Multi-target intelligent hydraulic regulation and control method for tidal river network sluice group

Abstract

The invention discloses a multi-target intelligent hydraulic regulation and control method for a tidal river network sluice group. The method comprises the following steps: S1, acquiring historical detection data and hydrological data of a tidal river network; and S2, constructing a river network flow prediction model and a water level prediction model to obtain river network key section hydrodynamic data. And S3, based on a multi-objective optimization thought, nesting the flow prediction model and the water level prediction model into an improved optimization algorithm by using an objective function with hydrodynamic regulation as a core. S4, according to the multi-target requirements of river network flood control, water resources, water environment, water ecology and the like, the optimal gate opening degree is solved based on an optimization algorithm, and a real-time dynamic water gate group combined regulation and control scheme is generated. According to the multi-target intelligent hydraulic regulation and control method for the tidal river network sluice group, an optimal scheduling method and a machine learning technology are fused, a sluice group joint regulation and control scheme can be generated in real time, the river network water fluidity and the self-cleaning capacity are remarkably enhanced, and important support is provided for tidal river network water problem systematic treatment and intelligent management.

Description

Technical Field

[0001] The invention belongs to the field of hydraulics and river dynamics, and in particular relates to a multi-objective intelligent hydraulic control method for a tidal river network sluice group. Background Art

[0002] The tidal river network area is densely populated, economically developed, and often has an important strategic position. At the same time, the area is subject to basin floods above and high tides below, making drainage difficult and flood disasters frequent; the terrain is low-lying, water resources are insufficient, and the water resource carrying capacity is low; a large amount of agricultural, production, and domestic water discharge leads to a large pollution load in the river network, and water ecological environment problems are prominent. It can be seen that the tidal river network area faces multiple goals such as flood control, water resources, and water ecological environment. At present, a large number of gate and pump projects have been built in the river network. Joint scheduling of these projects is an important means to ensure the water safety of the river network. However, the joint scheduling schemes of the projects are usually massive, and it is very difficult to make quick decisions and obtain the optimal scheme. The traditional river network gate scheduling method relies on numerical simulation software, which forms a rough water gate control operation plan by listing different opening and closing combinations of each gate under characteristic boundary conditions. This decision-making method is time-consuming and labor-intensive, especially in large river network systems, with low calculation efficiency and poor adaptability to real-time changing hydrological parameters. Therefore, it is urgent to build a multi-objective intelligent hydraulic control method for sluice groups in tidal river networks, generate real-time joint control plans for sluice groups, meet the multi-objective needs of tidal river networks, and provide important support for the systematic governance and intelligent management of river network water problems. Summary of the invention

[0003] In view of the problems existing in the prior art, the present invention provides a multi-objective intelligent hydraulic control method for a sluice group in a tidal river network, which is an intelligent control method for a sluice group that can reconstruct the spatiotemporal pattern of the hydrodynamics of the river network.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: a multi-objective intelligent hydraulic control method for a tidal river network sluice group, comprising the following steps:

[0005] S1. Obtain and pre-process the historical detection data and hydrological data of the tidal river network, including gate opening data, tidal water level data, key section flow data, and water level data in the gate;

[0006] S2. Using gate opening data and tidal water level data as input and key section flow data as output, a flow prediction model is constructed and trained; using gate opening data and tidal water level data as input and gate water level data as output, a water level prediction model is constructed and trained;

[0007] S3, construct a multi-objective function based on the water exchange rate and water level to evaluate the gate opening scheme, and add gate opening constraints;

[0008] S4. Use genetic algorithm to solve the most objective functions, specifically: embed the flow prediction model and the water level prediction model into the genetic algorithm, pass the key section flow data obtained by the flow prediction model under the current gate opening scheme to the genetic algorithm, calculate the objective function value, and the genetic algorithm determines whether the water exchange rate is the largest under the current scheme. If so, the optimal river network sluice opening combination is obtained. Otherwise, the water level prediction model predicts the water level inside the gate under the current scheme, and then compares the water level inside the gate with the tidal water level of the outer river to obtain the water level difference based on the tidal water level of the outer river, returns to the optimization algorithm to readjust the gate opening, and then calculates the new key section flow through the flow prediction model, and iterates the calculation until the objective function reaches the maximum value, that is, the optimal river network sluice opening combination is obtained.

[0009] Furthermore, the aforementioned step S1 specifically includes the following sub-steps:

[0010] S1.1. Deploy IoT monitoring nodes at preset locations of the tidal river network, including water level sensors, flow monitors and gate height sensors;

[0011] S1.2. Establish a data transmission network, use wireless communication technology to build a real-time data transmission network, and upload the monitoring data of each device in the monitoring node to the central data platform in real time;

[0012] S1.3. Integrate historical monitoring data, including water level, flow, and gate opening records, establish a database with multiple temporal and spatial scales, collect tidal water level data predicted by hydrological stations, combine historical trends with real-time monitoring, and generate hydrological data corresponding to historical monitoring data;

[0013] S1.4. Build a data management system, use the big data platform to collect, store and visualize monitoring data, update it in real time, clean and label historical and real-time data, and eliminate outliers.

[0014] Furthermore, in the aforementioned step S2, the traffic prediction model is specifically: constructing a traffic prediction model based on a time fusion transformer, the time fusion transformer includes: a gated residual network GRN, a variable selection network, and a multi-layer masked attention mechanism layer; the gated residual network GRN realizes dynamic adjustment of input features through nonlinear transformation; the variable selection network, through the variable selection mechanism, combines the gated residual network to dynamically assign weights to the input features and extract important features; the multi-layer masked attention mechanism layer captures the dependencies between complex time series features, and at the same time enhances the effectiveness of attention through the mask mechanism, and improves the modeling ability of the complex dynamic changes of the tidal river network through multi-layer stacking.

[0015] Furthermore, the aforementioned gated residual network GRN is specifically:

[0016] h1=W1·x+b1

[0017] h2=ReLU(h1)

[0018] g=σ(W g ·h2+b g )

[0019] y=g·h2+(1-g)x

[0020] Among them, h1 is the linear transformation output of the hidden layer, W1 is the weight matrix input to the hidden layer, b1 is the bias vector input to the hidden layer, h2 is the output of the hidden layer after the activation function, g is the output of the gating mechanism, and W g is the weight matrix of the gating mechanism, b g is the bias vector of the gating mechanism; σ and ReLU are activation functions, x is the input feature, and y is the output feature;

[0021] Variable selection network, specifically:

[0022]

[0023]

[0024]

[0025]

[0026] In the formula, z i is the i-th input eigenvalue; It is z i Perform exponential operations to enhance the differences between values; is the original value of the jth feature at time t; It is to flatten multidimensional data into a one-dimensional vector.

[0027] Multi-layer mask attention mechanism layer, specifically:

[0028]

[0029] MultiHead(Q,K,V)=Concat(head1,…,head h )W O

[0030]

[0031] Where Q, K, and V represent query, key, and value vectors, respectively, which are connected by input X and weight matrix W. Q ,W K ,W V Sex transformation generation; d kis the key vector dimension, used to scale the dot product result; M is the mask matrix, used to mask future time steps or invalid positions; W O is the output weight matrix, is the query weight matrix of the ith head, is the key weight matrix of the ith head, The value weight matrix of the i-th head; head i is the i-th attention head, softmax is a normalization operation used to generate attention weights, and Concat is to concatenate the outputs of multiple attention heads into a complete vector.

[0032] Furthermore, the aforementioned water level prediction model is constructed based on the long short-term memory network LSTM, specifically:

[0033] The forget gate determines whether to keep or forget by outputting the value [0,1] of the σ activation function of the formula, as shown below:

[0034] f t =σ(W f ·[h t-1 ,x t +b f ])

[0035] The input gate introduces new information and updates the memory unit as follows:

[0036] i t =σ(W i ·[h t-1 ,x t +b i ])

[0037]

[0038] The current memory unit combines the memory of the previous moment and the current new information as follows:

[0039]

[0040] The output gate determines the output and memory content at the current moment through the formula, as follows:

[0041] o t =σ(W o ·[h t-1 ,x t +b o ])

[0042] h t =o t *tanh(C t )

[0043] Where σ is the activation function; ht-1 is the output at time t-1; x t is the input at time t; C t is the information memory unit at time t, f t is the output of the forget gate, W f is the weight matrix of the forget gate, b f is the bias of the forget gate, i t is the input gate output, W i is the weight matrix of the input gate, b i is the bias of the input gate, is the new memory candidate value, W C is the weight matrix of the new memory candidate value, o t is the output of the output gate, W o is the weight matrix of the output gate, b C is the bias of the new memory candidate value, b o is the bias of the output gate.

[0044] Furthermore, the multi-objective function in the aforementioned step S3 is:

[0045] Z min ≤Z i ≤Z max

[0046] V i =∫Q i

[0047]

[0048] Among them, Z min The lowest water level to meet landscape requirements, in meters; Z max The highest water level to meet flood control needs, unit: m; Q i is the flow rate of the i-th section of the target river, in m 3 / s; V i is the water flow of the i-th section of the target river within a certain period of time, in m 3 ; V0 is the total channel storage capacity of the target river section, unit: m 3 ; γ is the water exchange rate;

[0049] The constraints are:

[0050] H(i)∈[0,x i ],i=1,2,3,…,I

[0051] Where H(i) is the opening constraint of the i-th gate, 0 means the gate is closed, x i Indicates that the gates are fully open, and I indicates the number of gates.

[0052] Furthermore, the aforementioned step S4 includes the following sub-steps:

[0053] S4.1. Determine the gate opening range. According to the gate opening value range corresponding to different gate settings, set the gate opening range as [0, x i ], where x i is the maximum opening of the i-th gate, and the first generation of gate opening quantity is generated within the allowable gate opening range;

[0054] S4.2, let the gate opening measure be N, where N is a natural number between 1 and M, and M is the value of the set algebra, perform non-dominated sorting on the first generation of gate opening measures, and then perform selection, crossover and mutation operations to generate the second generation of gate opening measures;

[0055] S4.3, merging the gate opening measure of the Nth generation with the gate opening measure of the N+1th generation, performing non-dominated sorting on the merged gate opening measures, performing selection, crossover and mutation operations on the new gate opening measures to generate the gate opening measure of the N+2th generation, where the initial value of N is 1;

[0056] S4.4, input the N+2 generation gate opening as new input data into the flow prediction model and water level prediction model, predict the flow process of key sections of the river in the basin, and the water level process in the gate, calculate the river flow and water exchange rate according to the objective function, if the water exchange rate has reached the optimal value, output the corresponding gate opening and end, if not, enter S4.5.

[0057] S4.5, determine whether the current algebra is less than M. If so, set N=N+1 and return to execute S4.3; if not, execute M=2M, that is, adjust the set algebra to 2M, then set N=N+1 and return to execute S4.3.

[0058] Compared with the prior art, the beneficial technical effects of the above technical solutions adopted by the present invention are as follows: First, by integrating the Internet of Things technology and the machine learning model, a real-time channel is built for hydrological and hydrodynamic data and data platforms, and data intelligent management is completed, deep potential is tapped, and real-time applications are applied. Second, the intelligent optimization algorithm is deeply embedded in the multi-objective function with hydrodynamic regulation as the core, breaking through the bottleneck that the traditional scheduling method is difficult to respond to the influence of multivariable coupling in real time; third, facing the reciprocating flow problem and water environment pressure unique to the tidal river network, it is proposed to optimize and reconstruct the hydrodynamic spatiotemporal pattern, enhance the fluidity and self-purification ability of the water body, and provide an intelligent and systematic solution for the governance of water environment problems in the tidal river network. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A diagram of the intelligent dispatching framework of a group of sluice gates reconstructed for spatiotemporal optimization of hydrodynamics in the constructed river network.

[0060] Figure 2The flow chart of using genetic algorithm to solve the optimal system future gate opening plan.

[0061] Figure 3 Build a high-precision hydrological monitoring network map for IoT technology and smart sensing devices.

[0062] Figure 4 This is a schematic diagram of the research area of ​​the Yunnan section of Shanghai’s river network.

[0063] Figure 5 This is a flow prediction and error comparison chart for key sections of the river.

[0064] Figure 6 These are the water level prediction and error comparison diagrams inside the gates. In the figure, (a) is the water level prediction and error comparison diagram inside the Dongjiaojing Gate, (b) is the water level prediction and error comparison diagram inside the Xisitang Gate, and (c) is the water level prediction and error comparison diagram inside the Haoqiaogang Gate.

[0065] Figure 7 Schematic diagram of gate opening changes before and after optimization of group gate intelligent scheduling method. (a) is the gate opening change diagram of Dongjiaojing Gate before and after optimization, (b) is the gate opening change diagram of Xisitang Gate before and after optimization, and (c) is the gate opening change diagram of Haoqiaogang Gate before and after optimization.

[0066] Figure 8 To study the schematic diagram of the regulation results in the time domain of river channel prediction, (a) is the flow process diagram of the target river section before gate optimization, and (b) is the flow process diagram of the target river section after gate optimization. DETAILED DESCRIPTION

[0067] In order to better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0068] Various aspects of the invention are described herein with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the invention are not limited to those described in the accompanying drawings. It should be understood that the invention is implemented by any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed in the invention are not limited to any implementation. In addition, some aspects disclosed in the invention may be used alone or in any appropriate combination with other aspects disclosed in the invention.

[0069] refer to Figure 1 The present invention provides a multi-objective intelligent hydraulic control method for a tidal river network sluice group, comprising the following steps:

[0070] S1. Obtain and pre-process the historical detection data and hydrological data of the tidal river network, including gate opening data, tidal water level data, key section flow data, and water level data in the gate;

[0071] S2. Using gate opening data and tidal water level data as input and key section flow data as output, a flow prediction model is constructed and trained; using gate opening data and tidal water level data as input and gate water level data as output, a water level prediction model is constructed and trained;

[0072] S3, construct a multi-objective function based on the water exchange rate and water level to evaluate the gate opening scheme, and add gate opening constraints;

[0073] S4. Use genetic algorithm to solve the most objective functions, specifically: embed the flow prediction model and the water level prediction model into the genetic algorithm, pass the key section flow data obtained by the flow prediction model under the current gate opening scheme to the genetic algorithm, calculate the objective function value, and the genetic algorithm determines whether the water exchange rate is the largest under the current scheme. If so, the optimal river network sluice opening combination is obtained. Otherwise, the water level prediction model predicts the water level inside the gate under the current scheme, and then compares the water level inside the gate with the tidal water level of the outer river to obtain the water level difference based on the tidal water level of the outer river, returns to the optimization algorithm to readjust the gate opening, and then calculates the new key section flow through the flow prediction model, and iterates the calculation until the objective function reaches the maximum value, that is, the optimal river network sluice opening combination is obtained.

[0074] Furthermore, as a preferred embodiment of the present invention, in step S1, the Internet of Things technology and intelligent sensing equipment are used to obtain historical detection data of the tidal river network, including gate opening data, tidal water level data, and key section flow data. The tidal water level data is provided by the hydrological station forecast and can be used as future input conditions. The specific process is as follows Figure 3 The specific implementation process is as follows:

[0075] like Figure 3As shown in the figure, first, IoT monitoring nodes are deployed at key locations of the tidal river network. These nodes include water level sensors, flow monitors, and gate height sensors, which are used to monitor important parameters such as water level, flow, and gate operation status. Relying on wireless communication technologies such as 4G / 5G, LoRa, or NB-IoT, real-time data transmission of distributed sensing equipment is realized. The hydrological data collected by each monitoring node is uploaded to the central data platform through a wireless network to ensure the stability and reliability of real-time transmission and provide timely data support for scheduling decisions. Subsequently, historical data and hydrological forecast information are integrated to form a complete data foundation. The monitored real-time data are integrated with historical monitoring records to build a multi-temporal and spatial scale database containing information such as water level, flow, and gate opening. At the same time, the tidal water level forecast data provided by the hydrological station is introduced, combined with historical trend analysis and real-time monitoring data, to generate a preliminary forecast of future hydrological conditions, providing an important basis for subsequent analysis and modeling. Finally, a data management system is built to use the big data platform to comprehensively manage the collected hydrological and hydrodynamic data. The system supports real-time dynamic updates and data visualization functions, which are convenient for intuitive analysis and display. At the same time, historical and real-time data are cleaned and annotated, and outliers are removed to ensure the reliability and integrity of data quality. Efficient data storage, processing and analysis are achieved through big data technology, further improving the intelligent level of river network hydrological monitoring and regulation. The input data and model parameter settings used by the prediction model are shown in the data table in Table 1 below.

[0076] Table 1

[0077]

[0078] Further, as a preferred embodiment of the present invention, the flow prediction model in step S2 is constructed based on the Temporal Fusion Transformers (TFT), and the water level prediction model is constructed based on the Long Short-Term Memory Network (LSTM). Both the Temporal Fusion Transformers (TFT) and the Long Short-Term Memory Network (LSTM) are existing models, among which the Temporal Fusion Transformer is a recently proposed prediction model, which has unique advantages in non-stationary sequence prediction and feature capture, and is suitable for complex runoff prediction; the Long Short-Term Memory Network is a classic prediction model, which is suitable for relatively simple water level prediction and saves computing resources. The prediction model in this example is constructed by calling the Temporal Fusion Transformer and the Long Short-Term Memory Network. It should be noted that since the specific calculation and solution process of the relevant modules belongs to the prior art, this embodiment only lists the model construction principle, and does not list the detailed solution process, as follows:

[0079] The gate opening data and tidal water level data are used as input data, and the input data is divided into training set and validation set according to actual needs. In order to facilitate the subsequent solution of practical problems, the length of the validation set is the length of time that the gate needs to be optimized in the future;

[0080] (1) The flow prediction model is specifically: a flow prediction model is constructed based on a time fusion transformer, which includes: a gated residual network (GRN), a variable selection network, and a multi-layer masked attention mechanism layer; the gated residual network (GRN) realizes dynamic adjustment of input features through nonlinear transformation; the variable selection network uses a variable selection mechanism and a gated residual network to dynamically assign weights to input features and extract important features; the multi-layer masked attention mechanism layer captures the dependencies between complex time series features, while enhancing the effectiveness of attention through a mask mechanism, and improving the modeling ability of the model for complex dynamic changes in tidal river networks through multi-layer stacking. Flow prediction model

[0081] Gated residual network GRN, specifically:

[0082] h1=W1·x+b1

[0083] h2=ReLU(h1)

[0084] g=σ(W g ·h2+b g )

[0085] y=g·h2+(1-g)x

[0086] Among them, h1 is the linear transformation output of the hidden layer, W1 is the weight matrix input to the hidden layer, b1 is the bias vector input to the hidden layer, h2 is the output of the hidden layer after the activation function, g is the output of the gating mechanism, and W g is the weight matrix of the gating mechanism, b g is the bias vector of the gating mechanism; σ and ReLU are activation functions, x is the input feature, and y is the output feature;

[0087] Variable selection network, specifically:

[0088]

[0089]

[0090]

[0091]

[0092] In the formula, z i is the i-th input eigenvalue; It is z iPerform exponential operations to enhance the differences between values; is the original value of the jth feature at time t; It is to flatten multidimensional data into a one-dimensional vector.

[0093] Multi-layer mask attention mechanism layer, specifically:

[0094]

[0095] MultiHead(Q,K,V)=Concat(head1,…,head h )W O

[0096]

[0097] Where Q, K, and V represent query, key, and value vectors, respectively, which are connected by input X and weight matrix W. Q ,W K ,W V Sex transformation generation; d k is the key vector dimension, used to scale the dot product result; M is the mask matrix, used to mask future time steps or invalid positions; W O is the output weight matrix, is the query weight matrix of the ith head, is the key weight matrix of the ith head, The value weight matrix of the i-th head; head i is the i-th attention head, softmax is a normalization operation used to generate attention weights, and Concat is to concatenate the outputs of multiple attention heads into a complete vector. The main parameters of the model are set as shown in Table 2:

[0098] Table 2

[0099]

[0100] (2) A water level prediction model is constructed based on the long short-term memory network LSTM, specifically:

[0101] The forget gate determines whether to keep or forget by outputting the value [0,1] of the σ activation function of the formula, as shown below:

[0102] f t =σ(W f ·[h t-1 ,x t +b f ])

[0103] The input gate introduces new information and updates the memory unit as follows:

[0104] it =σ(W i ·[h t-1 ,x t +b i ])

[0105]

[0106] The current memory unit combines the memory of the previous moment and the current new information as follows:

[0107]

[0108] The output gate determines the output and memory content at the current moment through the formula, as follows:

[0109] o t =σ(W o ·[h t-1 ,x t +b o ])

[0110] h t =o t *tanh(C t )

[0111] Where σ is the activation function; h t-1 is the output at time t-1; x t is the input at time t; C t is the information memory unit at time t. t is the output of the forget gate, W f is the weight matrix of the forget gate, b f is the bias of the forget gate, i t is the input gate output, W i is the weight matrix of the input gate, b i is the bias of the input gate, is the new memory candidate value, W C is the weight matrix of the new memory candidate value, o t is the output of the output gate, W o is the weight matrix of the output gate, b C is the bias of the new memory candidate value, b o is the bias of the output gate. The main parameter settings of the LSTM model are shown in Table 3 below:

[0112] Table 3

[0113]

[0114] Further, as a preferred embodiment of the present invention, in S3, the primary goal is to improve the hydrodynamics of the river network, and the objective function constructed by evaluating the gate opening scheme with water exchange rate and water level parameters is:

[0115] Z min ≤Z i ≤Z max

[0116] V i =∫Q i

[0117]

[0118] In the formula, Z min The minimum water level to meet landscape requirements, m; Z max The highest water level to meet flood control needs, m; Q i is the flow rate of the i-th section of the target river, m 3 / s; V i is the water flow of the i-th section of the target river within a certain period of time, m 3 ; V0 is the total channel storage capacity of the target river section, m 3 ; γ is the water exchange rate;

[0119] Constraints:

[0120] H(i)∈[0,x i ],i=1,2,3,…,I

[0121] Where H(i) is the opening constraint of the i-th gate, 0 means the gate is closed, and x i Indicates that the gate is fully open. Since each gate is dispersed in space and the riverbed elevation is different, the maximum opening height of each gate is not exactly the same and needs to be adapted to local conditions. I represents the number of gates.

[0122] It should be noted that in order to construct the objective function, it is necessary to obtain the future river section flow, and an accurate runoff prediction model is needed to ensure the accuracy of the objective function. Therefore, the time fusion transformer is used to predict the river section flow; the river water level outside the gate can be obtained through the tide forecasting station, while the river network water level inside the gate cannot be known in advance, and it is difficult to open and close the gate by the water level difference. Therefore, the long short-term memory network is used to predict the water level inside the gate, and then compared with the water level outside the gate to obtain the water level difference to control the gate opening and closing time.

[0123] Further, as a preferred embodiment of the present invention, step S4 is implemented by using a genetic algorithm (GA), such as Figure 2 As shown in the flow chart, the steps of using genetic algorithm to solve the optimal future gate opening scheme of the river network are as follows:

[0124] S4.1, determine the gate opening range, according to the gate opening value range corresponding to different gate settings, set the gate opening range as [0, x i ], where x iis the maximum opening of the i-th gate, and the first generation of gate opening quantity is generated within the allowable gate opening range;

[0125] S4.2, let the gate opening measure be N, where N is a natural number between 1 and M, and M is the value of the set algebra, perform non-dominated sorting on the first generation of gate opening measures, and then perform selection, crossover and mutation operations to generate the second generation of gate opening measures;

[0126] S4.3, merging the gate opening measure of the Nth generation with the gate opening measure of the N+1th generation, performing non-dominated sorting on the merged gate opening measures, performing selection, crossover and mutation operations on the new gate opening measures to generate the gate opening measure of the N+2th generation, where the initial value of N is 1;

[0127] S4.4, input the N+2 generation gate opening as new input data into the flow prediction model and water level prediction model, predict the flow process of the key section of the river in the basin, the water level process in the gate, calculate the river water exchange rate according to the objective function, if the river water exchange rate has reached the optimal value and cannot be improved, then output the corresponding gate opening and end, if not, enter S4.5.

[0128] S4.5, determine whether the current generation is less than the set generation M. If so, set N=N+1 and return to execute S43; if not, execute M=2M, that is, adjust the set generation to 2M, then set N=N+1 and return to execute S43.

[0129] It should be noted that when a person skilled in the art knows that a genetic algorithm is used to solve the optimal gate control quantity, the specific operation methods of non-dominated sorting, selection, crossover, mutation and fitness calculation should be known, which will not be described in detail here.

[0130] The method of the present invention is further described below by taking the Shanghai River Network, the Southern Section, as an example. The total area of ​​the Southern Section is 173.37 km 2 There are 69 rivers in the area with a total length of 159.51 km. The water system in this area is relatively complex. The main rivers have three longitudinal river sections (Nansitang-Shajinggang, Xisitang-Yujingpu, Dongjiaojing-Pengyuepu) and one transverse river section (Zoumatang). The main drainage gates and water diversion gates are as follows: Figure 4As shown in the annotation. In this embodiment, the input data of the tidal river network prediction model uses the main diversion and drainage gate opening data, tidal water level data and key section flow data of the Zoumatang River section from January 1 to June 22, 2022. Among them, the data from 0:00 on January 1 to 0:00 on June 21, 2022 is used for prediction model training, and the data from 0:00 on June 21 to 12:00 on June 21 is used for prediction model verification. The error of the water level prediction results of the verification data set is generally small, with the maximum error controlled within 20cm, 65.28% of the errors controlled within 5cm, and 95.83% of the errors controlled within 10cm. The flow prediction determination coefficient R 2 The model has good prediction performance, and has shown a high level of prediction for the future flow trend and peak and valley values, which can be used for river network regulation calculation and analysis. Figure 5 and Figure 6 As shown, Figure 6 (a) is the water level prediction and error comparison diagram inside Dongjiaojing Gate, (b) is the water level prediction and error comparison diagram inside Xisitang Gate, and (c) is the water level prediction and error comparison diagram inside Haoqiaogang Gate.

[0131] This embodiment considers the sunny day scenario, that is, the flow and water level of the urban river network are predicted during the rainless period, and the opening and closing process of the river network gate is determined. The prediction time domain is determined to be 12 hours, and the prediction model is used to predict the water level process in the gate and the flow process of the key section of the river in the prediction time domain.

[0132] In the constructed multi-objective intelligent hydraulic control method for tidal river network sluice groups, the genetic algorithm, flow prediction model and water level prediction model are coupled into a system model; the gate opening data and the water level data outside the gate at the current moment and the past moment are the system input values, and the predicted river network gate opening and the water level inside the gate in the time domain are the future output values ​​of the system. Through the constructed objective function and constraints, the genetic algorithm is used to solve the predicted gate opening in the time domain, in which the value of the algebra M is set to 100. In this embodiment, the time step of the gate pump control is 0.5h. Of course, it can be adjusted to 1h, 1.5h, etc. according to the requirements of accuracy. By solving, the predicted gate opening in the time domain can be obtained as follows Figure 7 As shown in the figure, (a) is the gate opening change diagram of Dongjiaojing Gate before and after optimization, (b) is the gate opening change diagram of Xisitang Gate before and after optimization, and (c) is the gate opening change diagram of Haoqiao Port Gate before and after optimization. Comparison of flow process before and after optimization of key sections of the river Figure 8 As shown in the figure, (a) is the flow process diagram of the target river section before gate optimization, and (b) is the flow process diagram of the target river section after gate optimization. The selected target river section has serious reciprocating flow phenomenon from 15:30 on June 21, 2022 to 3:00 on June 22, 2022, with a positive flow of 5893.2m 3 , reverse flow 25687.8m 3, total flow rate 19794.6m 3 After the gate opening was optimized by the group gate intelligent scheduling model, the forward flow of the target river section was reduced to 753.08m 3 , a decrease of 87.22%, and reverse flow decreased to 24262.4m 3 , the decrease is 5.55%, but the total flow rate is 23509.3m 3 , an increase of 18.77% compared to before gate optimization. By optimizing the gate with intelligent models, the water exchange rate increased from 16.1% to 19.1%, and the water exchange time decreased from 6.2 days to 5.2 days, providing technical support for water environment problems such as sewage and waterlogging that require rapid water exchange.

[0133] Although the present invention has been described above with preferred embodiments, it is not intended to limit the present invention. A person skilled in the art of the present invention may make various modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the definition of the claims.

Claims

1. A multi-objective intelligent hydraulic control method for a tidal river network sluice group, characterized in that: The following steps are involved: S1. Obtain and pre-process the historical detection data and hydrological data of the tidal river network, including gate opening data, tidal water level data, key section flow data, and water level data in the gate; S2. Using gate opening data and tidal water level data as input and key section flow data as output, a flow prediction model is constructed and trained; using gate opening data and tidal water level data as input and gate water level data as output, a water level prediction model is constructed and trained; S3, construct a multi-objective function based on the water exchange rate and water level to evaluate the gate opening scheme, and add gate opening constraints; S4. Use genetic algorithm to solve the most objective functions, specifically: embed the flow prediction model and the water level prediction model into the genetic algorithm, pass the key section flow data obtained by the flow prediction model under the current gate opening scheme to the genetic algorithm, calculate the objective function value, and the genetic algorithm determines whether the water exchange rate is the largest under the current scheme. If so, the optimal river network sluice opening combination is obtained. Otherwise, the water level prediction model predicts the water level inside the gate under the current scheme, and then compares the water level inside the gate with the tidal water level of the outer river to obtain the water level difference based on the tidal water level of the outer river, returns to the optimization algorithm to readjust the gate opening, and then calculates the new key section flow through the flow prediction model, and iterates the calculation until the objective function reaches the maximum value, that is, the optimal river network sluice opening combination is obtained.

2. The multi-objective intelligent hydraulic control method for a tidal river network sluice group according to claim 1 is characterized in that: Step S1 specifically includes the following sub-steps: S1.

1. Deploy IoT monitoring nodes at preset locations of the tidal river network, including water level sensors, flow monitors and gate height sensors; S1.

2. Establish a data transmission network, use wireless communication technology to build a real-time data transmission network, and upload the monitoring data of each device in the monitoring node to the central data platform in real time; S1.

3. Integrate historical monitoring data, including water level, flow, and gate opening records, establish a database with multiple time and space scales, collect tidal water level data predicted by hydrological stations, combine historical trends with real-time monitoring, and generate hydrological data corresponding to historical monitoring data; S1.

4. Build a data management system and use the big data platform to collect, store and visualize monitoring data, update it in real time, clean and label historical and real-time data, and remove outliers.

3. The multi-objective intelligent hydraulic control method for a tidal river network sluice group according to claim 1 is characterized in that: In step S2, the flow prediction model is specifically: building a flow prediction model based on a time fusion transformer, the time fusion transformer includes: a gated residual network GRN, a variable selection network, and a multi-layer masked attention mechanism layer; the gated residual network GRN realizes dynamic adjustment of input features through nonlinear transformation; the variable selection network, through the variable selection mechanism, combines the gated residual network to dynamically assign weights to the input features and extract important features; the multi-layer masked attention mechanism layer captures the dependencies between complex time series features, and at the same time enhances the effectiveness of attention through the mask mechanism, and improves the modeling ability of the model for the complex dynamic changes of tidal river networks through multi-layer stacking.

4. The multi-objective intelligent hydraulic control method for a tidal river network sluice group according to claim 3 is characterized in that: Gated residual network GRN, specifically: h1=W1·x+b1 h2=ReLU(h1) g=σ(W g ·h2+b g ) y=g·h2+(1-g)x Among them, h1 is the linear transformation output of the hidden layer, W1 is the weight matrix input to the hidden layer, b1 is the bias vector input to the hidden layer, h2 is the output of the hidden layer after the activation function, g is the output of the gating mechanism, and W g is the weight matrix of the gating mechanism, b g is the bias vector of the gating mechanism; σ and ReLU are activation functions, x is the input feature, and y is the output feature; Variable selection network, specifically: In the formula, z i is the i-th input eigenvalue; It is z i Perform exponential operations to enhance the differences between values; is the original value of the jth feature at time t; It is to flatten multidimensional data into a one-dimensional vector. Multi-layer mask attention mechanism layer, specifically: MultiHead(Q,K,V)=Concat(head1,...,head h )W O Where Q, K, and V represent query, key, and value vectors, respectively, which are connected by input X and weight matrix W. Q , W K , W V Sex transformation generation; d k is the key vector dimension, used to scale the dot product result; M is the mask matrix, used to mask future time steps or invalid positions; W O is the output weight matrix, is the query weight matrix of the ith head, is the key weight matrix of the ith head, The value weight matrix of the i-th head; head i is the i-th attention head, softmax is a normalization operation used to generate attention weights, and Concat is to concatenate the outputs of multiple attention heads into a complete vector.

5. The multi-objective intelligent hydraulic control method for a tidal river network sluice group according to claim 1 is characterized in that: In step S2, a water level prediction model is constructed based on the long short-term memory network LSTM, specifically: The forget gate determines whether to keep or forget by outputting the value [0, 1] of the σ activation function of the formula, as shown below: f t =σ(W f ·[h t-1 ,x t +b f ]) The input gate introduces new information and updates the memory unit as follows: i t =σ(W i ·[h t-1 ,x t +b i ]) The current memory unit combines the memory of the previous moment and the current new information as follows: The output gate determines the output and memory content at the current moment through the formula, as follows: the t =σ(W o ·[h t-1 ,x t +b o ]) h t =o t *tanh(C t ) Where σ is the activation function; h t-1 is the output at time t-1; x t is the input at time t; C t is the information memory unit at time t, f t is the output of the forget gate, W f is the weight matrix of the forget gate, b f is the bias of the forget gate, i t is the input gate output, W i is the weight matrix of the input gate, b i is the bias of the input gate, is the new memory candidate value, W C is the weight matrix of the new memory candidate value, o t is the output of the output gate, W o is the weight matrix of the output gate, b C is the bias of the new memory candidate value, b o is the bias of the output gate.

6. The multi-objective intelligent hydraulic control method for a tidal river network sluice group according to claim 1 is characterized in that: The multi-objective function in step S3 is: WITH min ≤Z i ≤Z max In i =∫Q i Among them, Z min The lowest water level to meet landscape requirements, in meters; Z max The highest water level to meet flood control needs, unit: m; Q i is the flow rate of the i-th section of the target river, in m 3 / s; V i is the water flow of the i-th section of the target river within a certain period of time, in m 3 ; V0 is the total channel storage capacity of the target river section, unit: m 3 ; γ is the water exchange rate; The constraints are: H(i)∈[0,x i ],i=1,2,3,...,I Where H(i) is the opening constraint of the i-th gate, 0 means the gate is closed, x i Indicates that the gates are fully open, and I indicates the number of gates.

7. The multi-objective intelligent hydraulic control method for a tidal river network sluice group according to claim 1 is characterized in that: Step S4 includes the following sub-steps: S4.

1. Determine the gate opening range. According to the gate opening value range corresponding to different gate settings, set the gate opening range as [0, x i ], where x i is the maximum opening of the i-th gate, and the first generation of gate opening quantity is generated within the allowable gate opening range; S4.2, let the gate opening measure be N, where N is a natural number between 1 and M, and M is the value of the set algebra, perform non-dominated sorting on the first generation of gate opening measures, and then perform selection, crossover and mutation operations to generate the second generation of gate opening measures; S4.3, merging the gate opening measure of the Nth generation with the gate opening measure of the N+1th generation, performing non-dominated sorting on the merged gate opening measures, performing selection, crossover and mutation operations on the new gate opening measures to generate the gate opening measure of the N+2th generation, where the initial value of N is 1; S4.4, input the N+2 generation gate opening as new input data into the flow prediction model and water level prediction model, predict the flow process of key sections of the river in the basin, and the water level process in the gate, calculate the river flow and water exchange rate according to the objective function, if the water exchange rate has reached the optimal value, output the corresponding gate opening and end, if not, enter S4.

5. S4.5, determine whether the current algebra is less than M. If so, set N=N+1 and return to execute S4.3; if not, execute M=2M, that is, adjust the set algebra to 2M, then set N=N+1 and return to execute S4.3.

Images