Regional water supply emergency scheduling method and system

Abstract

The application discloses a regional water supply emergency scheduling method and system, and the method comprises the following steps: fusing GIS and IoT data to construct a dynamic three-dimensional topological model; wavelet multi-scale decomposition is used to decompose historical water consumption and meteorological data, and a prediction model is used to obtain a water consumption prediction value of a scheduling day period; based on real-time collected reservoir storage capacity, flow, water quality, precipitation and pollution source data, a flow-water quality coupling model is used to predict the reservoir inflow and water quality of the scheduling day; a multi-objective function is established by combining water consumption prediction, inflow prediction and current storage capacity, a multi-objective optimization model with water quality constraint is embedded, and a NSGA-II algorithm is used to solve and obtain a Pareto optimal scheduling scheme set; finally, a scheme is selected and issued for execution. The method realizes intelligent scheduling decision of data-physical cooperation, meets the water supply demand under the condition of ensuring the stability of water supply water quality, avoids the shortage of water consumption at the user end, and guarantees that the water quality at the user end can continuously meet the standard.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water resources scheduling, and particularly relates to a regional water supply emergency scheduling method and system. BACKGROUND

[0002] The safe and reliable operation of the urban water supply system is directly related to the normal operation of the residents' life and the urban economy. In recent years, affected by the frequent occurrence of extreme weather, the expansion of the city scale and the increase of the user water peak-valley difference, emergency scheduling mechanisms have been established in various places to ensure continuous water supply in the event of accidents or water shortage.

[0003] The current water supply project in the receiving area takes the reservoir group as the core. The reservoir not only bears the daily water supply task, but also bears the emergency water supply task in the event of an emergency. The traditional emergency water supply scheduling strategy mainly relies on experience to formulate, and it is difficult to ensure that the water supply quality is safe while meeting the water supply demand. SUMMARY

[0004] Based on the above, the purpose of the present application is to provide a regional water supply emergency scheduling method and system. The scheduling method can meet the water supply demand while ensuring the stability of the water quality, avoid water shortage at the user end, and ensure that the water quality at the user end can continue to meet the standards.

[0005] In order to achieve the above technical purpose, the technical scheme adopted by the present application is:

[0006] In a first aspect, the present application provides a regional water supply emergency scheduling method, comprising:

[0007] 1) Based on geographic information system GIS and Internet of Things IoT data, a dynamic three-dimensional topological model of the water supply system is constructed, and the topological model represents the spatial position and real-time running state of the water source node, the water distribution network, the water plant, the user terminal and the water quality monitoring node;

[0008] 2) The water consumption, meteorological information of the target area are obtained from the water supply system historical database, and wavelet multi-scale decomposition is performed to obtain high-frequency components and low-frequency components as input features for water consumption prediction;

[0009] 3) The input features of step 2) are sent to the water consumption prediction model, and the water consumption prediction value of each time period of the target area scheduling day is output;

[0010] 4) Real-time acquisition of reservoir storage capacity, flow, water quality, precipitation and pollution source monitoring data, and input of the reservoir storage capacity, flow, water quality, precipitation and pollution source monitoring data into the reservoir inflow-flow quality coupling prediction model to obtain the inflow prediction value and water quality prediction value of each time period of the scheduling day;

[0011] 5) According to the prediction results of steps 3) and 4) and the current water storage of the reservoir, a multi-objective optimization scheduling model is established, with the maximum water supply reliability, the minimum energy consumption and the maximum water quality standard compliance rate as the target, and the water quality constraint as the limiting condition, and the non-dominated sorting genetic algorithm NSGA-II is used to solve, to obtain a set of Pareto optimal scheduling schemes;

[0012] 6) The final scheduling scheme is selected from the set of Pareto optimal schemes and issued to the water supply system execution unit for execution.

[0013] In some embodiments, step 1) specifically includes:

[0014] The three-dimensional coordinates and elevations of the water source nodes, water plants, pipe network nodes, user terminals and monitoring nodes in the water supply system are obtained from the geographic information system GIS; the real-time operation data of each node, including flow , pressure , meteorological data vector and water quality index vector , are obtained from the Internet of Things IoT sensor, and then the GIS spatial data and IoT real-time data are filtered, time-synchronized and coordinate-interpolated;

[0015] A dynamic three-dimensional topological graph is generated according to the pipe segment connection relationship with the water source nodes, water supply and distribution pipe network, water plants, user terminals and water quality monitoring nodes as vertices ;

[0016] Wherein, represents the node set of the water supply system, each node has a spatial position and a state vector , represents the node type; the edge set , represents the pipe from node to , only when there is a pipe connection between the two nodes ;

[0017] The topological model is dynamically adjusted through the actual collected data of each node and pipe.

[0018] In some embodiments, step 2) specifically includes:

[0019] The multi-source time series sequence of the target area is obtained from the historical database, aligned and missing value interpolated according to the time step to obtain the following original time series sequence:

[0020] water consumption sequence , meteorological information vector sequence ; wherein, This represents the k-th time step. k=1,..., K ;

[0021] For each sequence Perform zero-mean and variance normalization to obtain the normalized sequence. ;

[0022] The Discrete Wavelet Transform (DWT) is used for each normalized sequence. conduct J Decomposition of level 1 yields the first level 2. J Approximation coefficient sequence and the j Level detail coefficient sequence ;

[0023] From approximation coefficients Extracting low-frequency statistical features and and for each level of detail coefficient Calculate high frequency energy Entropy ;in, , High-frequency energy The expression is: ; The expression is:

[0024] ;

[0025] In the above formula, No. j Level wavelet detail coefficient sequence The first in Each coefficient value; Indicates the first j Level detail coefficient sequence Length; Indicates an incremental index. Used to traverse all the first... j Grade coefficient; Indicates the first j The proportion of the k-th coefficient in the total energy of that level;

[0026] Each sequence Concatenate according to source type to form multi-source, multi-scale feature vectors. Principal component analysis was used to... Dimensionality reduction is performed to obtain the final input features. .

[0027] In some embodiments, the prediction model consists of a parallel CNN-LSTM and a CNN-GRU, and the model parameters are updated online using a policy gradient algorithm; step 3) specifically includes:

[0028] The dimension-reduced feature vectors of the previous L time points are assembled into a matrix in the form of a sliding window with length L and dimension P As follows:

[0029]

[0030] are input into two parallel branches of CNN-LSTM and CNN-GRU at the same time;

[0031] In the CNN-LSTM branch, the input is first processed by a one-dimensional convolutional layer to extract features, and ReLU activation is applied on the convolutional output; then the result is sent to an LSTM network with hidden dimension D , and the hidden state at the last time is taken as the output feature vector of the branch.

[0032] In the CNN-GRU branch, another set of one-dimensional convolution and ReLU activation is used to extract features, and then the result is input into a GRU network with hidden dimension D' , and the hidden state at the final time is also taken as the branch output .

[0033] The output features of the two branches are concatenated into a vector , which is mapped to a prediction vector of length H through a fully connected layer, as follows:

[0034]

[0035] where represents the weight matrix of the fully connected layer, represents the bias vector of the fully connected layer.

[0036] The prediction vector is the predicted value of the water consumption of the target area for the next H time periods.

[0037] The prediction model takes the negative square norm of the prediction error as the immediate reward, and uses the policy gradient method to update all trainable parameters θ online; where

[0038]

[0039] where represents the true observation vector, represents the predicted value vector.

[0040] ​​​In some embodiments, step 4) specifically comprises:

[0041] Real-time acquisition of reservoir storage capacity, reservoir inflow observation value, reservoir outflow, water quality index vector, precipitation and upstream pollution source monitoring data vector set, and obtaining reservoir evaporation loss data form multi-source data, and the multi-source data is synchronized according to a unified time step Time synchronization interpolation, filtering, linear interpolation are performed to obtain 、 、 、 、 、 、 ; Reservoir storage capacity at , Reservoir inflow at , Reservoir outflow at , Water quality index vector at , Precipitation at , Upstream pollution source monitoring data vector set at , Evaporation loss of the reservoir at ;

[0042] The simplified SWAT hydrology-water quality model is used to preliminarily predict the flow and water quality, including:

[0043] The reservoir inflow prediction value is calculated according to the reservoir storage capacity change, evaporation loss and reservoir inflow , The expression is as follows:

[0044]

[0045] In the formula, , which represents the change of the reservoir storage volume, , represents the water area in the reservoir;

[0046] The upstream pollution source monitoring data vector is a set of all upstream monitoring points flowing into the target reservoir at the corresponding time positioned from the dynamic three-dimensional topological model The measured pollutant concentration vector; wherein, , , which represents the pollutant concentration vector of the i th monitoring point at ; representing the type of pollutant;

[0047] The flow rate , upstream inflow concentration , water storage volume at the end of the step and the pollutant decay coefficient , the reservoir water quality is estimated according to the discretized transport-decay equation , The expression of

[0048]

[0049] wherein, The expression of

[0050]

[0051] wherein, represents the inflow of the i th monitoring point at ;

[0052] A BP neural network with adaptive learning rate and momentum factor is constructed with as input to predict the flow correction amount and the water quality correction amount of the future H time periods .

[0053] In some embodiments, the BP neural network introduces adaptive learning rate and momentum factor in the back propagation process to improve the convergence speed; wherein the adaptive learning rate mechanism satisfies:

[0054]

[0055] wherein, is the mean square error of the current round, is the mean square error of the previous round; is the current learning rate, is the learning rate used in the next iteration; is the amplification factor, ; is the reduction factor, ;

[0056]

[0057] wherein, and represent the predicted values of the current inflow and water quality respectively, and These represent the actual observed values ​​of the current inflow and water quality, respectively.

[0058] The BP neural network introduces a momentum term to correct the weight update, as shown in the following expression:

[0059]

[0060]

[0061] in, This represents the neural network weight parameters at the current time t. Indicates the next moment The weights; This represents the weight update amount at the current time t; This represents the learning rate at the current moment; Indicates the momentum factor; Indicates the mean square error at the current time. Weights The gradient.

[0062] In some embodiments, step 5) specifically includes:

[0063] The scheduling day is discretized into H time periods, denoted as t=1~H; decision variables are set for each time period t. ;

[0064] in, Indicates the first i The outflow of water from each reservoir This represents the start / stop status of the j-th pump, where 1 indicates on and 0 indicates off. The total number of adjustable reservoir water sources; This represents the total number of controlled pump units.

[0065] Using the target area daily water consumption forecast value obtained in step 3) Based on the inflow rate and water quality coupling prediction results obtained in step 4), the following objectives and constraints are established;

[0066] The objectives include:

[0067] A) Maximizing water supply reliability: ;

[0068] in, This represents the predicted total water demand for the target area during the corresponding time period t.

[0069] B) Minimize energy consumption: ;

[0070] in, Where is the density of water; g is the acceleration due to gravity; This represents the head of the j-th pump during time period t; Let be the operating efficiency of the j-th pump; This represents the water flow rate of the j-th pump at time t. , The water delivery volume corresponding to the water source node p is determined based on the flow rate of water from each adjustable reservoir flowing through node p. This indicates the number of pumps at node p;

[0071] C) Maximize the rate of effluent quality compliance: ;

[0072] in, Indicates the water quality assessment threshold; This is an indicator function; it takes the value 1 if the condition is met, and 0 otherwise. The predicted water quality value of the mixed water source at time t is expressed as follows:

[0073]

[0074] In the formula, Let be the water quality prediction vector for the i-th reservoir at time t;

[0075] The constraints include:

[0076] a) Water volume is flat: ;

[0077] b) Water supply: ;in, For the first i Prediction of available water flow from reservoirs; , t=1~H; where This represents the loss coefficient during the transportation process; For the first i The reservoir is there Forecast of inbound flow at the time; For the first i Current water storage capacity of each reservoir;

[0078] c) Pump flow rate: ;in, This represents the maximum delivery flow rate of pump j;

[0079] d) Hard constraints on water quality: ;in, This indicates the hard constraint threshold for water quality;

[0080] by The temporal combinations are represented by chromosomes, and the non-dominated sorting genetic algorithm NSGA-II is used to solve the problem, obtaining a set of Pareto optimal scheduling schemes. .

[0081] In some embodiments, in step 6), the final scheduling scheme is selected from the Pareto optimal scheme set according to water supply reliability, energy consumption and water quality standard preference, and is issued to the water supply system execution unit for execution, and the scheduling instruction and its execution result are notarized by using blockchain technology.

[0082] In a second aspect, the present application provides a regional water supply emergency scheduling system which applies the above-mentioned regional water supply emergency scheduling method, and the system comprises:

[0083] A data acquisition unit is configured to acquire geographic information system (GIS) data and Internet of Things (IoT) sensor data, and to collect spatial coordinates and real-time operating states of water source nodes, water transmission and distribution pipe networks, water plants, user terminals and water quality monitoring nodes.

[0084] A topology construction unit is configured to construct a dynamic three-dimensional topology model of the water supply system according to the data collected by the data acquisition unit.

[0085] A multi-scale feature extraction unit is configured to acquire water consumption sequences and meteorological information of a target region from a water supply system historical database, and to generate corresponding high-frequency components and low-frequency components by using a wavelet multi-scale decomposition method, so as to serve as input features for water consumption prediction.

[0086] A water consumption prediction unit is configured to predict water consumption of each time period of a scheduling day in the target region according to the features output by the multi-scale feature extraction unit.

[0087] A reservoir multi-source data acquisition unit is configured to acquire reservoir storage capacity, flow, water quality, precipitation and pollution source monitoring data in real time, so as to form multi-source data.

[0088] A inflow-flow and water quality coupling prediction unit is configured to predict inflow and water quality of each time period of a scheduling day according to the multi-source data output by the reservoir multi-source data acquisition unit.

[0089] A multi-objective optimization scheduling unit is configured to establish a multi-objective optimization model with the maximum water supply reliability, the minimum energy consumption and the maximum water quality standard compliance rate as the target, and with water quality constraints as the limiting condition, according to the prediction results of water consumption, inflow and water quality, and the current storage capacity of the reservoir, and to call a non-dominated sorting genetic algorithm II (NSGA-II) to obtain a Pareto optimal scheduling scheme set.

[0090] An instruction issuing unit is configured to select a final scheduling scheme from the Pareto optimal scheduling scheme set, and to issue the selected scheduling instruction to a water supply system execution unit.

[0091] The water supply system execution unit is configured to implement corresponding flow regulation and pump start-stop related instructions according to the selected scheduling scheme. ​

[0092] In a third aspect, the present application also provides a computer-readable storage medium, wherein at least one instruction, at least one program, a code set or an instruction set is stored in the storage medium, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to realize the regional water supply emergency scheduling method.

[0093] Compared with the prior art, the present application has the beneficial effects that:

[0094] The present application realizes the intelligent transition of scheduling decision from experience type to data-physical coordination and global trade-off. The method enables the scheduling scheme to meet the instantaneous water demand while automatically avoiding the risk of water quality exceeding the standard and significantly reducing the power consumption of the pumping station; the Pareto frontier provides diversified alternative schemes for the decision maker, which can be flexibly switched between "energy saving", "stable supply" and "high quality" according to the actual preference, ensures that there is still a feasible solution in the sudden scenario, and comprehensively improves the safety, economy and emergency resilience of the regional water supply system. BRIEF DESCRIPTION OF DRAWINGS

[0095] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0096] Figure 1 is a brief implementation flowchart of the regional water supply emergency scheduling method of the present application;

[0097] Figure 2 is a system implementation structure diagram of the regional water supply emergency scheduling method corresponding to the present application scheme. DETAILED DESCRIPTION

[0098] The present application will be further described in detail below with reference to the drawings and embodiments. It is particularly pointed out that the following embodiments are only used to illustrate the present application, but do not limit the scope of the present application. Similarly, the following embodiments are only part of the embodiments of the present application, not all embodiments, and all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0099] Referring to the drawings, Figure 1 The present embodiment provides a regional water supply emergency scheduling method, which comprises:

[0100] 1) Based on Geographic Information System (GIS) and Internet of Things (IoT) data, construct a dynamic three-dimensional topology model of the water supply system. The topology model represents the spatial location and real-time operating status of water source nodes, water transmission and distribution networks, water plants, user terminals, and water quality monitoring nodes; specifically, it includes the following steps:

[0101] The three-dimensional coordinates and elevations of water source nodes, water plants, pipeline nodes, user terminals, and monitoring nodes in the water supply system are obtained from a Geographic Information System (GIS); real-time operational data of each node, including flow rate, is obtained from Internet of Things (IoT) sensors. ,pressure Meteorological data vector and water quality index vector The water quality index vector includes concentration data of various pollutants such as ammonia nitrogen, total nitrogen, total phosphorus, and heavy metals in the water body, and the meteorological data vector... This includes data such as temperature, precipitation, and humidity; then, the GIS spatial data and IoT real-time data are filtered, synchronized in time, and interpolated.

[0102] A dynamic three-dimensional topology map is generated based on the connections between water source nodes, water transmission and distribution networks, water plants, user terminals, and water quality monitoring nodes. ;

[0103] in, This represents the set of nodes in a water supply system, where each node... Having spatial location and state vector , Represents node type (e.g., 0-water source, 1-water plant, 2-user, 3-monitoring point, etc.); edge set , Indicates from node arrive A pipe is defined as follows: a pipe is defined as a connection between two nodes if and only if a pipe connection exists between them. The topology model is dynamically adjusted based on the actual data collected from each node and pipeline.

[0104] 2) Obtain water consumption and meteorological information for the target area from the historical database of the water supply system, perform wavelet multi-scale decomposition, and obtain high-frequency and low-frequency components as input features for water consumption prediction; specifically, the following steps are included:

[0105] Multi-source time series sequences of the target region are obtained from historical databases and uniformly sorted by time step. After alignment and missing value interpolation (specifically, linear interpolation is used in this embodiment), the following original time series sequence is obtained:

[0106] Water consumption sequence Meteorological information vector sequence ;in, This represents the k-th time step. k=1,..., K ;

[0107] For each sequence Perform zero-mean and variance normalization to obtain the normalized sequence. The water consumption is obtained through this step. Meteorological information vector Normalized time series data. Specifically, ;in, and The expression is as follows:

[0108]

[0109] This represents the original time-series signal (representing any of the above indicators). This represents the normalized time series signal; by using zero-mean and variance normalization, the original sequences from different sources and with different dimensions are brought into the same numerical range, while the amplification effect of large outliers in subsequent feature calculations is weakened, thus laying a clean and comparable numerical foundation for multi-source fusion.

[0110] The Discrete Wavelet Transform (DWT) (preferably Daubechies4) is used to process each normalized sequence. conduct J Level decomposition ( J (This refers to the number of decomposition layers, typically 2-4 layers), resulting in the [number of layers]. J Approximate (low-frequency) coefficient sequence and the j Level detail (high frequency) coefficient sequence ;

[0111] Wavelet decomposition breaks down each time series into low-frequency approximations (long-term trends) and several high-frequency details (short-term fluctuations), avoiding the information loss caused by mutual masking in conventional moving averages. This allows the model to capture both slow-changing patterns and rapid disturbances such as sudden increases in water consumption and instantaneous rises in pollutants.

[0112] From approximation coefficients Extracting low-frequency statistical features and and for each level of detail coefficient Calculate high frequency energy Entropy ;in, , ; This indicates calculating the mean. Indicates the standard deviation; high-frequency energy The expression is: ; The expression is:

[0113] ;

[0114] In the above formula, The j The first coefficient value in the sequence of wavelet detail coefficients at the The first coefficient value in the sequence of wavelet detail coefficients at the The first coefficient value in the sequence of wavelet detail coefficients at the The first coefficient value in the sequence of wavelet detail coefficients at the The length of the sequence of detail coefficients at the j The length of the sequence of detail coefficients at the The length of the sequence of detail coefficients at the The cumulative index, , is used to traverse all the coefficients at the j The cumulative index, , is used to traverse all the coefficients at the The proportion of the j The proportion of the j The proportion of the

[0115] After multi-scale processing of various time series signals (water consumption, weather) using wavelet decomposition, the approximate coefficients mainly reflect long-term trends, while the detail coefficients retain rapid fluctuations that occur at different time scales. Using only the mean, variance, and other low-frequency statistics of will cause the model to ignore these "short-term anomalies" or "periodic spikes", thereby weakening the ability to perceive features such as sudden water consumption peaks.

[0116] Measuring the amplitude of the detail coefficients by energy and the "distribution dispersion" of the energy ratio by entropy can sensitively reveal sudden spikes or periodic pulses, providing a basis for distinguishing between "normal fluctuations" and "abnormal shocks" in the prediction network.

[0117] The of each sequence is spliced according to the source type to form a multi-source multi-scale feature vector ; principal component analysis (PCA) is used to reduce the dimensionality of to obtain the final input feature . By compressing to a small number of principal components, the network input dimension can be reduced without losing much information, the training convergence time can be shortened, and the model's generalization ability on new samples can be improved.

[0118] The input tensor information density is higher after multi-scale decomposition and dimensionality reduction, and the gradient is more stable after feeding into the parallel CNN-LSTM / GRU, the fitting is more sufficient, and finally the water consumption prediction error is significantly lower than the baseline scheme without multi-scale processing.

[0119] 3) input the features of step 2) into the water consumption prediction model, and output the predicted water consumption of each time period of the target region on the scheduling day; specifically including the following steps:

[0120] In this embodiment, the prediction model is composed of parallel CNN-LSTM and CNN-GRU, and the model parameters are updated online through the policy gradient algorithm.

[0121] in the form of a sliding window with a length of L and a dimension of P, where the dimension P is the dimension of the feature vector after dimension reduction in step 2);

[0122] The last L dimension-reduced feature vectors are assembled into a matrix as follows:

[0123]

[0124] are input into two parallel branches CNN-LSTM and CNN-GRU at the same time;

[0125] In the CNN-LSTM branch, first, a one-dimensional convolution layer is used to extract features from the input, and ReLU activation is applied to the convolution output; then the result is input into an LSTM network with a hidden dimension of D (the hidden dimension D is usually 30-60% of the length of the input vector), and the hidden state at the last time is taken as the output feature vector of the branch; the sequence length after convolution is , where K represents the width of the convolution kernel.

[0126] In the CNN-GRU branch, another set of one-dimensional convolution and ReLU activation is used to extract features, which are then input into a GRU network with a hidden dimension of D' (the hidden dimension D' is usually 20-40% of the length of the input vector), and the hidden state at the last time is taken as the branch output .

[0127] The output features of the two branches are concatenated into a vector , which is mapped to a prediction vector with a length of H through a fully connected layer, as follows:

[0128]

[0129] where represents the weight matrix of the fully connected layer, represents the bias vector of the fully connected layer;

[0130] ​​Prediction vector The water consumption prediction value of the target area for the next H time periods is scheduled.

[0131] The above parallel use of CNN-LSTM and CNN-GRU can complement the advantages of the two types of recurrent units: CNN first extracts local spatiotemporal features, LSTM is good at capturing long-term trends, and GRU responds quickly to short-term fluctuations and has strong robustness. After the output of the two branches is spliced, it is equivalent to a lightweight integrated model that can simultaneously perceive "slow change + rapid disturbance" and reduce the risk of underfitting or overfitting of a single model; when the online policy gradient is fine-tuned, the gradient difference between the two paths provides complementary update directions, allowing the model to quickly converge under sudden loads. This significantly reduces the prediction error and makes it more adaptable to abnormal shocks, providing more accurate and stable benchmark data for subsequent multi-target scheduling.

[0132] In this embodiment, the prediction model uses the negative square norm of the prediction error as the immediate reward, and uses the REINFORCE policy gradient method to update all trainable parameters θ (including convolution kernels, LSTM / GRU weights, and fully connected layer parameters, etc.) online; wherein, The expression is as follows:

[0133]

[0134] Wherein, represents the true observation value vector (i.e., the actual value vector), represents the prediction value vector.

[0135] 4) Real-time acquisition of reservoir storage capacity, flow, water quality, precipitation, and pollution source monitoring data, and input into the reservoir inflow-water quality coupling prediction model to obtain the inflow prediction value and water quality prediction value of the reservoir at each time period on the scheduling day; including the following steps:

[0136] Real-time acquisition of reservoir storage capacity, inflow observation value, outflow, water quality index vector, precipitation, and upstream pollution source monitoring data vector set, and obtain the evaporation loss data of the reservoir to form a multi-source data, and the multi-source data is synchronized according to a unified time step Time synchronization interpolation, filtering, and linear interpolation are performed to obtain , , , , , , ; represents the storage capacity of the reservoir at , represents the inflow of the reservoir at , represents the outflow of the reservoir at real-time outbound flow Indicates that the reservoir is Water quality index vector at time point, This indicates that the reservoir is in Rainfall at any given time Indicates in Vector set of monitoring data of pollution sources upstream of the reservoir at any time Indicates that the reservoir is Evaporation loss (unit: m³) 3 In this embodiment, It is calculated using the Penman–Monteith equation combined with predicted temperature, humidity, wind speed, radiation, etc.

[0137] Preliminary predictions of flow and water quality were made using a simplified SWAT hydrological-water quality model, including:

[0138] Calculate the predicted inflow rate based on changes in water storage, evaporation loss, and inflow rate. , The expression is as follows:

[0139]

[0140] In the formula, It represents the change in the volume of water stored in the reservoir. Indicates the water area in the reservoir; precipitation. Meteorological vectors from a dynamic 3D topology model Extract from;

[0141] Among them, upstream pollution source monitoring data vector This refers to the set of all upstream monitoring points located from the dynamic 3D topology model that flowed into the target reservoir at the corresponding time. A vector composed of the measured concentrations of each pollutant; where, , , which indicates the first i Each monitoring point is at The pollutant concentration vector; This indicates the types of pollutants, including concentrations of ammonia nitrogen, total nitrogen, total phosphorus, and heavy metals in the water; among them, the water quality index vector... It contains concentration data for each pollutant. The pollutant concentration in the dynamic three-dimensional topological model is derived from the water quality index vector. Obtained from [the source].

[0142] This traffic Upstream inflow concentration The water storage volume at the end of the previous step and pollutant attenuation coefficient , the water quality of the reservoir is estimated according to the discretized transport-decay equation , The expression of is as follows:

[0143]

[0144] wherein, The expression of is as follows:

[0145]

[0146] wherein, represents the inflow of the th monitoring point at the th time point; i That is, The data can be obtained from the dynamic three-dimensional topological model. The BP neural network with an adaptive learning rate and a momentum factor is constructed for the input, to predict the flow correction amount

[0147] and the correction amount of water quality in the future H time periods , the SWAT model output is added to the network correction amount, to obtain the high-precision inflow and water quality coupled prediction results in the future H time periods . wherein, ,...,

[0148] ; similarly, ,..., ; In the training process of the BP neural network, the obtained in the historical data is taken as the input, and the measured data of the inflow and the water quality index vector of the th time point in the historical data

[0149] is taken as the output, and the model is trained until the model converges. In this embodiment, the BP neural network introduces an adaptive learning rate and a momentum factor in the back propagation process to improve the convergence speed; wherein, the adaptive learning rate mechanism satisfies:

[0150]

[0151]

[0152] wherein, is the mean square error of the current round, is the mean square error of the previous round; is the current learning rate, is the learning rate used in the next iteration; is an amplification factor, (usually 1.05~1.2);​​​ a reduction factor, (usually 0.5-0.9).

[0153]

[0154] wherein, and respectively represent the predicted values of the current inflow and water quality, and respectively represent the actual observed values of the current inflow and water quality.

[0155] The BP neural network introduces a momentum term to correct the weight update, and the expression is as follows:

[0156]

[0157]

[0158] wherein, represents the neural network weight parameter at the current time t, represents the weight at the next time , which is the result of updating the current weight according to the correction amount; represents the weight update amount at the current time t; represents the learning rate at the current time; represents a momentum factor, which is used to control the influence degree of the last update on the current update, and the value is usually in [0.5, 0.9]; represents the mean square error at the current time , and represents the gradient of the weight , which represents the "error reduction guide" in the current direction, and the larger this item is, the more sensitive the error is to the change of the weight.

[0159] In this step, the SWAT hydrological-water quality mechanism model is coupled with the BP neural network with adaptive learning rate and momentum factor online, realizing the deep integration of physical constraints and data-driven advantages: the SWAT equation retains the explainability of the reservoir water quantity change and the pollutant transport-decay process, so that the model still has robust physical consistency under different water regime conditions; in addition, the BP network uses the latest observation to systematically correct the SWAT output error in real time, which can capture the nonlinear deviation caused by sudden intervention, illegal discharge or meteorological anomaly in the basin, while maintaining physical consistency, significantly reducing random residual error and system drift. Through this coupling mechanism, the inflow and water quality prediction mean square error is significantly lower than that of single mechanism model or pure data model, and it can continuously track the intra-day change, providing more accurate, timely and reliable water source conditions for multi-objective optimization scheduling, thereby effectively improving the water supply reliability and water quality safety guarantee capability.

[0160] 5) According to the prediction results of steps 3) and 4) and the current storage capacity of the reservoir, a multi-objective optimization scheduling model is established, with the maximum water supply reliability, the minimum energy consumption and the maximum water quality compliance rate as the target, and the water quality constraint as the limiting condition, and the non-dominated sorting genetic algorithm NSGA-II is used to solve, to obtain a set of Pareto optimal scheduling schemes; Specifically, the following steps are included:

[0161] Discretize the scheduling day into H time periods, denoted as t = 1 ~ H; Set decision variables in each time period t

[0162] Wherein, represents the water source outflow of the jth reservoir, i is the start-stop state of the jth pump, 1 means on, and 0 means off; is the total number of adjustable water sources; is the total number of controlled pump groups; On the basis of the target area water consumption prediction value of the scheduling day obtained in step 3)

[0163] and the inflow and water quality coupling prediction results obtained in step 4), the following objectives and constraints are established;

[0164] Wherein, the objectives include:

[0165] A) Maximize water supply reliability:

[0166] Wherein, is the total demand prediction value of the water consumption of the target area in period t, from step 3).

[0167] B) Minimize energy consumption:

[0168] Wherein, is the water density; g is the acceleration of gravity; represents the lift of the jth pump in period t; is the operating efficiency of the jth pump; represents the water flow of the jth pump in period t, , is the water flow corresponding to the water source node p, determined according to the flow of the outflow of each adjustable reservoir flowing through the node p; For example: b reservoirs are selected to deliver water to the target area, and the outflow of a reservoirs among the b reservoirs needs to flow through the p node, then the water flow corresponding to the p node is the sum of the outflow of the a reservoirs; represents the number of pumps at the node p.

[0169] C) Maximize water quality compliance rate:

[0170] ​in, Indicates the water quality assessment threshold; This is an indicator function; it takes the value 1 if the condition is met, and 0 otherwise. The predicted water quality value of the mixed water source at time t is expressed as follows:

[0171]

[0172] In the formula, The water quality prediction vector for the i-th reservoir at time t is obtained through step 4).

[0173] The constraints include:

[0174] a) Water volume is flat: .

[0175] b) Water supply: ;in, For the first i Prediction of available water flow from reservoirs; , t=1~H; where This represents the loss coefficient during the transportation process (usually taken as 0.7~0.95). This formula can be used to calculate the predicted available water flow from the reservoir for H time periods; where, For the first i The reservoir is there The inbound flow forecast at that time is obtained from step 4); For the first i The current water storage capacity of each reservoir is obtained from step 4); For the first i The outflow rate of each reservoir's water source is determined by the decision variables. It was obtained from the middle.

[0176] c) Pump flow rate: ;in, This indicates the maximum flow rate of pump j.

[0177] d) Hard constraints on water quality: ;in, This indicates the hard constraint threshold for water quality; Can be with Equal or more stringent.

[0178] by The temporal combinations are represented by chromosomes, and the non-dominated sorting genetic algorithm NSGA-II is used to solve the problem, obtaining a set of Pareto optimal scheduling schemes. .

[0179] 6) According to the water supply reliability, energy consumption and water quality standard preference, the final scheduling scheme is selected from the Pareto optimal scheme set, and the scheduling instruction and its execution result are stored by using the blockchain technology, so that the tamper-proof and traceability of the scheduling data are ensured.

[0180] Further, if there is an unsatisfied item in the constraint condition under the current scheduling scheme, the user can select an alternative scheduling scheme from the Pareto optimal scheduling scheme set, which satisfies all the constraint conditions.

[0181] In summary, the scheduling decision is intelligently migrated from experience type to data-physical collaboration and global trade-off. The method enables the scheduling scheme to automatically avoid water quality exceeding the standard while meeting the instantaneous water demand, and significantly reduces the power consumption of the pumping station; the Pareto frontier provides diversified alternative schemes for the decision maker, which can be flexibly switched between 'energy saving','stable supply' and 'high quality' according to the actual preference, and still has a feasible solution in the emergency scenario, thereby comprehensively improving the safety, economy and emergency resilience of the regional water supply system.

[0182] Referring to the accompanying drawings, Figure 2 The embodiment also provides a regional water supply emergency scheduling system, which applies the above-mentioned regional water supply emergency scheduling method, and the system comprises:

[0183] A data acquisition unit is configured to acquire geographic information system (GIS) data and Internet of Things (IoT) sensor data, and collect spatial coordinates and real-time operating states of water source nodes, water distribution pipe networks, water plants, user terminals and water quality monitoring nodes;

[0184] A topology construction unit is configured to construct a dynamic three-dimensional topology model of the water supply system according to the data collected by the data acquisition unit;

[0185] A multi-scale feature extraction unit is configured to acquire a water consumption sequence and meteorological information of a target region from a water supply system historical database, and generate corresponding high-frequency components and low-frequency components by using a wavelet multi-scale decomposition method, so as to serve as input features for water consumption prediction;

[0186] A water consumption prediction unit is configured to predict water consumption of each period of a scheduling day of the target region according to the features output by the multi-scale feature extraction unit;

[0187] A reservoir multi-source data acquisition unit is configured to acquire reservoir storage capacity, flow, water quality, precipitation and pollution source monitoring data in real time, and form multi-source data;

[0188] A reservoir inflow-water quality coupling prediction unit is configured to predict the inflow and water quality of each period of a scheduling day according to the multi-source data output by the reservoir multi-source data acquisition unit;

[0189] The multi-objective optimization scheduling unit is configured to establish a multi-objective optimization model with the maximum water supply reliability, the minimum energy consumption and the maximum water quality standard compliance rate as the target and the water quality constraint as the limiting condition according to the prediction results of the water consumption, the inflow and the water quality and the current water storage of the reservoir, and to call a non-dominated sorting genetic algorithm II (NSGA-II) to obtain a Pareto optimal scheduling scheme set;

[0190] The instruction issuing unit is configured to select a final scheduling scheme from the Pareto optimal scheduling scheme set and issue the selected scheduling instruction to the water supply system execution unit;

[0191] The water supply system execution unit is configured to implement corresponding flow regulation and pump start-stop related instructions according to the selected scheduling scheme.

[0192] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0193] If the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application, essentially or the part that contributes to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor execute all or part of the steps of the embodiments of the method of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0194] The above only describes some embodiments of the present application, and does not limit the protection scope of the present application. Any equivalent device or equivalent process transformation, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A regional water supply emergency dispatching method, characterized in that, Comprise: 1) Based on geographic information system GIS and Internet of Things IoT data, a dynamic three-dimensional topological model of the water supply system is constructed, and the topological model represents the spatial position and real-time running state of the water source node, water distribution pipe network, water plant, user terminal and water quality monitoring node; 2) Obtain the water consumption and meteorological information of the target area from the historical database of the water supply system, and perform wavelet multi-scale decomposition to obtain high-frequency components and low-frequency components as input features for water consumption prediction; 3) The input features of step 2) are sent to the water consumption prediction model, and the predicted values of the water consumption of the target area at each time period of the scheduling day are output; 4) Real-time acquisition of reservoir storage capacity, flow, water quality, precipitation and pollution source monitoring data, and input into the reservoir inflow flow-water quality coupling prediction model to obtain the predicted values of the inflow flow and water quality at each time period of the scheduling day; 5) According to the prediction results of steps 3) and 4) and the current storage capacity of the reservoir, a multi-objective optimization scheduling model is established, which takes the maximum water supply reliability, the minimum energy consumption and the maximum water quality standard rate as the target, and takes the water quality constraint as the restriction condition, and uses the non-dominated sorting genetic algorithm NSGA-II to solve, and obtains a set of Pareto optimal scheduling schemes; 6) Select the final scheduling scheme from the Pareto optimal scheme set and issue it to the water supply system execution unit for execution.

2. The regional water supply emergency dispatching method according to claim 1, characterized in that, Step 1) specifically comprises: The three-dimensional coordinates and elevations of the water source nodes, water plants, pipe network nodes, user terminals and monitoring nodes in the water supply system are obtained from a geographic information system (GIS); real-time operation data of each node, including flow , pressure , a meteorological data vector and a water quality index vector , are obtained from an Internet of Things (IoT) sensor, and then the GIS spatial data and the IoT real-time data are filtered, time-synchronized and coordinate-interpolated. A dynamic three-dimensional topological graph is generated with water source nodes, water distribution pipe networks, water plants, user terminals and water quality monitoring nodes as vertices and according to pipe segment connection relationships ; wherein, denotes a set of nodes of the water supply system, each node having a spatial position and a state vector , denotes a node type; a set of edges , denotes a pipe from a node to only if there is a pipe connection between the two nodes ; The topological model is dynamically adjusted through the actual collected data on each node and pipeline.

3. The regional water supply emergency dispatching method according to claim 1, characterized in that, Step 2) specifically comprises: Obtain multi-source time series of target area from historical database, unify by time step Align and interpolate missing values to obtain the following original time series: a sequence of water usage a sequence of weather information vectors ; wherein denotes the k-th time step, k=1,...,K ; For each sequence zero-meaning and variance normalization are performed to obtain normalized sequence ; The normalized sequence is decomposed by discrete wavelet transform (DWT) to obtain a first approximation coefficient sequence and a first detail coefficient sequence J J j ;​​​​​ from the approximation coefficients extracting low frequency statistical features with and for each level of detail coefficients calculating high frequency energy and entropy ; wherein, , ; the expression of high frequency energy is: ; the expression of ； In the above formula, Indicates the first j Level wavelet detail coefficient sequence The first in Each coefficient value; Indicates the first j Level detail coefficient sequence Length; Indicates an incremental index. Used to traverse all the first... j Grade coefficient; Indicates the first j The proportion of the k-th coefficient in the total energy of that level; Concatenate the sequences of each source type to form a multi-source multi-scale feature vector Concatenate, form multi-source multi-scale feature vector ; Adopt principal component analysis to reduce dimensionality to obtain the final input feature .

4. The regional water supply emergency dispatching method according to claim 3, characterized in that, The prediction model is composed of parallel CNN-LSTM and CNN-GRU, and the model parameters are updated online through the policy gradient algorithm; Step 3) specifically comprises: The dimension-reduced feature vectors of the previous L time instants are assembled into a matrix in the form of a sliding window of length L and dimension P assembled into a matrix as follows: Will Both parallel branches CNN-LSTM and CNN-GRU are input simultaneously; In the CNN-LSTM branch, the input is first feature-extracted by a one-dimensional convolutional layer, and ReLU activation is applied on the convolutional output; then the result is sent to an LSTM network with a hidden dimension of D , and the hidden state at the last time is taken as the output feature vector of this branch ; In the CNN-GRU branch, another set of one-dimensional convolution and ReLU activation is used to extract features, which are then input into a GRU network with a hidden dimension of D' The hidden state at the final time is taken as the output of the branch ; concatenate the output features of the two branches into a vector , mapped to a prediction vector of length H through a fully connected layer , expressed as follows: In the formula, denotes a weight matrix of the fully connected layer, denotes a bias vector of the fully connected layer; Predicted vector scheduling a water consumption prediction value for the target region for the next H time periods the prediction model in terms of a negative squared norm of the prediction error as an immediate reward and performing online updates of all trainable parameters θ using a policy gradient method; wherein The expression is as follows: wherein, represents a true observation vector, represents a predicted value vector.

5. The regional water supply emergency dispatching method according to claim 1, characterized in that, Step 4) specifically comprises: Real-time acquisition of reservoir storage capacity, observation value of inflow, outflow, water quality index vector, precipitation and upstream pollution source monitoring data vector set, and obtain the evaporation loss data set of the reservoir to form multi-source data Time synchronization interpolation, filtering, linear interpolation are performed to obtain 、 、 、 、 、 、 ; Reservoir storage capacity at , Inflow of the reservoir at , Outflow of the reservoir at , Water quality index vector of the reservoir at , Precipitation of the reservoir at , Upstream pollution source monitoring data vector set of the reservoir at , Evaporation loss of the reservoir at ; The simplified SWAT hydrology-water quality model is used to preliminarily predict the flow and water quality, including: The inflow forecast value is calculated according to the water storage variation, evaporation loss and inflow , The expression is as follows: In the formula, represents the change in the volume of the reservoir water storage, represents the water area in the reservoir; said upstream pollution source monitoring data vector all upstream monitoring points of the corresponding time inflow target reservoir located from the dynamic three-dimensional topological model measured concentration of each pollutant constitutes a vector; wherein, , , which represents the pollutant concentration vector of the i th monitoring point at ; represents the type of pollutant; The flow rate , upstream inflow concentration , reservoir volume at the end of the step and pollutant decay coefficient , the water quality of the reservoir is estimated according to the discretized transport-decay equation , The expression is as follows: wherein The expression of the above is as follows: in, Indicates the first i Each monitoring point is at Inbound traffic; Construct a BP neural network with adaptive learning rate and momentum factor for input , predict the flow correction amount of the future H time periods and water quality correction amount , add the network correction amount to the SWAT model output to obtain high-precision coupled prediction results of inflow and water quality in the future H time periods .

6. The regional water supply emergency dispatching method according to claim 5, characterized in that, The BP neural network introduces adaptive learning rate and momentum factor in the back propagation process to improve the convergence speed; wherein, the adaptive learning rate mechanism satisfies: wherein, is the mean squared error for the current epoch, is the mean squared error for the previous epoch; is the current learning rate, is the learning rate to be used for the next iteration; is the amplification factor, ; is the reduction factor, ; wherein, and respectively represent the predicted value of current inflow and water quality, and respectively represent the actual observed value of current inflow and water quality; The BP neural network introduces a momentum term to correct the weight update, and the expression is as follows: in, This represents the neural network weight parameters at the current time t. Indicates the next moment The weights; This represents the weight update amount at the current time t; This represents the learning rate at the current moment; Indicates the momentum factor; Indicates the mean square error at the current time. Weights The gradient.

7. The regional water supply emergency dispatching method according to claim 5, characterized in that, Step 5) specifically comprises: Discretize the scheduling day into H time periods, denoted as t = 1 ~ H; set decision variables ; wherein, represents the water source outflow of the jth reservoir, i represents the jth pump start-stop state, 1 represents on, and 0 represents off; represents the total number of adjustable reservoir water sources; represents the total number of controlled pump groups;​ On the basis of the target area daily water consumption prediction value obtained in step 3) and the warehouse inflow and water quality coupling prediction result obtained in step 4), the following target and constraint conditions are established On the basis of the target area daily water consumption prediction value obtained in step 3) and the warehouse inflow and water quality coupling prediction result obtained in step 4), the following target and constraint conditions are established Wherein, the target includes: A) Water supply reliability maximization: ; wherein, is the total demand forecast of water consumption for the target area for the period t; B) Minimization of energy consumption: ; wherein, is the water density; g is the acceleration of gravity; represents the head of the jth pump at time period t; is the operating efficiency of the jth pump; represents the water flow rate of the jth pump at time t, , is the water flow rate corresponding to the water source node p, which is determined according to the flow rate of water flowing through the node p from each adjustable reservoir; represents the number of pumps at the node p; C) Maximize the rate of water quality compliance: ; wherein, represents the water quality evaluation threshold value; is an indicator function, which takes 1 if the condition is satisfied, and 0 otherwise; represents the predicted value of the water quality of the mixed water source at time t, and its expression is as follows: In the formula, is the water quality prediction vector of the ith reservoir at time t; Wherein, the constraint condition includes: a) water level is horizontal: ; b) water supply: ; wherein, is the i reservoir water supply flow prediction; , t = 1 ~ H; wherein represents the loss coefficient in the delivery process; is the i reservoir inflow prediction at ; is the i current reservoir storage capacity; c) Pump flow rate: ; wherein, represents the maximum delivery flow rate of the jth pump; d) a water quality hardness constraint: ; wherein, represents a water quality hardness constraint threshold; With the timing combination as chromosome, the non-dominated sorting genetic algorithm NSGA-II is used for solution, and a Pareto optimal scheduling scheme set is obtained .

8. The regional water supply emergency dispatching method according to claim 7, characterized in that, In step 6), according to the water supply reliability, energy consumption and water quality standard preference, the final scheduling scheme is selected from the Pareto optimal scheme set and is issued to the water supply system execution unit for execution, and the blockchain technology is used to store the scheduling instruction and its execution result.

9. A regional water supply emergency dispatching system, characterized in that, Apply the regional water supply emergency scheduling method according to any one of claims 1-8, the system comprises: A data acquisition unit for acquiring geographic information system GIS data and Internet of Things IoT sensor data, and acquiring spatial coordinates and real-time running states of water source nodes, water distribution pipe networks, water plants, user terminals and water quality monitoring nodes; A topological construction unit for constructing a dynamic three-dimensional topological model of the water supply system according to the data collected by the data acquisition unit; A multi-scale feature extraction unit for obtaining water consumption sequences and meteorological information of a target area from a water supply system historical database, and generating corresponding high-frequency components and low-frequency components using a wavelet multi-scale decomposition method as input features for water consumption prediction; A water consumption prediction unit for predicting the water consumption of a target area at each time period of the scheduling day according to the features output by the multi-scale feature extraction unit; A reservoir multi-source data acquisition unit for acquiring real-time reservoir storage capacity, flow, water quality, precipitation and pollution source monitoring data to form multi-source data; The reservoir inflow-flow quality coupling prediction unit is configured to predict the inflow and water quality of each time period of the dispatch day according to the multi-source data output by the reservoir multi-source data acquisition unit. The multi-objective optimization scheduling unit is configured to establish a multi-objective optimization model with the maximum water supply reliability, the minimum energy consumption, and the maximum water quality standard compliance rate as the target and the water quality constraint as the limiting condition according to the prediction results of the water consumption, the inflow, and the water quality and the current water storage capacity of the reservoir, and to obtain a Pareto optimal scheduling scheme set by calling a non-dominated sorting genetic algorithm NSGA-II for solution. The instruction issuing unit is configured to select a final scheduling scheme from the Pareto optimal scheduling scheme set and issue the selected scheduling instruction to the water supply system execution unit. The water supply system execution unit is configured to implement corresponding flow regulation and pump start-stop related instructions according to the selected scheduling scheme.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the regional water supply emergency scheduling method according to any one of claims 1-8. The program is executed by the processor to implement the regional water supply emergency scheduling method according to any one of claims 1-8.

Images