Real-time prediction method for reservoir desilting of multi-sand river based on multi-parameter measurement and machine learning algorithm

Abstract

The present application relates to a kind of multi-sand river reservoir desilting real-time prediction method based on multi-parameter measurement and machine learning algorithm, including reservoir multi-parameter data acquisition device, integrated controller, data wireless sending device, reservoir desilting real-time prediction monitoring system, database and application server, six parts of field feedback prompt device;Wherein integrated controller is embedded with five kinds of machine learning core algorithms based on Python language programming limit gradient promotion, support vector machine, K nearest neighbor algorithm, random forest, Gaussian process regression, can analyze the data collected, realize the real-time prediction analysis of reservoir desilting;The prediction model established can realize the real-time prediction of different multi-sand river reservoir desilting process, and the applicability and generalization are strong, to provide strong theoretical and technical support for the comparison and optimization of reservoir scheduling scheme.

Description

Technical Field

[0001] This invention relates to a real-time prediction method for sediment discharge from reservoirs in sandy rivers based on multi-parameter measurement and machine learning algorithms, belonging to the field of reservoir sediment research. Background Technology

[0002] Constructing reservoirs on sandy rivers inevitably leads to siltation problems. Siltation hinders the reservoir's ability to regulate water flow, shortens its lifespan, and negatively impacts its flood control function and the safety of the dam itself. Reservoir sediment discharge is a widely adopted method worldwide for mitigating siltation and extending reservoir lifespan. Practice has proven that utilizing floodwaters during the flood season or pre-flood discharge to remove accumulated sediment from the reservoir can achieve excellent results and benefits. Therefore, research on reservoir sediment discharge prediction is of great significance.

[0003] Currently, the main research methods used in domestic and international studies on reservoir sediment discharge processes include measured data analysis, physical model testing, and numerical simulation.

[0004] The measured data analysis method is a statistical analysis method based on a large amount of measured data. It uses long-term series of reservoir hydrological data to summarize and analyze the sedimentation and sediment discharge processes of various reservoirs. While this method is fundamental to reservoir sediment discharge research, it is limited in several ways. First, it often only analyzes the correlation of individual factors, making it difficult to examine the combined effects of numerous influencing factors. Second, it focuses on trend analysis of sedimentation and sediment discharge changes in specific reservoirs, thus limiting its applicability and accuracy. Furthermore, its application in wider adoption also has limitations.

[0005] The physical model testing method is a method that uses certain similarity theories of water flow and sediment movement (geometric similarity, gravity similarity, resistance similarity, sediment transport similarity, settlement similarity, etc.) and parameters such as boundary conditions and dynamic conditions of actual engineering projects to establish a reservoir sediment engineering model at a certain scale. Then, based on the model tests conducted, the actual water and sediment changes are simulated and analyzed. While the physical model testing method can guide engineering practice to a certain extent, it is limited by the complexity of actual engineering projects and cannot completely simulate real-world conditions. Furthermore, it is relatively costly, requiring significant time, financial, and material resources.

[0006] Numerical simulation is a method based on the hydrodynamic principles of sediment movement and water flow. It involves establishing mathematical models of water and sediment in different dimensions (one-dimensional, two-dimensional, or three-dimensional) under certain initial and boundary conditions to simulate and predict sediment changes in reservoirs. While numerical simulation has clear physical meaning, it suffers from drawbacks such as the need to meet certain assumptions, complexity, numerous computational parameters, and difficulty in parameter tuning. Therefore, its application in the actual management of reservoirs remains somewhat limited.

[0007] In recent years, machine learning (ML) algorithms, which excel at solving big data analysis problems, have developed rapidly and have been widely applied in various scientific fields such as air quality forecasting, geographic information prediction, disease diagnosis, and fault analysis, achieving excellent results. Machine learning algorithms are essentially data analysis methods that automatically analyze and model data, and then use those models to make predictions. They are effective methods for classifying or regressing nonlinear systems. Considering the numerous influencing factors of reservoir sediment discharge, which involve multi-factor and nonlinear relationships, using machine learning methods can effectively overcome the challenge of establishing complex relationships among multiple factors.

[0008] In addition, radar level gauges, Doppler flow meters, sediment concentration measuring instruments, and multibeam echo sounders are all mature data acquisition equipment and technologies for reservoir sediment, which can provide advanced technical support for real-time prediction of reservoir sediment discharge.

[0009] In summary, in order to more effectively solve the problem of reservoir siltation and fully realize the benefits of reservoirs, it is necessary to propose a real-time prediction method for sediment discharge from reservoirs in sediment-rich rivers based on multi-parameter measurement and machine learning algorithms. This method can further meet the scheduling and operation needs of reservoirs in sediment-rich rivers in my country and provide technical support for the rational scheduling and safe operation of reservoirs. Summary of the Invention

[0010] This invention aims to overcome the shortcomings of existing research methods and further provide technical support for the rational scheduling and safe operation of reservoirs. It combines reservoir measurement with machine learning algorithms to propose a real-time prediction method for sediment discharge processes in reservoirs of sediment-rich rivers.

[0011] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0012] The real-time prediction method for sediment discharge from reservoirs in sandy rivers based on multi-parameter measurement and machine learning algorithms includes six parts: a reservoir multi-parameter data acquisition device, an integrated controller, a wireless data transmission device, a real-time prediction and monitoring system for sediment discharge from the reservoir, a database and application server, and a field feedback and prompting device.

[0013] The reservoir multi-parameter data acquisition device is responsible for collecting reservoir sediment data, which is received in real time by the integrated controller. The integrated controller incorporates five core machine learning algorithms written in Python: Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Gaussian Process Regression (GPR). Based on these algorithms, the collected data is analyzed to achieve real-time prediction and analysis of reservoir sediment discharge. The wireless data transmission device receives data from the integrated controller and sends it to the database and application server. The database and application server receive and store the data transmitted by the wireless data transmission device and determine whether the sediment concentration at the discharge meets the standard requirements. The determination result is displayed in real time in the reservoir sediment discharge real-time prediction and monitoring system and is received in real time by the on-site feedback device.

[0014] The reservoir multi-parameter data acquisition device comprises four parts: a radar level gauge, a Doppler flow meter, a sediment concentration meter, and a multibeam echo sounder. Each part collects different sediment data from the reservoir. The radar level gauge is used to collect the real-time water level W upstream of the dam; the Doppler flow meter is used to collect the real-time inflow rate into the reservoir. and outbound flow The sediment concentration measuring instrument is used to collect the sediment concentration of water entering the reservoir in real time. Multibeam echo sounders are used to collect the siltation volume G of reservoirs.

[0015] The main component of the integrated controller is the data controller, which acquires the upstream water level W from the radar water level gauge and the inflow rate from the Doppler flow meter in real time via an RS485 interface. and outbound flow The sediment content of the incoming material collected by the sediment content measuring instrument The data includes the sedimentation volume G obtained from the multibeam echo sounder, and the real-time data acquisition time t; simultaneously, the water level difference in front of the dam is calculated from the water level data W in front of the dam. Combined with machine learning algorithm prediction models, the real-time predicted sediment content of the reservoir is calculated. .

[0016] The real-time predicted sediment content of the reservoir The specific calculation method is as follows:

[0017] (1) Using existing water and sediment data, preprocess and segment the existing data to obtain training data, and substitute them into five pre-programmed machine learning algorithms: XGBoost, SVM, KNN, RF, and GPR. Specifically, the existing training data... , , , , , These six variables, used as input variables, will include the sand content in the existing training data. As the sole output variable, five different reservoir sediment discharge prediction models were established, comprehensively considering various influencing factors. Simultaneously, the predicted sediment concentration at the reservoir outlet was obtained by substituting test data into the models. The established model takes the following form:

[0018]

[0019] In the formula, For regression functions; The amount of sand released from the existing training data; The inbound traffic from the existing training data; The amount of sand content input into the existing training data; The outbound traffic from the existing training data; This refers to the amount of data accumulated in the existing training data. The water level in front of the dam is from the existing training data; The water level difference in front of the dam is given in the existing training data.

[0020] (2) Next, an error comparison analysis of the prediction results of five different models was conducted, taking the prediction of sediment concentration in the reservoir as an example. Compared with existing test data on sand content in the outflow The mean absolute error (MAE) and root mean square error (RMSE) between the two are used as evaluation criteria, and the algorithm model with the smallest error and highest accuracy is selected accordingly. The formulas for calculating the mean absolute error (MAE) and root mean square error (RMSE) are as follows:

[0021]

[0022]

[0023] In the formula, n is the number of test data samples; To predict the sediment content at the outlet; The sand content at the point of release from the warehouse is based on existing test data;

[0024] (3) Input the selected algorithm model into the integrated controller module, and input the real-time acquired W, , , G data, and the calculated variable data Substituting the values ​​into the model yields the real-time predicted sediment concentration at the outlet. .

[0025] The wireless data transmission device is responsible for receiving data sent by the integrated controller, including the upstream water level W and inflow rate collected by the reservoir multi-parameter data acquisition device. Outbound flow Sand content entering the reservoir The amount of siltation G, and the water level difference in front of the dam calculated by the integrated controller. Predicted sediment content at the outlet The data is collected in real time, t. The collected data is wirelessly transmitted to a remote database and application server via a 4G or 5G network.

[0026] The database and application server are mainly used to receive data sent from the wireless data transmission device and store it in a timely manner;

[0027] Meanwhile, the database and application server use the standard value of outflow sediment concentration input from the reservoir sediment discharge real-time prediction and monitoring system. Determine the predicted sediment concentration of the reservoir as calculated currently. Whether it meets the standard requirements; the criteria for judgment are as follows: ,in %, represents the allowable value for relative error. The calculation formula is as follows:

[0028]

[0029] In the formula, To predict the relative error between the outflow sand content and the standard value of outflow sand content; The predicted sediment content at the outlet; This is the standard value for the sand content at the outlet.

[0030] If the requirements are not met, a message indicating that the sediment content data leaving the reservoir does not meet the standard requirements will be sent to the monitoring system via the Internet; at the same time, a message will be sent to the on-site feedback device of the reservoir to promptly inform the on-site management personnel so that they can take appropriate measures.

[0031] The real-time sediment discharge prediction and monitoring system for reservoirs is installed on a desktop computer or mobile laptop connected to the Internet at the reservoir site. Before monitoring begins, operators use the system to set the current basic information of the reservoir and the standard for sediment concentration at discharge, and store this information in a database and application server via the Internet for subsequent use. During monitoring, the system interface displays the real-time predicted sediment concentration at discharge. Process curve of change over time t;

[0032] In addition, the monitoring system receives alerts from the database and application server indicating that the sediment content at the reservoir outlet does not meet the requirements, and alerts the on-site reservoir management personnel so that they can take timely countermeasures.

[0033] The on-site feedback and prompting device receives real-time prompts from the database and application server indicating that the sediment concentration at the outlet does not meet the requirements. Based on the feedback and prompts received by the device, the reservoir on-site management personnel take timely measures to change the sediment concentration at the outlet.

[0034] Beneficial effects of the invention

[0035] (1) This invention combines reservoir measurement with machine learning algorithms to overcome the shortcomings of existing research methods, such as the need to meet certain assumptions, complex processes, and numerous calculation parameters.

[0036] (2) The present invention comprises six parts: a reservoir multi-parameter data acquisition device, an integrated controller, a data wireless transmission device, a reservoir sediment discharge real-time prediction and monitoring system, a database and application server, and a field feedback and prompting device. It can collect reservoir sediment data in real time and automatically, and establish an excellent prediction model for the sediment discharge process of multi-sand reservoirs based on machine learning algorithms, which comprehensively considers various influencing factors, thereby improving the accuracy and precision of the prediction.

[0037] (3) This invention is based on five core machine learning algorithms written in Python: extreme gradient boosting, support vector machine, K nearest neighbor algorithm, random forest, and Gaussian process regression. Based on the core machine learning algorithms, the collected data is analyzed to realize real-time prediction and analysis of reservoir sediment discharge. The established prediction model can be applied to the sediment discharge process of various sediment-rich rivers and reservoirs, and has strong applicability and promotion.

[0038] (4) The prediction model established by this invention can realize the real-time prediction of the sediment discharge process of reservoirs in different sediment-rich rivers, thereby providing strong theoretical and technical support for the comparison and optimization of reservoir scheduling schemes. Attached Figure Description

[0039] Figure 1 A schematic diagram of the structural components of the method of the present invention.

[0040] Figure 2 A flowchart illustrating the method of this invention.

[0041] Figure 3 A schematic diagram illustrating the integration principle of the method of this invention.

[0042] Figure 4 The real-time prediction of sediment content at the outlet in the method of this invention A schematic diagram of the process curve as a function of time t. Detailed Implementation

[0043] The specific embodiments of the present invention will be further described in detail below with reference to the examples. Unless otherwise specified, the instruments and equipment involved in the examples are all conventional instruments and equipment; unless otherwise specified, the reagents involved are all commercially available conventional reagents; unless otherwise specified, the experimental methods involved are all conventional methods.

[0044] Example

[0045] This invention addresses the research on reservoir sediment discharge prediction, aiming to achieve real-time and accurate prediction and monitoring of sediment discharge processes in sediment-rich rivers. An example of this invention provides a real-time prediction method applicable to sediment discharge processes in reservoirs of sediment-rich rivers. The method's structural diagram, flowchart, and integrated principle diagram are provided in the respective examples. Figure 1 , Figure 2 and Figure 3 .

[0046] This method comprises six parts: a reservoir multi-parameter data acquisition device, an integrated controller, a wireless data transmission device, a real-time reservoir sediment discharge prediction and monitoring system, a database and application server, and a field feedback and prompting device. It also incorporates five core machine learning algorithms embedded in the integrated controller, all written in Python: Extreme Gradient Boosting (XGBoost), Support Vector Machines (SVM), K Nearest Neighbor (KNN), Random Forest (RF), and Gaussian Progress Regression (GPR). Based on these machine learning algorithms, real-time prediction and analysis of reservoir sediment discharge can be achieved.

[0047] Multi-parameter data acquisition device for a reservoir

[0048] The reservoir multi-parameter data acquisition device consists of four parts: a radar level gauge (e.g., the F-LD100 integrated radar level gauge), a Doppler flow meter (e.g., the HQ-610 ultrasonic Doppler flow meter), a sediment concentration meter (e.g., the TKCS2012 online sediment concentration meter), and a multibeam echo sounder (e.g., the EM2040C multibeam echo sounder). Each part is used to collect different reservoir sediment data. The radar level gauge is used to collect the real-time water level W upstream of the dam; the Doppler flow meter is used to collect the real-time inflow into the reservoir. and outbound flow The sediment concentration measuring instrument is used to collect the sediment concentration of water entering the reservoir in real time. Multibeam echo sounders are used to collect the siltation volume G of reservoirs.

[0049] Meanwhile, each part of the data acquisition device provides the acquired data to the integrated controller via the RS485 interface.

[0050] Two integrated controllers

[0051] The main components of the integrated controller are the data controller (including a microcontroller unit (MCU) and a field-programmable gate array (FPGA)), and an external power supply continuously powers the data controller within the integrated controller. The data controller acquires real-time water level data (W) from the radar level gauge and inflow data from the Doppler flow meter via an RS485 interface. and outbound flow data The sediment concentration data of the incoming material collected by the sediment concentration measuring instrument The data includes the sedimentation volume G collected by the multibeam echo sounder, and the real-time acquisition time t; simultaneously, the water level difference data in front of the dam is calculated from the water level data W in front of the dam. And further based on the above data (W, , , G By combining the selected machine learning algorithm prediction model, real-time predicted sediment content data for reservoir outflow is obtained. The specific application methods are as follows:

[0052] (1) Existing water and sediment data were preprocessed and segmented beforehand. The resulting training data was then fed into five pre-programmed machine learning algorithms (XGBoost, SVM, KNN, RF, GPR). , , , , , These six variables, used as input variables, will include the sand content in the existing training data. As the sole output variable, a comprehensive prediction model for sediment discharge from different reservoirs is established, taking into account various influencing factors. Simultaneously, the predicted sediment concentration at the reservoir outlet is obtained by substituting test data into the model. The established model takes the following form:

[0053]

[0054] In the formula, For regression functions; The amount of sand released from the existing training data; The inbound traffic from the existing training data; The amount of sand content input into the existing training data; The outbound traffic from the existing training data; This refers to the amount of data accumulated in the existing training data. The water level in front of the dam is from the existing training data; The difference in water level in front of the dam is given in the existing training data.

[0055] It should be noted here that: "Water level difference in front of the dam" "This takes into account the timeliness of the impact of changes in the water level in front of the dam on sediment discharge, and calculates the water level before the current day." The average water level of the day and the difference between the average water level of the day are the input variables introduced to the impact on sediment discharge on that day.

[0056] (2) Next, an error comparison analysis of the prediction results of different models is conducted, taking the predicted sediment concentration in the reservoir as an example. Compared with existing test data on sand content in the outflow The mean absolute error (MAE) and root mean square error (RMSE) between the two are used as evaluation criteria, and the optimal algorithm model with the smallest error and highest accuracy is selected accordingly. The formulas for calculating the mean absolute error (MAE) and root mean square error (RMSE) are as follows:

[0057]

[0058]

[0059] In the formula, n is the number of test data samples; To predict the sediment content at the outlet; This refers to the sand content of the material leaving the warehouse based on existing test data.

[0060] (3) Input the selected optimal algorithm model into the integrated controller module, and input the real-time acquired and calculated data of each variable (W, , , G Substituting the values ​​into the optimal model, real-time predicted sediment concentration data for reservoir discharge can be obtained. .

[0061] (4) All the above data and calculation results are cached in the FLASH memory, and this data (W, , , G , (t) is transmitted to the wireless data transmitting device.

[0062] Three-data wireless transmission device

[0063] The wireless data transmitter is responsible for receiving data sent by the integrated controller, including: the upstream water level data W and the inflow data collected by the reservoir multi-parameter data acquisition device. Outbound flow data Data on sediment content in the database The data includes siltation volume G and the water level difference in front of the dam calculated by the integrated controller. Predicted sediment content data for outflow. The system acquires various types of data in real time, and first stores the data in the FLASH memory. After caching, the data is sent to the Central Processing Unit (CPU). Then, the CPU encapsulates the data and encrypts the relevant data according to a customized TCP / IP protocol. At certain time intervals, the data is wirelessly transmitted by the wireless communication module to the remote database and application server via a 4G (4th Generation) or 5G (5th Generation) network.

[0064] Four databases and application servers

[0065] The database and application server are mainly used to receive and store data sent from the wireless data transmission device, including: upstream water level data W and inflow data. Outbound flow data Data on sediment content in the database Siltation volume data G, and calculated water level difference data in front of the dam. Predicted sediment content data for outflow. This includes all data, including the time t when acquiring various types of data in real time. In this embodiment of the invention, an Industrial Personal Computer (IPC) can be used as the database and application server hardware, and a database such as MariaDB can be used to support data applications. The server is connected to the Internet via fiber optic cable, and receives and stores all data sent to the specified IP address by the wireless data transmission device according to the set server IP address.

[0066] Meanwhile, the database and application server can predict the standard value of sediment concentration at the reservoir discharge based on the real-time sediment discharge monitoring system. To determine the currently calculated predicted sediment concentration data for reservoir discharge. Whether it meets the standard requirements. The criteria for judgment are as follows: ,in To predict the relative error between the outflow sediment concentration and the standard value of outflow sediment concentration, This is the allowable value for relative error (e.g., 10%). The calculation formula is as follows. If the requirements are not met, a notification message indicating that the outflow sediment concentration data does not meet the standard requirements will be sent to the monitoring system via the Internet. This message mainly includes: the time when the outflow sediment concentration does not meet the standard requirements and its corresponding sediment concentration value, the standard sediment concentration value, and the extent of exceedance. Simultaneously, a notification SMS message can be sent to the reservoir's on-site feedback device. This SMS message contains information such as the time when the outflow sediment concentration does not meet the standard requirements and its corresponding sediment concentration value, the standard sediment concentration value, and the extent of exceedance, providing timely notification to the reservoir's on-site management personnel so that appropriate countermeasures can be taken.

[0067]

[0068] In the formula, To predict the relative error between the outflow sand content and the standard value of outflow sand content; The predicted sediment content at the outlet; This is the standard value for the sand content at the outlet.

[0069] Real-time Prediction and Monitoring System for Sediment Discharge of Five Reservoirs

[0070] The real-time sediment discharge prediction and monitoring system for reservoirs is installed on a desktop computer or mobile laptop connected to the Internet at the reservoir site. Before real-time prediction and monitoring begins, operators can use the system to set basic reservoir information, sediment concentration standards for outflow, etc., and store this information in a database and application server via the Internet for later use. During monitoring, the system interface can display the real-time sediment concentration prediction results graph, that is, the process line of the real-time predicted outflow sediment concentration S over time t.

[0071] In addition, the monitoring system can also receive alerts from the database and application server indicating that the outgoing sand content does not meet the requirements, and mark the data points (called anomalies) showing that the outgoing sand content does not meet the requirements on the system interface, as follows: Figure 4 This serves as a visual indication; simultaneously, it alerts on-site reservoir management personnel so that timely countermeasures can be taken (such as adjusting the gate opening to change the outflow, lowering the water level in front of the dam, etc.) to change the outflow sediment content to meet the set standard requirements.

[0072] Six on-site feedback and prompting devices

[0073] The on-site feedback and alert device, powered by a mobile terminal, can receive real-time alerts from the database and application server indicating that the outflow sediment concentration does not meet requirements. Reservoir site management personnel can then take timely measures (such as adjusting gate opening to change the outflow rate or lowering the water level in front of the dam) to adjust the outflow sediment concentration based on the feedback received from the device. In this embodiment of the invention, the on-site feedback and alert device uses a smartphone; however, this invention does not impose specific limitations on its use in practical applications.

Claims

1. A real-time prediction method for sediment discharge from reservoirs in sandy rivers based on multi-parameter measurement and machine learning algorithms, characterized in that, It consists of six parts: a reservoir multi-parameter data acquisition device, an integrated controller, a wireless data transmission device, a reservoir sediment discharge real-time prediction and monitoring system, a database and application server, and a field feedback and prompting device. The reservoir multi-parameter data acquisition device is responsible for collecting reservoir sediment data, which is then received in real time by the integrated controller. The integrated controller incorporates five core machine learning algorithms written in Python: Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Gaussian Process Regression (GPR). Based on these algorithms, the acquired data is analyzed to achieve real-time prediction and analysis of reservoir sediment discharge. The main component of the integrated controller is the data controller, which acquires real-time data via an RS485 interface, including the upstream water level W from the radar level gauge, the inflow flow Q1 and outflow flow Q2 from the Doppler flowmeter, and the inflow sediment concentration S1 from the sediment concentration meter. The sedimentation volume G and the real-time acquisition time t are collected by the multibeam echo sounder; the water level difference ΔW in front of the dam is calculated from the water level data W in front of the dam. Then, the water level W in front of the dam, the inflow rate Q1, the outflow rate Q2, the inflow sediment concentration S1, the sedimentation volume G, and the water level difference ΔW in front of the dam are input into the model predicted by the machine learning algorithm to calculate the real-time predicted outflow sediment concentration S′; the wireless data transmission device is responsible for receiving the data sent by the integrated controller and sending the data to the database and application server; the database and application server are responsible for receiving and storing the data sent by the wireless data transmission device, and judging whether the outflow sediment concentration meets the standard requirements. The judgment result can be displayed in real time in the reservoir sediment discharge real-time prediction and monitoring system, and is received in real time by the on-site feedback prompt device; The water level difference ΔW in front of the dam is an input variable that considers the timeliness of the impact of water level changes in front of the dam on sediment discharge. It is calculated by comparing the average water level of the previous n days with the water level of the current day, and the difference between the two values. Here, n = 1, 2, 3, 4, 5… The reservoir sediment discharge real-time prediction and monitoring system is installed on a desktop computer or mobile laptop computer that can connect to the Internet at the reservoir site. Before the monitoring begins, the operator sets the current basic information of the reservoir and the standard for sediment concentration at discharge through the system, and stores this information in the database and application server through the Internet for subsequent use. During the monitoring process, the system interface displays the real-time process line of the predicted sediment concentration at discharge S′ changing with time t. In addition, the monitoring system receives alerts from the database and application server indicating that the sediment content at the reservoir outlet does not meet the requirements, and alerts the on-site reservoir management personnel so that they can take timely countermeasures.

2. The method for real-time prediction of sediment discharge from reservoirs in sandy rivers as described in claim 1, characterized in that, The reservoir multi-parameter data acquisition device includes four parts: a radar level gauge, a Doppler flow meter, a sediment concentration meter, and a multibeam echo sounder. Each part of the device collects different reservoir sediment data. The radar level gauge is used to collect the reservoir's upstream water level W in real time; the Doppler flow meter is used to collect the reservoir's inflow flow Q1 and outflow flow Q2 in real time; the sediment concentration meter is used to collect the inflow sediment concentration S1 in real time; and the multibeam echo sounder is used to collect the reservoir's siltation volume G.

3. The real-time prediction method for sediment discharge from reservoirs in sandy rivers as described in claim 1, characterized in that, The specific calculation method for the real-time predicted sediment concentration S′ in the reservoir is as follows: (1) Using existing water and sediment data, preprocess and segment the existing data to obtain training data, and substitute them into five pre-programmed machine learning algorithms: XGBoost, SVM, KNN, RF, and GPR. Among them, W0 and Q in the existing training data are used to perform the following operations: 10 Q 20 S 10 Using six variables—G0, ΔW0, and ΔW0—as input variables, and the sediment concentration S0 from the existing training data as the sole output variable, five different reservoir sediment discharge prediction models are established, comprehensively considering various influencing factors. Simultaneously, the predicted sediment concentration S0 from the outflow is obtained by substituting test data into the models. The established model form is as follows: S0=f(Q 10 ,S 10 ,Q 20 ,G0,W0,ΔW0) In the formula, f() is the regression function; S0 is the sediment content in the existing training data; Q 10 S represents the inbound traffic from the existing training data. 10 Q represents the sand content input into the existing training data; 20 G0 represents the outflow from the dam in the existing training data; W0 represents the siltation in the existing training data; ΔW0 represents the water level in front of the dam in the existing training data; and ΔW0 represents the water level difference in front of the dam in the existing training data. (2) Next, an error comparison analysis of the prediction results of five different models is conducted. Here, the mean absolute error (MAE) and root mean square error (RMSE) between the predicted sediment concentration S0” and the sediment concentration S0’ in the existing test data are used as the judgment indicators, and the algorithm model with the smallest error and the highest accuracy is selected accordingly. The formulas for calculating the mean absolute error (MAE) and root mean square error (RMSE) are as follows: ； ； In the formula, n is the number of test data samples; S0” is the predicted sediment content of the outflow; S0' is the sediment content of the outflow in the existing test data; (3) Input the selected algorithm model into the integrated controller module, and substitute the real-time acquired W, Q1, Q2, S1, G data, and the calculated variable data ΔW into the model to obtain the real-time predicted sediment content S′.

4. The method for real-time prediction of sediment discharge from reservoirs in sandy rivers as described in claim 1, characterized in that, The wireless data transmission device is responsible for receiving data sent by the integrated controller, including the upstream water level W, inflow Q1, outflow Q2, inflow sediment concentration S1, and siltation G collected by the reservoir multi-parameter data acquisition device, as well as the upstream water level difference ΔW, the predicted outflow sediment concentration S′, and the real-time data acquisition time t calculated by the integrated controller; the acquired data is wirelessly transmitted to a remote database and application server via a 4G or 5G network.

5. The method for real-time prediction of sediment discharge from reservoirs in sandy rivers as described in claim 1, characterized in that, The database and application server are mainly used to receive data sent from the wireless data transmission device and store it in a timely manner; Meanwhile, the database and application server determine whether the currently calculated predicted sediment concentration S′ meets the standard requirements based on the standard value [S] of the outflow sediment concentration input from the reservoir sediment discharge real-time prediction and monitoring system. The judgment criterion is: ξ≤[ξ], where [ξ]=10%, which is the allowable value for relative error. The formula for calculating ξ is as follows: ； In the formula, ξ is the relative error between the predicted sediment concentration at discharge and the standard value of sediment concentration at discharge; S′ is the predicted sediment concentration at discharge; [S] is the standard value of sediment concentration at discharge. If the requirements are not met, a message indicating that the sediment content data leaving the reservoir does not meet the standard requirements will be sent to the monitoring system via the Internet; at the same time, a message will be sent to the on-site feedback device of the reservoir to promptly inform the on-site management personnel so that they can take appropriate measures.

6. The method for real-time prediction of sediment discharge from reservoirs in sandy rivers as described in claim 1, characterized in that, The on-site feedback and prompting device receives real-time prompts from the database and application server indicating that the sediment concentration at the outlet does not meet the requirements. Based on the feedback and prompts received by the device, the reservoir on-site management personnel take timely measures to change the sediment concentration at the outlet.

Images