaly detection module is configured to acquire water quality and quantity data, generate a preliminary anomaly point set by comparing the water quality and quantity data with preset safety thresholds in real time, wherein the preliminary anomaly point set contains anomaly type codes, location coordinates and over-standard amplitude data, calculate a time change gradient by a sliding window algorithm according to the preliminary anomaly point set, delete transient interference points by calling a pre-trained historical time series model, and generate a standardized anomaly report;

[0038] An edge intelligent decision-making module is configured to generate initial control commands by inferring through a pruned LSTM lightweight model deployed at the edge based on the standardized anomaly report and correcting output values in combination with rainfall prediction data, convert the initial control commands into Modbus protocol control words after redundancy verification and device register state verification, and generate device operation instructions;

[0039] The device control and cooperative scheduling module is configured to drive the target water conservancy device to operate by executing the device operation instruction, collect the actual response state of the device, the actual response state of the device including a response delay time and an execution deviation value, and generate a unified scheduling command by resolving an instruction conflict through a Raft distributed consistency algorithm based on the initial control commands of the multiple edge nodes and the response state of the device.

[0040] The parameter self-learning updating module is configured to collect flood discharge feedback data after executing the unified scheduling command, analyze the deviation between the set value of the flood discharge and the actual flood discharge through a Q-learning optimizer, generate dynamic regulation parameters and a safety threshold offset, and update the pruning LSTM lightweight model and the preset safety threshold according to the dynamic regulation parameters and the safety threshold offset, the flood discharge feedback data including instantaneous flow and cumulative flow.

[0041] In a third aspect, the present application further provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the steps in the first aspect when executing the computer program.

[0042] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps in the first aspect.

[0043] The above water conservancy facility Internet of Things real-time monitoring method, device, equipment and medium can quickly filter out preliminary abnormal points by comparing the water quality and quantity data with the preset safety threshold in real time, and further improve the accuracy of abnormal identification by using a pre-trained historical time series model to process data and filter out instantaneous interference factors. Secondly, based on the standardized abnormal report, the initial control command is generated by using the model deployed on the edge and combining the rainfall prediction data, which improves the dynamic adjustment ability and control precision of the water conservancy facility control instruction under extreme weather. Moreover, the command is converted into a Modbus protocol control word after redundancy verification and device register state verification, which ensures the reliability of instruction transmission and execution. Further, the instruction conflict is resolved based on the initial control commands of the multiple edge nodes and the response state of the device through the Raft distributed consistency algorithm, which solves the water power conflict problem of the upstream and downstream water conservancy facility control instructions. In addition, after executing the unified scheduling command, the model and the preset safety threshold are updated by analyzing the flood discharge deviation, which improves the adaptive ability and reliability of monitoring and ensures the monitoring and control of the water conservancy facility under different working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to make the technical solutions in the embodiments of the present application or the related art clearer, the accompanying drawings needed in the embodiments or the related art description will be briefly introduced. Obviously, the accompanying drawings in the following description only aim to explain the present application, and for those skilled in the art, other drawings can be obtained without creative effort.

[0045] Figure 1 A water conservancy facility Internet of Things real-time monitoring method flow chart is provided for an exemplary embodiment of the present application.

[0046] Figure 2 A water conservancy facility Internet of Things real-time monitoring device structure schematic diagram is provided for an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the technical solutions in the embodiments of the present application or the related art clearer, the accompanying drawings needed in the embodiments or the related art description will be briefly introduced. Obviously, the accompanying drawings in the following description only aim to explain the present application, and for those skilled in the art, other drawings can be obtained without creative effort.

[0048] In one embodiment, as shown in Figure 1 A water conservancy facility Internet of Things real-time monitoring method is provided, and the present embodiment takes the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is realized through the interaction of the terminal and the server. In the present embodiment, the method includes the following steps:

[0049] S101: Obtain water quality and quantity data, compare the water quality and quantity data with a preset safety threshold in real time to generate a preliminary abnormal point set, wherein the preliminary abnormal point set contains an abnormal type code, a location coordinate and an over-standard amplitude data, calculate a time change gradient through a sliding window algorithm according to the preliminary abnormal point set, call a pre-trained historical time sequence model to delete transient interference points, and generate a standardized abnormal report.

[0050] Specifically, the water level and flow data can be obtained in real time by water level sensors and electromagnetic flow meters deployed at monitoring points such as rivers and reservoirs, forming a time series data set, and obtaining water quality and quantity data. By comparing the water level and flow data with the corresponding preset safety threshold, when the threshold is exceeded, an abnormality flag can be triggered, and a preliminary abnormal point set is generated by aggregating according to geographical location and timestamp. Then, a fixed-length data window can be selected by a sliding window algorithm, and the change rate of the data in the window is calculated in turn to obtain the time change gradient, so as to further analyze the evolution trend of the abnormal points over time and determine whether the abnormality is continuously aggravated, alleviated or stable. Through the pre-trained historical time series model, instantaneous abnormal data caused by accidental factors can be identified and removed, and finally a standardized abnormal report is generated to provide standardized and accurate data support for subsequent processing.

[0051] S102: Based on the standardized abnormal report, inference is performed through the pruned LSTM lightweight model deployed at the edge, and the output value is corrected in combination with the rainfall prediction data to generate an initial control command. After redundancy checking and device register state verification, the initial control command is converted into a Modbus protocol control word to generate a device operation instruction.

[0052] Specifically, the pruned LSTM lightweight model deployed at the edge is based on a traditional LSTM model, which reduces the number of model parameters and reduces the computational complexity through model pruning technology, so that it can run efficiently on edge computing devices. Furthermore, the model can take the standardized abnormal report and rainfall prediction data as input, learn the time series characteristics of abnormal data and the influence law of rainfall on water conservancy facility operation through the gating mechanism of the LSTM network, and make inference and prediction to make the generated initial control command more suitable for actual working conditions. Further, by performing redundancy checking and device register state verification on the initial control command, the effectiveness and reliability of the control can be ensured. The redundancy checking checks whether the control command has duplicate, conflicting or redundant information to ensure the simplicity and accuracy of the command, and the device register state verification verifies whether the command is compatible with the current register state of the water conservancy device to avoid execution failure due to mismatched device state. After verification, the command can be converted into a Modbus protocol control word. Modbus protocol is a communication protocol widely used in industrial automation field, which has good universality and compatibility, so that the device operation instruction can be generated to realize precise control of water conservancy devices.

[0053] S103: The target water conservancy device is driven to operate by executing the device operation instruction, and the actual response state of the device is collected, including the response delay time and the execution deviation value. Based on the initial control commands of multiple edge nodes and the device response state, the Raft distributed consistency algorithm is used to eliminate instruction conflicts to generate a unified scheduling command.

[0054] After device operation instructions are transmitted to the target hydraulic equipment via the communication network, they can be used to drive the corresponding operation. Response delay can be determined by recording the time difference between the instruction issuance and the actual device action. Execution deviation can be calculated by comparing the difference between the device's actual operating parameters and the instruction settings. Furthermore, when multiple edge nodes simultaneously issue initial control commands, command conflicts may arise, such as when two nodes request the same gate to be opened and closed, respectively. The Raft distributed consensus algorithm coordinates edge nodes by electing a master node, ensuring that only one master node can issue valid commands at any given time. Schematically, the algorithm first conducts a voting election among edge nodes. The node with the support of more than half of the nodes becomes the master node, while the remaining nodes serve as slave nodes. The master node then receives and processes all initial control commands. Based on the device's response status, it synchronizes unified scheduling commands to all slave nodes through a log replication mechanism, thereby resolving command conflicts and ensuring the coordinated operation of hydraulic equipment.

[0055] S104: After executing the unified dispatch command, collect flood discharge feedback data, analyze the deviation between the flood discharge setting value and the actual flood discharge through the Q-learning optimizer, generate dynamic control parameters and safety threshold offsets, and update the pruned LSTM lightweight model and the preset safety threshold according to the dynamic control parameters and the safety threshold offset. The flood discharge feedback data includes instantaneous flow and cumulative flow.

[0056] Specifically, instantaneous flow rate reflects the amount of water released per unit time, while cumulative flow rate reflects the total amount of water released during the release process. The Q-learning optimizer is an optimization algorithm based on reinforcement learning. It uses the deviation between the set discharge volume and the actual discharge volume as feedback. By updating the Q-value table, it learns the optimal operating strategy under different control parameters and generates dynamic control parameters and safety threshold offsets. The dynamic control parameters can be used to update the parameters of the pruned LSTM lightweight model, enabling it to adapt to dynamic changes during the operation of water conservancy facilities. The safety threshold offset is used to adjust the preset safety threshold to better reflect actual operating conditions.

[0057] In the above method, by acquiring water quality and quantity data and comparing with the preset safety threshold, a preliminary set of abnormal point positions is generated, and by calculating the time change gradient, transient interference points can be deleted, and discrete abnormal data can be converted into structured standardized abnormal reports. Secondly, the model deployed on the edge is used in combination with local rainfall prediction data for inference correction, to generate initial control commands and convert them into Modbus protocol control words after redundancy verification and device state verification, realizing intelligent generation and reliable transmission of control commands, and improving the dynamic adjustment accuracy of control commands in extreme weather. Moreover, based on the initial control commands and device response states of multiple edge nodes, the Raft distributed consistency algorithm is used to eliminate command conflicts and generate unified scheduling commands, which can realize global consistency of cross-node collaborative scheduling. In addition, after executing the unified scheduling command, the deviation is analyzed by analyzing the real-time flood discharge feedback data, and then the model parameters and threshold can be dynamically adjusted according to the real-time working condition, enhancing the adaptability and accuracy of long-term water conservancy facility monitoring.

[0058] In one embodiment, water quality and quantity data are acquired, and by comparing the water quality and quantity data with the preset safety threshold in real time, a preliminary set of abnormal point positions is generated, including:

[0059] Real-time water level data and real-time flow data are collected respectively to obtain water quality and quantity data;

[0060] The real-time water level data and the real-time flow data are compared with the preset water level safety threshold and the preset flow safety threshold respectively to obtain comparison results;

[0061] When the comparison result is that there are data points exceeding the preset water level safety threshold or the preset flow safety threshold, the data points are aggregated according to geographical location and timestamp to generate a preliminary set of abnormal point positions.

[0062] Specifically, for example, the water level safety threshold can be determined according to the design flood level and the dead water level of the water conservancy facility, and the flow safety threshold can be set comprehensively according to the flood discharge capacity, water conveyance capacity and flood carrying capacity of the downstream river channel and other factors. For example, for a reservoir dam, the water level safety threshold range in different periods can be determined according to the design flood control high water level, normal storage level and other parameters of the reservoir dam, combined with historical flood data and basin hydrological model analysis. In the actual comparison process, real-time data can be obtained according to the set sampling frequency, and compared with the corresponding threshold. If the real-time water level data is higher than the upper limit of the preset water level safety threshold, or the real-time flow data exceeds the preset flow safety threshold range, it can be determined that the data point is abnormal, and a comparison result containing abnormal identification, data value, collection time and other information is generated. Based on the result, abnormal data points in the same area can be classified by using geographic information system technology based on the latitude and longitude coordinates of the sensor. And the abnormal data points are divided according to the set time window, and the abnormal data generated in the same time window is merged. Through the above process, the scattered abnormal data points can be structured and integrated according to the geographical position and time to generate a preliminary abnormal point set. The set can include detailed information of each abnormal point, such as abnormal type, position coordinates, abnormal occurrence time, abnormal data value, etc., providing a clear and orderly data basis for subsequent abnormal analysis and processing.

[0063] In one embodiment, according to the preliminary abnormal point set, the time change gradient is calculated by a sliding window algorithm, and the pre-trained historical time series model is called to remove transient interference points, and a standardized abnormal report is generated, including:

[0064] The time change gradient of adjacent time points is calculated for the preliminary abnormal point set by using a sliding window algorithm;

[0065] The time change gradient is compared and processed by the pre-trained historical time series model to remove interference points with time change gradient exceeding the preset fluctuation threshold in the preliminary abnormal point set, to obtain screened abnormal points;

[0066] The screened abnormal points are structured and packaged to obtain a standardized abnormal report, and the standardized abnormal report includes timestamp, abnormal level code, confidence score and abnormal type code.

[0067] Specifically, a fixed-length data window is selected by a sliding window algorithm, which is sequentially slid in the abnormal point set. For each data point in the window, the change gradient in the time period can be obtained by calculating the ratio of the difference between the data values of adjacent time points and the time interval. The time change gradient set of the entire abnormal data sequence can be obtained by traversing all data points in the preliminary abnormal point set. The selection of the window length can be set according to the actual monitoring requirements and the data update frequency. The pre-trained historical time series model can be constructed by a long short-term memory network, which is trained by inputting the time change gradient, abnormal type, environmental factors and other time series characteristics of the historical abnormal data, and outputs the judgment result of whether it is an interference point. The model parameters are continuously adjusted through the back propagation algorithm, so that the model can accurately identify the interference data pattern. When the model receives the time change gradient calculated by the sliding window algorithm, it compares and analyzes the gradient characteristics in the similar scenario in the historical data. If the time change gradient of a data point exceeds the preset fluctuation threshold, and the model judges that the change does not conform to the normal abnormal fluctuation pattern through learning historical data, it can be determined as a transient interference point and removed from the preliminary abnormal point set. Finally, the screened abnormal point set is obtained.

[0068] Finally, the screened abnormal points are structured and encapsulated to generate a standardized abnormal report, which can make the abnormal data more convenient for subsequent processing and analysis. The timestamp can be directly used as the original collection time of the abnormal point to accurately record the time when the abnormality occurs. The abnormal level code can be divided according to the degree of deviation of the abnormal data from the safety threshold, and then quantitatively expressed by a unified coding rule. The confidence score is generated by the historical time series model during the screening process, which can reflect the reliability of the model's judgment on the abnormal point. The value range is 0-1, and the closer the value is to 1, the more accurate the model's judgment of the abnormal point. By encapsulating the above information in a specific data structure such as JSON format, a standardized abnormal report can be formed to ensure accurate and efficient transmission and interaction of data between different modules and systems.

[0069] In one embodiment, based on the standardized abnormal report, inference is performed by a pruning LSTM lightweight model deployed on the edge, and the output value is corrected in combination with rainfall prediction data to generate an initial control command, including:

[0070] The abnormal type code and abnormal level code in the standardized abnormal report are input into the pruning LSTM lightweight model for forward inference to generate a basic gate opening adjustment coefficient and a basic pump station power adjustment amount;

[0071] Obtain rainfall prediction data of the target area, including rainfall intensity level and duration data;

[0072] And according to the rainfall prediction data, the basic gate opening adjustment coefficient and the basic pump station power adjustment amount are graded and corrected to obtain the corrected adjustment parameters.

[0073] Based on the corrected adjustment parameters, initial control commands are generated, including gate opening adjustment values, pump station power adjustment values, and time validity tags.

[0074] Specifically, based on the structured pruning algorithm, on the basis of the traditional LTSM model, according to the absolute value of the connection weight, a threshold is set to delete the connections below the threshold while minimizing the performance loss of the model. Then, a pruned LSTM lightweight model can be obtained. During the forward inference process of the model, the abnormal type code and the abnormal level code are first processed by one-hot encoding and converted into a vector form suitable for model input. During the training stage, the model can learn the mapping relationship between historical abnormal data and corresponding control operations. For the input abnormal information, the basic gate opening adjustment coefficient and the basic pump station power adjustment amount can be calculated to provide basic parameters for subsequent control.

[0075] Further, rainfall prediction data of the target area can be obtained, considering the influence of meteorological factors on the operation of water conservancy facilities. The rainfall intensity level can be divided into several levels such as light rain, moderate rain, heavy rain, etc. according to the meteorological industry standard, and the duration data represents the expected duration of the rainfall. Different rainfall intensity and duration have different effects on the operation of water conservancy facilities, for example, heavy rain may cause a sharp increase in inflow in a short period of time, while continuous rainfall may cause the water level to gradually rise. Then, according to the pre-set correction rule library, the basic gate opening adjustment coefficient and the basic pump station power adjustment amount are graded and corrected in combination with the rainfall prediction data. Illustratively, first, the interval of rainfall intensity level and duration can be determined, and the corresponding correction coefficient in the rule library is used to adjust the basic adjustment parameters. And different combinations of rainfall intensity level and duration correspond to different weight values. By multiplying the basic adjustment parameters by the corresponding weight values, the corrected adjustment parameters are obtained, so that the adjustment parameters can better fit the actual impact of rainfall on the operation of water conservancy facilities.

[0076] When generating initial control commands based on the corrected adjustment parameters, the gate opening adjustment value can be calculated according to the corrected gate opening adjustment coefficient combined with the current gate actual opening. Similarly, the pump station power adjustment value is calculated according to the corrected pump station power adjustment amount and the current pump station actual power. And a time validity tag can be generated based on the time range of rainfall prediction and the response delay time of water conservancy facilities to limit the effective execution time of the control command. Finally, these data are packaged according to a specific data protocol to form the initial control command, ensuring that the command can be accurately transmitted to the target water conservancy equipment and correctly executed.

[0077] In one embodiment, the initial control command is converted into a Modbus protocol control word after redundancy check and device register state verification, generating a device operation instruction, including:

[0078] According to the initial control command, a redundancy check calculation is performed and the difference of the check code is compared, and the difference between the time stamp of the initial control command and the current system time is verified to see if it exceeds a preset retransmission threshold. Instructions with a difference less than the preset retransmission threshold are screened out, and a set of valid control commands is generated.

[0079] Based on the set of valid control commands, the state register of the target device is obtained, and the state verification is performed based on the controllable mode flag, and the control command that passes the device state verification is generated.

[0080] For the control command that passes the device state verification, the device address, function code and control parameter data field are encapsulated according to the Modbus-RTU communication specification, and the device operation instruction is generated.

[0081] Specifically, according to the initial control command, a cyclic redundancy check calculation is performed, and the data is operated by a specific generating polynomial to obtain a check code. The check code is appended to the end of the initial control command to form a complete data packet containing data and check code. The data part in the data packet is recalculated at the receiving end, and the calculation result is compared with the check code carried in the data packet. If the difference is not zero, it indicates that the data has an error in the transmission process, and the initial control command is discarded. The difference between the time stamp of the initial control command and the current system time can be calculated. The preset retransmission threshold is set according to the real-time requirement of the water conservancy facility control scene, for example, in the emergency flood discharge scene, it can be set to 30 seconds. When the difference exceeds the threshold, it means that the instruction may have lost its timeliness due to transmission delay, etc., and is also screened out. Finally, only the instructions with a difference less than the preset retransmission threshold are retained to form a set of valid control commands, ensuring that the subsequent executed instruction data is accurate and has timeliness.

[0082] When performing device register state verification based on the set of valid control commands, a connection can be established with the controller of the target device through the industrial communication bus to read the controllable mode flag bit in the device state register. The device state register is a specific storage area in the target device controller for storing device running state information, and the controllable mode flag bit is one or more binary bits indicating whether the device is currently in a state where it can receive and execute control commands. Subsequently, for each instruction in the set of valid control commands, the controllable mode flag bit of the corresponding target device can be read in sequence, and a judgment can be made according to the pre-set verification rule. If the flag bit indicates that the device is in a controllable state, the control command passes the state verification, and if the device is in an uncontrollable state, the control command is marked as invalid, and the current state information of the device is recorded for subsequent troubleshooting or re-sending of the instruction.

[0083] Through the above verification process, control commands that pass the device state verification can be screened out to provide a reliable instruction basis for generating device running instructions. In addition, Modbus-RTU is a serial communication protocol widely used in the industrial field, and its data frame consists of four parts: device address, function code, data field, and check code. For control commands that pass the device state verification, a complete device running instruction can be packaged according to the Modbus-RTU communication specification. The instruction can be recognized and executed by the target water conservancy device, enabling precise control of the device and ensuring that the water conservancy facility operates according to the predetermined strategy, thereby improving the efficiency and safety of the water conservancy facility.

[0084] In one embodiment, the target water conservancy device is driven to operate by executing the device running instruction, and the actual response state of the device is collected, including:

[0085] According to the device address and function code in the device running instruction, the target water conservancy device is driven to perform operations, and the target water conservancy device includes a gate motor and a pump station frequency converter;

[0086] Real-time gate opening data and real-time pump station current data are obtained through a gate opening sensor and a pump station current transformer, respectively;

[0087] Based on the real-time gate opening data and the gate opening adjustment value in the device running instruction, an absolute difference value is calculated, and the response delay time from the sending time of the device running instruction to the running time of the target water conservancy device is recorded;

[0088] The absolute difference value is used as the opening execution deviation value, and the real-time pump station current data and the error value of the power adjustment value in the device running instruction are combined to generate the actual response state of the device.

[0089] Specifically, the target water conservancy equipment can include gate motor and pump station frequency converter and other key facilities. The device address of the device operation instruction can be used to explicitly point to the target of the instruction, and the function code further defines the specific operation type and requirement, such as the opening and closing operation of the gate motor, the power adjustment of the pump station frequency converter, etc. And in the process of device operation, the actual opening of the gate can be monitored in real time by using the gate opening sensor, the real-time gate opening data is obtained, and the current of the pump station during operation is measured in real time by using the pump station current transformer, and the real-time pump station current data is obtained. Then the real-time gate opening data and the gate opening adjustment value in the device operation instruction are compared, and the absolute difference value between the two is calculated. The difference value is the opening execution deviation value, which can measure the deviation degree of the actual opening of the gate from the instruction requirement. And the time difference between the sending time of the device operation instruction and the actual starting time of the target water conservancy equipment, that is, the response delay time, can be recorded. The response delay time reflects the response speed of the device to the instruction. Through the above indexes, the accuracy and timeliness of the device in executing the instruction can be comprehensively quantitatively evaluated. Finally, the opening execution deviation value, the current error value and the response delay time are structured and integrated, and are packaged according to a specific data format such as JSON format, so that the actual response state of the device can be obtained.

[0090] In one embodiment, based on the initial control commands of multiple edge nodes and the device response state, the instruction conflict is resolved by Raft distributed consistency algorithm to generate unified scheduling commands, including:

[0091] According to the geographic coordinate information in the initial control commands of multiple edge nodes, data channels are divided by 5G network slices according to the preset transmission period, and data aggregation processing is performed according to the basin partition to generate a cross-node collaborative data set;

[0092] Based on the preset water conservancy facility topology relationship diagram, the gate opening adjustment value and the pump station power adjustment value in the cross-node collaborative data set are analyzed for topology relationship, the water power dependence relationship of upstream and downstream devices is identified, the gate opening adjustment value and the pump station power adjustment value that produce flood discharge conflict risk are marked, a conflict instruction point set is generated, and the conflict instruction point set includes effective reserved instructions and conflict invalid identification;

[0093] The conflict instruction point set is processed by Raft distributed consistency algorithm for consistency voting to obtain a conflict resolution instruction set;

[0094] The flood control priority rule library is called, the execution deviation value in the device actual response state and the real-time measured water level difference value are used to calculate the weighted priority of the effective reserved instructions in the conflict resolution instruction set, and a unified scheduling command is generated, which includes facility adjustment parameters and instruction effective time window of each edge node.

[0095] Specifically, the 5G network slicing technology can virtualize physical network resources into multiple logically isolated network slices. Each slice can allocate corresponding bandwidth, delay, reliability, and other network resources according to the different needs of water conservancy monitoring and control to meet real-time requirements. And according to the basin partition, the data of each edge node is aggregated and processed, and the data of the edge nodes scattered in different geographical locations is integrated into a complete cross-node collaborative data set, providing a data basis for subsequent unified scheduling. And the preset water conservancy topology relationship diagram depicts the upstream and downstream connection relationship and hydraulic dependency relationship between devices such as gates and pump stations in water conservancy in the form of a graph structure, for example, the opening degree change of the upstream gate will affect the water level and flow of the downstream river. Subsequently, a graph traversal algorithm can be used to analyze the cross-node collaborative data set, starting from each water conservancy node, traversing its upstream and downstream related device nodes according to the connection edges in the topology relationship diagram, and calculating the influence of different device operations on water level and flow. And the preset hydraulic model can also be used to simulate the hydraulic state changes under different instruction combinations. If it is found that a combination of gate opening adjustment value and downstream pump station power adjustment value leads to water level exceeding the safety threshold or water flow backflow, etc., it can be determined that the combination has a flood discharge conflict risk, and the corresponding instruction is marked as a conflict instruction and recorded in the conflict instruction point set. The conflict instruction point set can be stored in a structured data format, and each instruction record contains instruction content, device identification involved, conflict type, effective reservation instruction identification, and conflict invalid identification, etc.

[0096] Further, the conflict instruction point set can be processed by the Raft distributed consistency algorithm for consistency voting to generate a conflict resolution instruction set, ensuring that all nodes have consistency in the processing results of the conflict instructions. In addition, the flood control priority rule base can be set based on the flood control design standards and historical experience of water conservancy facilities, defining the priority calculation rules of various control instructions under different working conditions. For each effective reservation instruction in the conflict resolution instruction set, the flood control priority rule base can be traversed to find rules that meet the current conditions. Each rule corresponds to a priority weight value. Then, according to the matched rule weight value, combined with the importance of the device involved in the instruction, the influence degree of instruction execution on the overall flood control effect, etc., the weighted priority score can be calculated. Then, according to the weighted priority score, the effective reservation instruction is sorted, and the sorting result is integrated with the facility adjustment parameters involved in the instruction, the instruction effective time window, etc. According to the preset data format, a unified scheduling command is packaged and generated to ensure that each edge node can execute the flood control scheduling operation in coordination according to the command.

[0097] For example, as shown in FIG. 7, the conflict instruction point set generated by the conflict detection module 702 can be sent to the conflict resolution module 703 for consistency voting and conflict resolution. The conflict resolution module 703 can generate a conflict resolution instruction set based on the conflict instruction point set, and the conflict resolution instruction set can be sent to the unified scheduling module 704 for unified scheduling. The unified scheduling module 704 can generate a unified scheduling command based on the conflict resolution instruction set, and the unified scheduling command can be sent to the edge node 701 for execution. Figure 2As shown, based on the same inventive concept, the embodiments of the present application also provide a water conservancy facility Internet of Things real-time monitoring device 200 for implementing the above-mentioned water conservancy facility Internet of Things real-time monitoring device. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more water conservancy facility Internet of Things real-time monitoring device embodiments provided below can refer to the limitations of the water conservancy facility Internet of Things real-time monitoring method in the foregoing, which will not be repeated here. The device comprises:

[0098] The data acquisition and anomaly detection module 201 is configured to acquire water quality and quantity data, generate a preliminary anomaly point set by comparing the water quality and quantity data with a preset safety threshold in real time, wherein the preliminary anomaly point set comprises an anomaly type code, a location coordinate, and an over-standard amplitude data, calculate a time change gradient through a sliding window algorithm according to the preliminary anomaly point set, call a pre-trained historical time series model to delete transient interference points, and generate a standardized anomaly report;

[0099] The edge intelligent decision-making module 202 is configured to infer through a pruned LSTM lightweight model deployed at the edge based on the standardized anomaly report, correct an output value in combination with rainfall prediction data, generate an initial control command, and convert the initial control command into a Modbus protocol control word after redundancy checking and device register state verification to generate a device operation instruction;

[0100] The device control and collaborative scheduling module 203 is configured to drive a target water conservancy device to operate by executing the device operation instruction, collect a device actual response state, wherein the device actual response state comprises a response delay time and an execution deviation value, and eliminate instruction conflicts through a Raft distributed consistency algorithm based on the initial control commands of multiple edge nodes and the device response state to generate a unified scheduling command.

[0101] The parameter self-learning update module 204 is configured to collect flood discharge feedback data after executing the unified scheduling command, analyze the deviation between a flood discharge set value and an actual flood discharge through a Q-learning optimizer to generate dynamic control parameters and a safety threshold offset, and update the pruned LSTM lightweight model and the preset safety threshold according to the dynamic control parameters and the safety threshold offset, wherein the flood discharge feedback data comprises instantaneous flow and cumulative flow.

[0102] In the above device, the data acquisition and anomaly detection module 201 can effectively filter interference data by acquiring water quality and quantity data and comparing with the preset safety threshold, combined with the sliding window algorithm and historical time series model processing, to provide data basis for subsequent decision-making. The edge intelligent decision-making module 202 generates control instructions by using the pruned LSTM model deployed on the edge combined with rainfall prediction, which are converted into device operation instructions after verification, significantly improving the control accuracy of the device under extreme weather. The device control and collaborative scheduling module 203 executes the instructions and collects the device response state, and uses the Raft algorithm to eliminate instruction conflicts, effectively solving the water power conflict problem of upstream and downstream facilities. The parameter self-learning update module 204 optimizes the model parameters and safety threshold according to the flood discharge feedback data, realizes closed-loop feedback optimization, and makes the system have self-adaptive ability.

[0103] In an exemplary embodiment, the present application also provides a computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the water conservancy facility Internet of Things real-time monitoring method of the present application when executing the computer program. Preferably, a multi-core processor is used to improve the parallel processing capability of the system. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate a large amount of supply information and computing tasks.

[0104] In an exemplary embodiment, the present application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the water conservancy facility Internet of Things real-time monitoring method of the present application.

[0105] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A real-time monitoring method for water conservancy facilities Internet of Things, characterized in that: The method comprises: Acquire water quality and quantity data, compare the water quality and quantity data with preset safety thresholds in real time, generate a preliminary abnormal point set, wherein the preliminary abnormal point set includes an abnormality type code, location coordinates, and exceedance magnitude data; calculate the time variation gradient based on the preliminary abnormal point set using a sliding window algorithm, call a pre-trained historical time series model to delete instantaneous interference points, and generate a standardized abnormality report; Based on the standardized exception report, the edge-deployed pruned LSTM lightweight model is used for inference, and the output value is corrected in combination with the rainfall forecast data to generate an initial control command. The initial control command is then converted into a Modbus protocol control word after redundancy check and device register status verification to generate a device operation instruction. The target water conservancy equipment is driven to operate by executing the equipment operation instruction, and the actual response status of the equipment is collected, wherein the actual response status of the equipment includes the response delay time and the execution deviation value. Based on the initial control command and the equipment response status of multiple edge nodes, the command conflict is resolved through the Raft distributed consensus algorithm to generate a unified scheduling command; After executing the unified scheduling command, flood discharge feedback data is collected, and the deviation between the flood discharge set value and the actual flood discharge is analyzed through the Q-learning optimizer to generate dynamic control parameters and safety threshold offsets. The pruned LSTM lightweight model and the preset safety threshold are updated according to the dynamic control parameters and the safety threshold offset respectively. The flood discharge feedback data includes instantaneous flow and cumulative flow.

2. The method according to claim 1, characterized in that The acquiring of water quality and quantity data and generating a preliminary abnormal point set by comparing the water quality and quantity data with a preset safety threshold in real time include: respectively collecting real-time water level data and real-time flow data to obtain the water quality and quantity data; Comparing the real-time water level data and the real-time flow data with a preset water level safety threshold and a preset flow safety threshold, respectively, to obtain a comparison result; When the comparison result shows that there are data points that exceed the preset water level safety threshold or the preset flow safety threshold, the data points are aggregated according to geographic location and timestamp to generate the preliminary abnormal point set.

3. The method according to claim 1, characterized in that The method calculates the time variation gradient based on the preliminary abnormal point set through a sliding window algorithm, calls a pre-trained historical time series model to delete instantaneous interference points, and generates a standardized abnormality report, including: Using a sliding window algorithm to calculate the time variation gradient of adjacent time points for the preliminary abnormal point set; Performing interference screening by comparing the time variation gradient with the pre-trained historical time series model, removing interference points in the preliminary abnormal point set whose time variation gradient exceeds a preset fluctuation threshold, and obtaining screened abnormal points; The screened abnormal points are subjected to structured packaging processing to obtain the standardized abnormality report, which includes a timestamp, an abnormality level code, a confidence score and the abnormality type code.

4. The method according to claim 1, wherein Based on the standardized exception report, the edge-deployed pruned LSTM lightweight model is used for reasoning, and the output value is corrected in combination with the rainfall forecast data to generate the initial control command, including: Inputting the abnormality type code and abnormality level code in the standardized abnormality report into the pruned LSTM lightweight model for forward reasoning to generate the basic gate opening adjustment coefficient and the basic pump station power adjustment amount; Acquiring the rainfall forecast data for the target area, the rainfall forecast data including rainfall intensity level and duration data; and performing graded correction on the basic gate opening adjustment coefficient and the basic pump station power adjustment amount according to the rainfall forecast data to obtain a correction adjustment parameter; The initial control command is generated based on the modified adjustment parameters, and the initial control command includes a gate opening adjustment value, a pump station power adjustment value and a time validity tag.

5. The method according to claim 1, wherein The initial control command is converted into a Modbus protocol control word after redundancy check and device register status verification to generate a device operation instruction, including: Perform redundancy check calculation and compare check code difference according to the initial control command, verify whether the difference between the timestamp of the initial control command and the current system time exceeds a preset retransmission threshold, filter out instructions whose difference is less than the preset retransmission threshold, and generate a valid control command set; Based on the valid control command set, obtaining a controllable mode flag in a status register of the target device, performing status verification based on the controllable mode flag, and generating a control command that passes the device status verification; For the control command that passes the device status verification, the device address, function code and control parameter data field are encapsulated according to the Modbus-RTU communication specification to generate the device operation instruction.

6. The method according to claim 5, characterized in that The step of driving the target water conservancy equipment to operate by executing the equipment operation instruction and collecting the actual response status of the equipment includes: According to the device address and function code in the device operation instruction, the target water conservancy equipment is driven to perform an operation, and the target water conservancy equipment includes a gate motor and a pump station frequency converter; Through the gate opening sensor and the pump station current transformer, real-time gate opening data and real-time pump station current data are obtained respectively; Calculating an absolute difference between the real-time gate opening data and the gate opening adjustment value in the equipment operation instruction, and recording the response delay time from the time the equipment operation instruction is sent to the time the target water conservancy equipment is operated; The absolute difference is used as the opening execution deviation value, and combined with the error value of the real-time pump station current data and the power adjustment value in the equipment operation instruction to generate the actual response state of the equipment.

7. The method according to claim 1, characterized in that The initial control command based on multiple edge nodes and the device response status is used to resolve instruction conflicts through the Raft distributed consensus algorithm to generate a unified scheduling command, including: Based on the geographic coordinate information in the initial control commands of the multiple edge nodes, the data channels are divided according to the preset transmission period through 5G network slicing, and data aggregation processing is performed according to the watershed partitions to generate a cross-node collaborative data set; Based on a preset water conservancy facility topology relationship diagram, a topological relationship analysis is performed on the gate opening adjustment values ​​and pump station power adjustment values ​​in the cross-node collaborative data set to identify the hydraulic dependency of upstream and downstream equipment, mark the gate opening adjustment values ​​and pump station power adjustment values ​​that generate flood discharge conflict risks, and generate a conflict instruction point set, which includes valid retention instructions and conflict invalidation flags; Performing a consensus vote on the conflicting instruction point set using the Raft distributed consensus algorithm to obtain a conflict resolution instruction set; Call the flood control priority rule library, and perform weighted priority calculation on the valid retention instructions in the conflict resolution instruction set based on the execution deviation value and the real-time measured water level difference in the actual response state of the device to generate the unified scheduling command, which includes the facility adjustment parameters of each edge node and the instruction effective time window.

8. A real-time monitoring device for water conservancy facilities Internet of Things, characterized in that: The device comprises: A data acquisition and anomaly detection module is used to obtain water quality and quantity data, generate a preliminary anomaly point set by comparing the water quality and quantity data with preset safety thresholds in real time, wherein the preliminary anomaly point set includes anomaly type code, location coordinates, and exceedance magnitude data; and calculate the time variation gradient based on the preliminary anomaly point set using a sliding window algorithm, call a pre-trained historical time series model to delete instantaneous interference points, and generate a standardized anomaly report; An edge intelligent decision-making module is configured to perform reasoning based on the standardized exception report using a pruned LSTM lightweight model deployed at the edge, and to correct the output value in combination with rainfall forecast data to generate an initial control command. The initial control command is then converted into a Modbus protocol control word after performing redundancy check and device register status verification to generate a device operation instruction. The device control and collaborative scheduling module is used to drive the operation of the target water conservancy equipment by executing the device operation instructions, collect the actual response status of the equipment, and resolve instruction conflicts based on the initial control commands and the device response status of multiple edge nodes using the Raft distributed consensus algorithm to generate a unified scheduling command; A parameter self-learning update module is used to collect flood discharge feedback data after executing the unified scheduling command, analyze the deviation between the flood discharge set value and the actual flood discharge through the Q-learning optimizer, generate dynamic control parameters and safety threshold offsets, and update the pruned LSTM lightweight model and the preset safety threshold according to the dynamic control parameters and the safety threshold offset. The flood discharge feedback data includes instantaneous flow and cumulative flow.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images