Multi-target water supply scheduling method and system based on big data driving

Abstract

The invention discloses a multi-target water supply scheduling method and system based on big data driving, and the method comprises the steps: collecting the multi-source data of a plurality of water supply targets, and enabling the scheme to obtain more reliable data support in the scheduling planning of the water supply targets; based on multi-source data serving as input features, the scheme combines the multi-source data with a graph space-time prediction model, a season statistical model and a gradient lifting regression model to obtain water consumption predicted by the three models respectively, prediction results of the three models are integrated through error adaptive weighting, more reliable predicted water consumption requirements are obtained, and the prediction efficiency is improved. And after the linear distribution model of the water source and the water supply targets is solved, initial water supply distribution schemes in one-to-one correspondence with the multiple water supply targets are obtained, the water supply accuracy of the water supply targets can be improved, multiple costs are used as targets to be minimized, and after multi-constraint combined solving, the water supply accuracy of the water supply targets can be improved. Therefore, the scheduling cost optimization is considered under the condition that the multi-target water supply scheduling meets flexibility and reliability.

Description

Technical Field

[0001] The present invention relates to the technical fields of water consumption forecasting technology and water supply scheduling technology, and in particular to a multi-objective water supply scheduling method and system driven by big data. Background Art

[0002] With the acceleration of urbanization and industrialization, the water supply system has become a complex pattern of "multiple water sources, multiple users, and multiple levels". This makes the implementation of traditional water supply scheduling methods inflexible and the limitations of lack of precision in supply increasingly obvious. Among them, traditional scheduling relies on the following methods: (1) static experience curves or time-based quotas - mainly based on historical statistical averages to determine the pump start and stop and water source distribution curves, which can only ensure "uninterrupted water supply" and is difficult to take into account energy consumption or backup flexibility in real time; (2) single-objective heuristic optimization - many researchers focus on minimizing electricity costs or minimizing energy consumption per unit water volume, but ignore the differences in water shortage penalties for users of different levels, resulting in high-priority users being passively reduced in tense situations; (3) the hydraulic model of the pipeline network is often separated from the scheduling model, and the scheduler needs to trial and error between simulation and manual scheduling, which has slow response speed and is easily affected by human experience bias; (4) data silos - SCADA high-frequency pressure / flow, meter reading historical water consumption, weather forecasts, holiday events and other data are located in different systems, lacking a unified time scale and quality control, resulting in limited prediction accuracy and optimization feasibility.

[0003] In summary, the pain points of existing technical solutions include: (1) Single objective or simple weighting: It is impossible to simultaneously measure the multi-dimensional indicators of "energy consumption-cost-reliability-resilience"; (2) Poor robustness to uncertain demand: Once the point prediction error is too large, it may cause large-scale pressure shortage or oversupply; (3) Conflict between rolling closed loop and start-stop wear: If there is no minimum operation / stop constraint during high-frequency re-optimization, pump start-stop oscillation will aggravate equipment wear and also increase maintenance costs. Therefore, there is an urgent need for a water supply scheduling technology that can integrate multi-source heterogeneous big data, predict demand in real time, and solve multi-objective optimization online. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to propose a multi-objective water supply scheduling method and system driven by big data, which is reliable in implementation, flexible in application, can integrate multi-source data and has good supply scheduling effect.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is: A multi-objective water supply scheduling method driven by big data, comprising: S01. Collect location information, historical water consumption data, water supply network operation data, meteorological environment data, and water supply rule parameters of multiple water supply targets within the water supply scheduling area, and construct a multi-source data set corresponding to multiple water supply targets; S02. Fill missing values, identify and repair outliers in multi-source data according to preset conditions to complete preprocessing of multi-source data. Then, generate a spatiotemporal correlation diagram based on the hydraulic distance between water supply targets, historical water consumption correlation, and pressure coupling relationship. S03. Based on the preprocessed multi-source data, combined with time series, meteorological and / or water supply target priorities as input features, the water demand of each water supply target in the future time period is predicted through the graph spatiotemporal prediction model to obtain a first prediction result. Then, the seasonal statistical model and the gradient boosting regression model are combined to output the second and third prediction results respectively. Then, the predicted water demand of each water supply target in the future time period is obtained through error adaptive weighted integration, and a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual samples; S04. Allocate corresponding margins for the predicted water demands corresponding to the multiple water supply targets based on their preset priorities, and then solve the source-water supply target linear allocation model to obtain initial water supply allocation plans corresponding to the multiple water supply targets; S05. Within a preset rolling optimization time domain, a multi-objective optimization function is constructed with multiple costs as the objectives to be minimized. Then, a solution is performed based on the preset decision variables and under the condition that multiple preset constraints are satisfied to obtain a scheduling solution. S06. Execute the scheduling plan according to the preset rolling cycle to supply water to multiple water supply targets within the water supply scheduling area.

[0006] As a possible implementation, further, this solution S06 also includes: The water demand of multiple water supply targets is monitored in real time, and then the deviation rate between the two is calculated in combination with the predicted water demand. When the deviation rate exceeds the trigger threshold corresponding to the water supply target priority, the scheduling plan for the remaining period of the preset rolling cycle is re-optimized to obtain the updated scheduling decision and execute it.

[0007] As a possible implementation method, further, in this solution S01, the historical water consumption data includes the measured water consumption of the water supply target at the historical time t; the water supply network operation data includes the measured water supply pressure of the node where the water supply target is located, and the measured flow rate of the corresponding pipe section; the meteorological environment data includes one or more data of temperature, rainfall, relative humidity or evaporation; and the water supply rule parameters include one or more of the minimum pressure, maximum pressure, priority water shortage upper limit, and minimum duration of pump start and stop.

[0008] As a better implementation option, preferably, in this solution S02, a low-rank matrix completion algorithm is used to fill the missing values ​​of historical water use data in multi-source data, which reconstructs the historical water use matrix using an optimization model that includes a projection operator of the observation position and nuclear norm regularization.

[0009] As a better implementation option, preferably, this solution uses a statistical threshold method based on median filter residuals and / or a multivariate anomaly detection based on isolation forest to identify outliers in multi-source data, and then repairs the outliers by repairing the outliers using local linear interpolation to complete the preprocessing of multi-source data.

[0010] As a better implementation option, preferably, this solution S01 includes: N Water supply targets, data collection is performed under discrete time index set and rolling forecast window to obtain N The multi-source dataset is constructed by combining the location information of each water supply target, historical water consumption data, water supply network operation data, meteorological environment data and water supply rule parameters, and then constructing a multi-source dataset corresponding to multiple water supply targets; the ... The definition is as follows:

[0011] in, = 1, 2, ... N , which is the water supply target index, For water supply goals Measured water consumption at historical time t; For water supply goals Corresponding pipe network node n The measured pressure, For water supply goals Corresponding pipe section e The measured flow rate; are the temperature, rainfall, relative humidity and evaporation at historical time t respectively; is a date indicator variable; is a set of water supply rule parameters; For water supply goals Water supply priority label, For water supply goals The geographical location or pipe network topological location.

[0012] As a preferred implementation option, preferably, this solution S01 also includes: mapping data with different sampling frequencies to a uniform time step , for high-frequency data Perform aggregation, which is defined as follows:

[0013] in, is the original timestamp set falling into the tth unified time window, which represents the t The set of original sampling points corresponding to the time step; To unify scheduling or preset time steps, is the original high-frequency data, is the unified time step data after aggregation.

[0014] As a preferred implementation option, preferably, this solution S02 includes: The low-rank matrix completion algorithm is used to fill the missing values ​​of historical water use data in multi-source data, and a historical water use matrix is ​​constructed based on the historical water use data. , for the set with missing positions , using nuclear norm minimization, the function is defined as follows:

[0015] in, is the matrix to be reconstructed, is a projection operator that only retains the observed position and sets the rest to zero; is the Frobenius norm, is the nuclear norm, which is the sum of the singular values, is the regularization trade-off coefficient.

[0016] The statistical threshold method based on median filter residual and / or multivariate anomaly detection based on isolation forest are used to identify outliers in multi-source data, and a sliding median filter is constructed for each sequence. , and its residual is defined as follows

[0017] Interquartile range (IQR) threshold using the statistical threshold method Identify outliers, or give outlier scores using multivariate anomaly detection with isolation forests , when the threshold is exceeded , then the value is judged to be abnormal.

[0018] in, For water supply goals The measured water consumption at historical time t is, is the window width The median filter, 、 are the first and third quartiles of the residual distribution, is the empirical coefficient, and its value ranges from 1.5 to 3; 、 are the anomaly score value and score threshold of the isolation forest algorithm respectively; After identifying the outliers, the outliers are repaired by local linear interpolation to complete the preprocessing of multi-source data, and then the data feature vector is constructed for subsequent prediction. , which is expressed as follows:

[0019] in, = 1, 2, ... N , which is the water supply target index, 、 Water supply targets In historical time 、 The measured water consumption, is the lag order; are temperature and rainfall respectively; is a date indicator variable; For water supply goals Water supply priority label; is the hour of time step t.

[0020] Generate a spatiotemporal correlation map based on the hydraulic distance between water supply targets, historical water use correlation, and pressure coupling relationship , where the weight function is constructed as follows:

[0021] It is normalized and defined as follows:

[0022] in, = 1, 2, ... N , which is the water supply target index, j is the node index, is the traversal node index during normalized summation, is the hydraulic distance of the shortest path in the pipe network, is the Pearson correlation coefficient of historical water supply demand, is the pressure coupling, which is the sensitivity of pressure changes to changes in demand at another node, 、 、 is a non-negative combination weight, and + + =1, is the final adjacency matrix after normalization, is the unnormalized comprehensive weight, It is the comprehensive weight corresponding to the traversal node index during normalized summation.

[0023] As a preferred implementation option, preferably, this solution S03 includes: Based on the pre-processed multi-source data, combined with time series, meteorological and / or water supply target priority as input features, the water demand of each water supply target in the future time period is predicted by the graph spatiotemporal prediction model to obtain the first prediction result, wherein the graph spatiotemporal prediction model uses the most recent window length Tensor input , which contains all the node features of the water supply target and is processed using K-layer graph convolution and temporal structure. The graph convolution function is defined as follows:

[0024]

[0025]

[0026] in, For the The graph convolution operation of the layer, For the l The input feature representation of the layer, is the order of the Chebyshev polynomial, For the Tier k Order filter parameters, is a Chebyshev polynomial, 、 They are graph Laplacian operator and scaled Laplacian operator in graph convolution respectively; for The maximum eigenvalue of is the adjacency matrix, represents the symmetric normalized adjacency matrix.

[0027] The temporal convolution function is defined as follows:

[0028] in, For the The temporal convolution operation of the layer, 、 are the temporal convolution kernel and bias respectively, For nodes In the The input feature time series of the layer, is the activation function.

[0029] The prediction output function is defined as follows:

[0030] in, The spatiotemporal prediction model for water supply targets The first prediction result obtained by predicting water demand in the future time period is is the adjacency matrix, The length of the most recent window The tensor input contains all the node features of the water supply target, is the prediction step length.

[0031] The function definition of the seasonal statistical model is as follows:

[0032] in, is a lag operator, satisfying ; 、 、 、 are non-seasonal and seasonal AR / MA polynomials; 、 is the difference order, For seasonal cycles, is the residual white noise, For water supply goals Water consumption at time t.

[0033] After performing differential detrending and deseasonalization through seasonal statistical models, we can obtain the water supply target. The corresponding steady-state series of water consumption ,in, is the lag operator, 、 is the difference order, For seasonal cycles, For water supply goals Water consumption at time t.

[0034] In single-step prediction, the prediction step size , for the stationary series , SARIMA degenerates into ARMA( ), the prediction function is defined as follows:

[0035] in, is the predicted value of the next step of the stationary sequence in single-step prediction, , which is the equivalent AR total order, , which is the total order of the equivalent MA; is the non-seasonal AR order, is the seasonal AR order, is the non-seasonal MA order, is the seasonal MA order, For seasonal cycles, 、 are the autoregressive coefficient and the moving average coefficient, which are obtained from the polynomial expansion coefficients. For stationary series In the The value of the lag order, is the residual white noise in the j The estimated value at the order lag.

[0036] When performing a single-step prediction, the prediction output of the second prediction result is defined as follows:

[0037] in, is the second prediction result during single-step prediction, Indicates the inverse transformation of the difference and seasonal difference. is the predicted value of the next stationary series in single-step forecasting.

[0038] In multi-step forecasting, , use the predicted value of the previous step to replace the unknown true value, and then perform recursion to obtain the first h Predicted value at step time , .

[0039] The objective function of the gradient boosting regression model is defined as follows:

[0040]

[0041] in, Indicates the m regression trees, is the number of trees, is the squared error loss, For the m Number of leaves, For the j The predicted weight of the leaf, 、 is the regularization parameter; For water supply goals Water consumption at time t; For water supply goals The feature vector of the data at time t.

[0042] When making a single-step prediction based on the gradient boosting regression model, it is assumed that M After rounds of boosting, the model is a collection of trees , then the prediction function is defined as follows:

[0043] in, , which is the water supply target The data feature vector at time t, where = 1, 2, ... N , which is the water supply target index, 、 Water supply targets In historical time 、 The measured water consumption, is the lag order; are temperature and rainfall respectively; is a date indicator variable; For water supply goals Water supply priority label; For the hour at time step t, each tree is a set of leaf weights With the split condition, when predicting, by putting Falling from top to bottom to a leaf , and then output the weight of the leaf ; is the third prediction result during single-step prediction.

[0044] In multi-step prediction, As the next feature Continue calling Functions, loops H Steps, get multi-step prediction value , , which is set as the third prediction result.

[0045] Obtain the first prediction result predicted by the graph spatiotemporal prediction model , the second forecast result predicted by the seasonal statistical model And the third prediction result predicted by the gradient boosting regression model Finally, the predicted water demand of each water supply target in the future time period is obtained through error adaptive weighted integration.

[0046] The error adaptive weighted integration includes calculating the most recent calibration window W The mean absolute percentage error (MAPE) of each model is defined as follows:

[0047] The ensemble weights are defined as follows:

[0048] in, , ; , For the model In the Recent Calibration window W Mean absolute percentage error MAPE within; The fusion prediction function is defined as follows:

[0049] in, N is the number of models involved in the prediction, For the model j Water supply targets The prediction result at time t is, For water supply goals The water consumption at time t, For water supply goals Across time steps h When the water demand is predicted by integrating the first prediction result, the second prediction result and the third prediction result; 、 、 is the fusion weight of the first prediction result, the second prediction result, and the third prediction result.

[0050] After obtaining the predicted water demand, a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual sample, where the residual sample formula is defined as follows:

[0051] in, is the prediction residual, Water supply targets In time t Actual and predicted water consumption.

[0052] The scenario construction through bootstrapping is defined as follows:

[0053] in, S =1, 2…… S ; S is the number of scenes, is the scene bias bootstrapped from the residual set; is the predicted water consumption after scenario correction; For water supply goals In time Forecasted water consumption.

[0054] If the distribution is approximately normal, the confidence interval is defined as follows:

[0055] in, 、 are the lower and upper confidence limits, respectively. is the quantile of the normal distribution, is the estimate of the residual standard deviation, For water supply goals In time Forecasted water consumption.

[0056] As a preferred implementation option, preferably, this solution S04 includes: According to the preset priorities of multiple water supply targets, the corresponding margin is configured for the predicted water demand , whose function is defined as follows:

[0057] in, For water supply goals The safety factor of the corresponding water supply priority label is greater than 0, and the higher the priority, the larger the value. For water supply goals In time Forecasted water consumption.

[0058] By solving the source-water supply target linear allocation model, the initial water supply allocation plan corresponding to multiple water supply targets is obtained. ; Among them, the source-water supply target linear allocation model is defined as follows: Assuming water source set , the decision is , then the objective function is as follows:

[0059] It includes the following constraints:

[0060]

[0061]

[0062] in, For water source k Towards the goal exist Water supply allocation for each time period; For water source k Unit water supply cost; For water source k exist Available water supply capacity during the period, For water supply goals The corresponding margin.

[0063] As a preferred implementation option, in this solution S05, a multi-objective optimization function is constructed with one or more costs including electricity consumption cost, water source cost, and pump start-up and shutdown cost as the objectives to be minimized.

[0064] S05 includes: within the preset rolling optimization time domain, according to the rolling optimization window Constructing decision variables ; It is expressed as follows: Rolling optimization window ; Decision variables ; in, is the number of optimization window steps, which represents the time span of the optimization window. is the execution time corresponding to the decision; For the moment The planned supply of the pump at the time is in the range of [0, ], For water supply goals Predicted water consumption at time t; For the moment The water shortage of the water source is [0, ], For water supply goals The tolerable water shortage ratio corresponding to the water supply priority is [0, 1]; is the kth water source at time The water intake range is [0, ], For water source k At the moment Available water supply capacity, For pipe section e At the moment of traffic, For nodes At the moment The node pressure, For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump; For the moment The amount of water stored at the water source.

[0065] Taking multiple costs as the objectives to be minimized, a multi-objective optimization function is constructed, which is defined as follows:

[0066]

[0067]

[0068]

[0069]

[0070] in, is the energy cost, is the water production cost, Punishment for lack of water, is the start-stop cost, are target weight coefficients respectively, and their cumulative sum is 1.

[0071] In energy costs In the calculation, the number of pumps is p , and their corresponding numbers are 1, 2... p , For the moment The electricity price during the period, For pumps p efficiency, For pumps p At the moment The lift, is the time step.

[0072] The cost of water production In the calculation, the amount of water source is k , and their corresponding numbers are 1, 2... k , For water source k The unit water supply cost, is the kth water source at time The water intake range is [0, ], For water source k At the moment available water supply capacity.

[0073] Punishment for lack of water Calculating, is the water shortage penalty coefficient, For the moment the amount of water shortage; Start-stop costs Calculating, For pumps p Start-up and shutdown costs; For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump.

[0074] In the default decision variables Under these conditions, it satisfies the following multiple constraints: (1) Constraints on demand and water shortage:

[0075] in, , ; For water supply goals At the moment Forecasted water consumption; For the moment The planned supply of the pump at the time is in the range of [0, ]; For the moment water shortage.

[0076] (2) Priority water shortage upper limit constraint:

[0077] in, For the moment the amount of water shortage; For water supply goals The tolerable water shortage ratio corresponding to the water supply priority is [0, 1]. For water supply goals At the moment Forecasted water consumption.

[0078] (3) Node traffic balance constraints

[0079] in, 、 are the collections of inflow and outflow pipe sections respectively; is the set of water sources connected to the node, is the water supply target set for taking water at the corresponding node; For pipe section e At the moment of traffic, is the kth water source at time The amount of water taken, For the moment The planned pump supply volume at that time.

[0080] (4) Dynamic constraints on water volume in water storage tanks

[0081]

[0082] in, For the moment The amount of water stored in the water source, 、 They are the upper and lower limits of the storage water volume respectively.

[0083] (5) Pump constraints and start-stop logic

[0084] in, For pumps p of traffic, For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump; For pumps p Rated maximum flow rate.

[0085] (6) Pressure and pipe section hydraulic approximation constraints Using linear constraints, let

[0086] in, , is the linearized friction coefficient, which is obtained from the derivative of the reference flow point; 、 Node n The maximum safety pressure and minimum pressure; For pipe section e At the moment of traffic, 、 Node 、 m At the moment Node pressure.

[0087] (7) Water source capacity constraints

[0088] in, For water source k At the moment Available water supply capacity, is the kth water source at time of water intake.

[0089] Based on constraints (1)-constraint (7), the decision variables are minimized to obtain the scheduling solution, which is defined as: .

[0090] Based on the above, this solution also proposes a multi-objective water supply scheduling system driven by big data, which includes: A data collection unit is used to collect location information, historical water consumption data, water supply network operation data, meteorological environment data and water supply rule parameters of multiple water supply targets in the water supply scheduling area, and construct a multi-source data set corresponding to multiple water supply targets; The data preprocessing unit is used to fill missing values, identify and repair outliers in multi-source data according to preset conditions to complete the preprocessing of multi-source data, and then generate a spatiotemporal correlation diagram based on the hydraulic distance between the water supply targets, the historical water consumption correlation and the pressure coupling relationship; A water use prediction unit is used to predict the water demand of each water supply target in the future time period based on preprocessed multi-source data, combined with time series, meteorological and / or water supply target priorities as input features, and obtain a first prediction result through a graph spatiotemporal prediction model. Then, a seasonal statistical model and a gradient boosting regression model are combined to output a second prediction result and a third prediction result respectively. Then, an error adaptive weighted integration is performed to obtain the predicted water demand of each water supply target in the future time period, and a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual sample. A water supply planning unit is used to allocate corresponding margins to the predicted water demands corresponding to the preset priorities of multiple water supply targets, and then solve the source-water supply target linear allocation model to obtain an initial water supply allocation plan corresponding to the multiple water supply targets; The scheduling optimization unit is used to construct a multi-objective optimization function within a preset rolling optimization time domain, using multiple costs as the objectives to be minimized. Then, the function is solved by combining preset decision variables and satisfying multiple preset constraints to obtain a scheduling solution. A water supply execution unit is used to execute the scheduling plan according to a preset rolling cycle to supply water to multiple water supply targets within the water supply scheduling area; The water supply monitoring unit is used to monitor the water demand of multiple water supply targets in real time, and then calculate the deviation rate between the two in combination with the predicted water demand. When the deviation rate exceeds the trigger threshold corresponding to the water supply target priority, the scheduling optimization unit is called to re-optimize the scheduling plan for the remaining period of the preset rolling cycle, obtain the updated scheduling decision and send it to the water supply execution unit for execution.

[0091] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: the present invention cleverly collects multi-source data of multiple water supply targets and then pre-processes the data to eliminate missing values ​​and abnormal values ​​in the multi-source data, so that the present invention can obtain more reliable data support in the scheduling and planning of water supply targets; on this basis, the present invention is based on multi-source data, combines time series, meteorological or water supply target priority as input features, and respectively combines them with the graph spatiotemporal prediction model, seasonal statistical model and gradient boosting regression model to obtain the water consumption predicted by the three models (the first prediction result, the second prediction result , the third prediction result), by integrating the prediction results of the three models through error adaptive weighting, a more reliable predicted water demand is obtained, and then the preset priority of the water supply target is configured with a margin. After solving the linear allocation model of water sources and water supply targets, an initial water supply allocation plan corresponding to multiple water supply targets is obtained, which can improve the accuracy and flexibility of water supply for water supply targets. By using multiple costs as the objectives to be minimized and jointly solving multiple constraints, a more economical and reliable scheduling plan can be obtained, so that the multi-objective water supply scheduling can take into account the optimization of scheduling costs while meeting the requirements of flexibility and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0093] Figure 1 This is a brief implementation flow chart of the scheduling method of this solution; Figure 2 This is the spatiotemporal association diagram of this solution, which shows the association between the water source S and the user U; Figure 3 The spatiotemporal association diagram of this scheme shows the association between water source S and user U in time series; Figure 4 This is a schematic diagram of the unit module connection of the water supply scheduling system of this scheme. DETAILED DESCRIPTION

[0094] The present invention will be described in further detail below with reference to the accompanying drawings and examples. It is particularly noted that the following examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention. Similarly, the following examples are only some embodiments of the present invention and are not intended to be exhaustive. All other embodiments obtained by those of ordinary skill in the art without creative effort are intended to fall within the scope of protection of the present invention.

[0095] Combine Figure 1 As shown, this embodiment provides a multi-objective water supply scheduling method based on big data driving, which includes: S01. Collect location information, historical water consumption data, water supply network operation data, meteorological environment data, and water supply rule parameters of multiple water supply targets within the water supply scheduling area, and construct a multi-source data set corresponding to multiple water supply targets; S02. Fill missing values, identify and repair outliers in multi-source data according to preset conditions to complete preprocessing of multi-source data. Then, generate a spatiotemporal correlation diagram based on the hydraulic distance between water supply targets, historical water consumption correlation, and pressure coupling relationship. S03. Based on the preprocessed multi-source data, combined with time series, meteorological and / or water supply target priorities as input features, the water demand of each water supply target in the future time period is predicted through the graph spatiotemporal prediction model to obtain a first prediction result. Then, the seasonal statistical model and the gradient boosting regression model are combined to output the second and third prediction results respectively. Then, the predicted water demand of each water supply target in the future time period is obtained through error adaptive weighted integration, and a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual samples; S04. Allocate corresponding margins for the predicted water demands corresponding to the multiple water supply targets based on their preset priorities, and then solve the source-water supply target linear allocation model to obtain initial water supply allocation plans corresponding to the multiple water supply targets; S05. Within a preset rolling optimization time domain, a multi-objective optimization function is constructed with multiple costs as the objectives to be minimized. Then, a solution is performed based on the preset decision variables and under the condition that multiple preset constraints are satisfied to obtain a scheduling solution. S06. Execute the scheduling plan according to the preset rolling cycle to supply water to multiple water supply targets within the water supply scheduling area.

[0096] In order to improve the flexibility of strategy optimization, as a possible implementation method, this solution S06 further includes: The water demand of multiple water supply targets is monitored in real time, and then the deviation rate between the two is calculated in combination with the predicted water demand. When the deviation rate exceeds the trigger threshold corresponding to the water supply target priority, the scheduling plan for the remaining period of the preset rolling cycle is re-optimized to obtain the updated scheduling decision and execute it.

[0097] In terms of data collection, as a possible implementation method, further, in this solution S01, the historical water consumption data includes the measured water consumption of the water supply target at the historical time t; the water supply network operation data includes the measured water supply pressure of the node where the water supply target is located, and the measured flow rate of the corresponding pipe section; the meteorological environment data includes one or more data of temperature, rainfall, relative humidity or evaporation; the water supply rule parameters include one or more of the minimum pressure, maximum pressure, priority water shortage upper limit, and minimum duration of pump start and stop.

[0098] Since there may be problems such as missing and anomalies in the data collection process, especially for historical water use data, the reliability and continuity of the data are more critical. As a better implementation option, preferably, in this solution S02, a low-rank matrix completion algorithm is used to fill the missing values ​​of historical water use data in multi-source data. It uses a projection operator including the observation position and an optimization model of nuclear norm regularization to reconstruct the historical water use matrix.

[0099] In terms of outlier identification and repair, as a better implementation option, preferably, in this solution S02, a statistical threshold method based on median filter residuals and / or multivariate anomaly detection based on isolation forest is used to identify outliers in multi-source data, and then the outliers are repaired by local linear interpolation to complete the preprocessing of multi-source data.

[0100] In order to better collect multi-source data of multiple water supply targets and facilitate their reuse, as a better implementation option, preferably, this solution S01 includes: N Water supply targets, data collection is performed under discrete time index set and rolling forecast window to obtain N The multi-source dataset is constructed by combining the location information of each water supply target, historical water consumption data, water supply network operation data, meteorological environment data and water supply rule parameters, and then constructing a multi-source dataset corresponding to multiple water supply targets; the ... The definition is as follows:

[0101] in, = 1, 2, ... N , which is the water supply target index, For water supply goals Measured water consumption at historical time t; For water supply goals Corresponding pipe network node n The measured pressure, For water supply goals Corresponding pipe section e The measured flow rate; are the temperature, rainfall, relative humidity and evaporation at historical time t respectively; is a date indicator variable; is a set of water supply rule parameters; For water supply goals Water supply priority label, For water supply goals The geographical location or pipe network topological location.

[0102] In order to avoid data misalignment in time due to different acquisition frequencies when data is reused, as a better implementation option, preferably, this solution S01 also includes: mapping data with different sampling frequencies to a unified time step , for high-frequency data Perform aggregation, which is defined as follows:

[0103] in, is the original timestamp set falling into the tth unified time window, which represents the t The set of original sampling points corresponding to the time step; To unify scheduling or preset time steps, is the original high-frequency data, is the unified time step data after aggregation.

[0104] In terms of missing value processing, as a better implementation option, preferably, this solution S02 includes: The low-rank matrix completion algorithm is used to fill the missing values ​​of historical water use data in multi-source data, and a historical water use matrix is ​​constructed based on the historical water use data. , for the set with missing positions , using nuclear norm minimization, the function is defined as follows:

[0105] in, is the matrix to be reconstructed, is a projection operator that only retains the observed position and sets the rest to zero; is the Frobenius norm, is the nuclear norm, which is the sum of the singular values, is the regularization trade-off coefficient.

[0106] The statistical threshold method based on median filter residual and / or multivariate anomaly detection based on isolation forest are used to identify outliers in multi-source data, and a sliding median filter is constructed for each sequence. , and its residual is defined as follows

[0107] Interquartile range (IQR) threshold using the statistical threshold method Identify outliers, or give outlier scores using multivariate anomaly detection with isolation forests , when the threshold is exceeded , then the value is judged to be abnormal.

[0108] in, For water supply goals The measured water consumption at historical time t is, is the window width The median filter, 、 are the first and third quartiles of the residual distribution, is the empirical coefficient, and its value ranges from 1.5 to 3; 、 are the anomaly score value and score threshold of the isolation forest algorithm respectively.

[0109] After identifying the outliers, the outliers are repaired by local linear interpolation to complete the preprocessing of multi-source data, and then the data feature vector is constructed for subsequent prediction. , which is expressed as follows:

[0110] in, = 1, 2, ... N , which is the water supply target index, 、 Water supply targets In historical time 、 The measured water consumption, is the lag order; are temperature and rainfall respectively; is a date indicator variable; For water supply goals Water supply priority label; is the hour of time step t.

[0111] Generate a spatiotemporal correlation map based on the hydraulic distance between water supply targets, historical water use correlation, and pressure coupling relationship , where the weight function is constructed as follows:

[0112] It is normalized and defined as follows:

[0113] in, = 1, 2, ... N , which is the water supply target index, j is the node index, is the traversal node index during normalized summation, is the hydraulic distance of the shortest path in the pipe network, is the Pearson correlation coefficient of historical water supply demand, is the pressure coupling, which is the sensitivity of pressure changes to changes in demand at another node, 、 、 is a non-negative combination weight, and + + =1, is the final adjacency matrix after normalization, is the unnormalized comprehensive weight, It is the comprehensive weight corresponding to the traversal node index during normalized summation.

[0114] in, Figure 2 This is the spatiotemporal association diagram of this solution, which shows the association between the water source S and the user U; Figure 2 An example of a node includes a water source (S1), a water tank / reservoir node (R), and relationships with multiple users (U1–U4). Figure 3 This is a spatiotemporal correlation diagram of this solution that uses time series to represent the relationship between water source S and user U within the same time layer (t-2, t-1, t). In the figure, the vertical (cross-time layer) dotted line represents the temporal dependency and autoregressive transmission (Temporal Link) of the same node between adjacent time layers, which is used to support the modeling of dynamic evolution by the spatiotemporal graph model. Figure 3 The juxtaposition of multiple time layers can reflect the combination of "spatial structure + temporal evolution"; in graph spatiotemporal neural networks or graph convolution + temporal convolution / LSTM, spatial convolution acts on the set of horizontal edges at each time layer, and temporal convolution / recurrent structure captures sequence inertia along the dotted line direction.

[0115] In terms of water demand forecasting, this solution combines multiple models for forecasting and then integrates them to obtain more reliable forecast results. As a better implementation option, preferably, this solution S03 includes: Based on the pre-processed multi-source data, combined with time series, meteorological and / or water supply target priority as input features, the water demand of each water supply target in the future time period is predicted by the graph spatiotemporal prediction model to obtain the first prediction result, wherein the graph spatiotemporal prediction model uses the most recent window length Tensor input , which contains all the node features of the water supply target and is processed using K-layer graph convolution and temporal structure. The graph convolution function is defined as follows:

[0116]

[0117]

[0118] in, For the The graph convolution operation of the layer, For the l The input feature representation of the layer, is the order of the Chebyshev polynomial, For the Tier k Order filter parameters, is a Chebyshev polynomial, 、 They are graph Laplacian operator and scaled Laplacian operator in graph convolution respectively; for The maximum eigenvalue of is the adjacency matrix, represents the symmetric normalized adjacency matrix.

[0119] The temporal convolution function is defined as follows:

[0120] in, For the The temporal convolution operation of the layer, 、 are the temporal convolution kernel and bias respectively, For nodes In the The input feature time series of the layer, is the activation function.

[0121] The prediction output function is defined as follows:

[0122] in, The spatiotemporal prediction model for water supply targets The first prediction result obtained by predicting water demand in the future time period is is the adjacency matrix, The length of the most recent window The tensor input contains all the node features of the water supply target, is the prediction step length.

[0123] After obtaining the first prediction result obtained by the graph spatiotemporal prediction model to predict the water demand of the water supply target in the future time period, this solution further predicts the second and third prediction results through the seasonal statistical model and the gradient boosting regression model.

[0124] Among them, the function definition of the seasonal statistical model is as follows:

[0125] in, is a lag operator, satisfying ; 、 、 、 are non-seasonal and seasonal AR / MA polynomials; 、 is the difference order, For seasonal cycles, is the residual white noise, For water supply goals Water consumption at time t.

[0126] After performing differential detrending and deseasonalization through seasonal statistical models, we can obtain the water supply target. The corresponding steady-state series of water consumption ,in, is the lag operator, 、 is the difference order, For seasonal cycles, For water supply goals Water consumption at time t.

[0127] When the seasonal statistical model is used for single-step prediction, the prediction step length is , for the stationary series , SARIMA degenerates into ARMA( ), the prediction function is defined as follows:

[0128] in, is the predicted value of the next step of the stationary sequence in single-step prediction, , which is the equivalent AR total order, , which is the total order of the equivalent MA; is the non-seasonal AR order, is the seasonal AR order, is the non-seasonal MA order, is the seasonal MA order, For seasonal cycles, 、 are the autoregressive coefficient and the moving average coefficient, which are obtained from the polynomial expansion coefficients. For stationary series In the The value of the lag order, is the residual white noise in the j The estimated value at the order lag.

[0129] When performing a single-step prediction, the prediction output of the second prediction result is defined as follows:

[0130] in, is the second prediction result during single-step prediction, Indicates the inverse transformation of the difference and seasonal difference. is the predicted value of the next stationary series in single-step forecasting.

[0131] When using seasonal statistical models for multi-step forecasting, , use the predicted value of the previous step to replace the unknown true value, and then perform recursion to obtain the first h Predicted value at step time , .

[0132] The objective function of the gradient boosting regression model in this scheme is defined as follows:

[0133]

[0134] in, Indicates the m regression trees, is the number of trees, is the squared error loss, For the m Number of leaves, For the j The predicted weight of the leaf, 、 is the regularization parameter; For water supply goals Water consumption at time t; For water supply goals The feature vector of the data at time t.

[0135] When making a single-step prediction based on the gradient boosting regression model, it is assumed that M After rounds of boosting, the model is a collection of trees , then the prediction function is defined as follows:

[0136] in, , which is the water supply target In time t The data feature vector of = 1, 2, ... N , which is the water supply target index, 、 Water supply targets In historical time 、 The measured water consumption, is the lag order; are temperature and rainfall respectively; is a date indicator variable; For water supply goals Water supply priority label; For the hour at time step t, each tree is a set of leaf weights With the split condition, when predicting, by putting Falling from top to bottom to a leaf , and then output the weight of the leaf ; is the third prediction result during single-step prediction.

[0137] In multi-step prediction, As the next feature Continue calling Functions, loops H Steps, get multi-step prediction value , , which is set as the third prediction result.

[0138] Obtain the first prediction result predicted by the graph spatiotemporal prediction model , the second forecast result predicted by the seasonal statistical model And the third prediction result predicted by the gradient boosting regression model Finally, the predicted water demand of each water supply target in the future time period is obtained through error adaptive weighted integration.

[0139] The error adaptive weighted integration includes calculating the most recent calibration window W The mean absolute percentage error (MAPE) of each model is defined as follows:

[0140] The ensemble weights are defined as follows:

[0141] in, , ; , For the model In the Recent Calibration window W Mean absolute percentage error MAPE within; The fusion prediction function is defined as follows:

[0142] in, N is the number of models involved in the prediction, For the model j Water supply targets The prediction result at time t is, For water supply goals The water consumption at time t, For water supply goals Across time steps h When the water demand is predicted by integrating the first prediction result, the second prediction result and the third prediction result; 、 、 is the fusion weight of the first prediction result, the second prediction result, and the third prediction result.

[0143] After obtaining the predicted water demand, a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual sample, where the residual sample formula is defined as follows:

[0144] in, is the prediction residual, Water supply targets In time t Actual and predicted water consumption.

[0145] The scenario construction through bootstrapping is defined as follows:

[0146] in, S =1, 2…… S ; S is the number of scenes, is the scene bias bootstrapped from the residual set; is the predicted water consumption after scenario correction; For water supply goals In time Forecasted water consumption.

[0147] If the distribution is approximately normal, the confidence interval is defined as follows:

[0148] in, 、 are the lower and upper confidence limits, respectively. is the quantile of the normal distribution, is the estimate of the residual standard deviation, For water supply goals In time Forecasted water consumption.

[0149] In terms of constructing a water supply distribution plan, as a preferred implementation option, preferably, this plan S04 includes: According to the preset priorities of multiple water supply targets, the corresponding margin is configured for the predicted water demand , whose function is defined as follows:

[0150] in, For water supply goals The safety factor of the corresponding water supply priority label is greater than 0, and the higher the priority, the larger the value. For water supply goals In time Forecasted water consumption.

[0151] By solving the source-water supply target linear allocation model, the initial water supply allocation plan corresponding to multiple water supply targets is obtained. ; Among them, the source-water supply target linear allocation model is defined as follows: Assuming water source set , the decision is , then the objective function is as follows:

[0152] It includes the following constraints:

[0153]

[0154]

[0155] in, For water source k Towards the goal exist Water supply allocation for each time period; For water source k Unit water supply cost; For water source k exist Available water supply capacity during the period, For water supply goals The corresponding margin.

[0156] For multi-objective optimization, as a better implementation option, preferably, in this solution S05, one or more costs including electricity consumption cost, water source cost, and pump start-up and shutdown cost are used as the objectives to be minimized to construct a multi-objective optimization function.

[0157] Specifically, the solution S05 includes: within the preset rolling optimization time domain, according to the rolling optimization window Constructing decision variables ; It is expressed as follows: Rolling optimization window ; Decision variables ; in, is the number of optimization window steps, which represents the time span of the optimization window. is the execution time corresponding to the decision; For the moment The planned supply of the pump at the time is in the range of [0, ], For water supply goals Predicted water consumption at time t; For the moment The water shortage of the water source is [0, ], For water supply goals The tolerable water shortage ratio corresponding to the water supply priority is [0, 1]; is the kth water source at time The water intake range is [0, ], For water source k At the moment Available water supply capacity, For pipe section e At the moment of traffic, For nodes At the moment The node pressure, For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump; For the moment The amount of water stored at the water source.

[0158] Taking multiple costs as the objectives to be minimized, a multi-objective optimization function is constructed, which is defined as follows:

[0159]

[0160]

[0161]

[0162]

[0163] in, is the energy cost, is the water production cost, Punishment for lack of water, is the start-stop cost, are target weight coefficients respectively, and their cumulative sum is 1.

[0164] In energy costs In the calculation, the number of pumps is p , and their corresponding numbers are 1, 2... p , For the moment The electricity price during the period, For pumps p efficiency, For pumps p At the moment The lift, is the time step.

[0165] The cost of water production In the calculation, the amount of water source is k , and their corresponding numbers are 1, 2... k , For water source k The unit water supply cost, is the kth water source at time The water intake range is [0, ], For water source k At the moment available water supply capacity.

[0166] Punishment for lack of water Calculating, is the water shortage penalty coefficient, For the moment the amount of water shortage; Start-stop costs Calculating, For pumps p Start-up and shutdown costs; For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump.

[0167] In the default decision variables Under these conditions, it satisfies the following multiple constraints: (1) Constraints on demand and water shortage:

[0168] in, , ; For water supply goals At the moment Forecasted water consumption; For the moment The planned supply of the pump at the time is in the range of [0, ]; For the moment water shortage.

[0169] (2) Priority water shortage upper limit constraint:

[0170] in, For the moment the amount of water shortage; For water supply goals The tolerable water shortage ratio corresponding to the water supply priority is [0, 1]. For water supply goals At the moment Forecasted water consumption.

[0171] (3) Node traffic balance constraints

[0172] in, 、 are the collections of inflow and outflow pipe sections respectively; is the set of water sources connected to the node, is the water supply target set for taking water at the corresponding node; For pipe section e At the moment of traffic, is the kth water source at time The amount of water taken, For the moment The planned pump supply volume at that time.

[0173] (4) Dynamic constraints on water volume in water storage tanks

[0174]

[0175] in, For the moment The amount of water stored in the water source, 、 They are the upper and lower limits of the storage water volume respectively.

[0176] (5) Pump constraints and start-stop logic

[0177] in, For pumps p of traffic, For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump; For pumps p Rated maximum flow rate.

[0178] (6) Pressure and pipe section hydraulic approximation constraints Using linear constraints, let

[0179] in, , is the linearized friction coefficient, which is obtained from the derivative of the reference flow point; 、 Node n The maximum safety pressure and minimum pressure; For pipe section e At the moment of traffic, 、 Node 、 m At the moment Node pressure.

[0180] (7) Water source capacity constraints

[0181] in, For water source k At the moment Available water supply capacity, is the kth water source at time of water intake.

[0182] Based on constraints (1)-constraint (7), the decision variables are minimized to obtain the scheduling solution, which is defined as: .

[0183] After solving the above constraints, an optimized scheduling plan can be obtained.

[0184] In order to improve the flexibility of strategy optimization, as a possible implementation method, further, this solution S06 also monitors the water demand of multiple water supply targets in real time to obtain real-time observed water consumption. , and then compare it with the predicted water consumption corresponding to the currently executed scheduling plan The deviation rate of the two is calculated together, which is defined as follows:

[0185] in, is the relative deviation rate, is a numerical stability constant and is a minimum value to avoid the denominator being 0.

[0186] In this scheme, the relative deviation rate is used to determine whether the current water supply strategy is appropriate. When the deviation rate exceeds the trigger threshold M corresponding to the water supply target priority (e.g. ), the scheduling plan for the remaining time period of the preset rolling cycle is re-optimized (i.e., S01-S04 are re-executed), and the updated scheduling decision is obtained and executed.

[0187] Combine Figure 4 As shown above, this solution also proposes a multi-objective water supply scheduling system driven by big data, which includes: A data collection unit is used to collect location information, historical water consumption data, water supply network operation data, meteorological environment data and water supply rule parameters of multiple water supply targets in the water supply scheduling area, and construct a multi-source data set corresponding to multiple water supply targets; The data preprocessing unit is used to fill missing values, identify and repair outliers in multi-source data according to preset conditions to complete the preprocessing of multi-source data, and then generate a spatiotemporal correlation diagram based on the hydraulic distance between the water supply targets, the historical water consumption correlation and the pressure coupling relationship; A water use prediction unit is used to predict the water demand of each water supply target in the future time period based on preprocessed multi-source data, combined with time series, meteorological and / or water supply target priorities as input features, and obtain a first prediction result through a graph spatiotemporal prediction model. Then, a seasonal statistical model and a gradient boosting regression model are combined to output a second prediction result and a third prediction result respectively. Then, an error adaptive weighted integration is performed to obtain the predicted water demand of each water supply target in the future time period, and a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual sample. A water supply planning unit is used to allocate corresponding margins to the predicted water demands corresponding to the preset priorities of multiple water supply targets, and then solve the source-water supply target linear allocation model to obtain an initial water supply allocation plan corresponding to the multiple water supply targets; The scheduling optimization unit is used to construct a multi-objective optimization function within a preset rolling optimization time domain, using multiple costs as the objectives to be minimized. Then, the function is solved by combining preset decision variables and satisfying multiple preset constraints to obtain a scheduling solution. A water supply execution unit is used to execute the scheduling plan according to a preset rolling cycle to supply water to multiple water supply targets within the water supply scheduling area; The water supply monitoring unit is used to monitor the water demand of multiple water supply targets in real time, and then calculate the deviation rate between the two in combination with the predicted water demand. When the deviation rate exceeds the trigger threshold corresponding to the water supply target priority, the scheduling optimization unit is called to re-optimize the scheduling plan for the remaining period of the preset rolling cycle, obtain the updated scheduling decision and send it to the water supply execution unit for execution.

[0188] In addition, the functional units in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0189] If the integrated unit is implemented 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 solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0190] The above descriptions are only some embodiments of the present invention and do not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A multi-objective water supply scheduling method based on big data, characterized in that: It includes: S01. Collect location information, historical water consumption data, water supply network operation data, meteorological environment data, and water supply rule parameters of multiple water supply targets within the water supply scheduling area, and construct a multi-source data set corresponding to multiple water supply targets; S02. Fill missing values, identify and repair outliers in multi-source data according to preset conditions to complete preprocessing of multi-source data. Then, generate a spatiotemporal correlation diagram based on the hydraulic distance between water supply targets, historical water consumption correlation, and pressure coupling relationship. S03. Based on the preprocessed multi-source data, combined with time series, meteorological and / or water supply target priorities as input features, the water demand of each water supply target in the future time period is predicted through the graph spatiotemporal prediction model to obtain a first prediction result. Then, the seasonal statistical model and the gradient boosting regression model are combined to output the second and third prediction results respectively. Then, the predicted water demand of each water supply target in the future time period is obtained through error adaptive weighted integration, and a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual samples; S04. Allocate corresponding margins for the predicted water demands corresponding to the multiple water supply targets based on their preset priorities, and then solve the source-water supply target linear allocation model to obtain initial water supply allocation plans corresponding to the multiple water supply targets; S05. Within a preset rolling optimization time domain, a multi-objective optimization function is constructed with multiple costs as the objectives to be minimized. Then, a solution is performed based on the preset decision variables and under the condition that multiple preset constraints are satisfied to obtain a scheduling solution. S06. Execute the scheduling plan according to the preset rolling cycle to supply water to multiple water supply targets within the water supply scheduling area.

2. The multi-objective water supply scheduling method based on big data drive according to claim 1 is characterized in that: S06. Also includes: The water demand of multiple water supply targets is monitored in real time, and then the deviation rate between the two is calculated in combination with the predicted water demand. When the deviation rate exceeds the trigger threshold corresponding to the water supply target priority, the scheduling plan for the remaining period of the preset rolling cycle is re-optimized to obtain the updated scheduling decision and execute it.

3. The multi-objective water supply scheduling method based on big data drive according to claim 1, characterized in that: In S01, the historical water consumption data includes the measured water consumption of the water supply target at the historical time t; The water supply network operation data includes the measured water supply pressure at the node where the water supply target is located and the measured flow rate of the corresponding pipe section; The meteorological environment data includes one or more data of temperature, rainfall, relative humidity or evaporation; The water supply rule parameters include at least one of the minimum pressure, the maximum pressure, the priority water shortage upper limit, and the minimum duration of pump start and stop.

4. The multi-objective water supply scheduling method based on big data drive according to any one of claims 1 to 3, characterized in that: In S02, a low-rank matrix completion algorithm is used to fill missing values ​​in the historical water use data from multi-source data. The historical water use matrix is ​​reconstructed using an optimization model that includes a projection operator with observation positions and nuclear norm regularization. The statistical threshold method based on median filter residuals and / or multivariate anomaly detection based on isolation forest are used to identify outliers in multi-source data, and then the outliers are repaired by local linear interpolation to complete the preprocessing of multi-source data.

5. The multi-objective water supply scheduling method based on big data drive according to claim 4 is characterized in that: S01 includes: N Water supply targets, data collection is performed under discrete time index set and rolling forecast window to obtain N The multi-source dataset is constructed by combining the location information of each water supply target, historical water consumption data, water supply network operation data, meteorological environment data and water supply rule parameters, and then constructing a multi-source dataset corresponding to multiple water supply targets; the ... The definition is as follows: in, = 1, 2, ... N , which is the water supply target index, For water supply goals Measured water consumption at historical time t; For water supply goals Corresponding pipe network node n The measured pressure, For water supply goals Corresponding pipe section e The measured flow rate; are the temperature, rainfall, relative humidity and evaporation at historical time t respectively; is a date indicator variable; is a set of water supply rule parameters; For water supply goals Water supply priority label, For water supply goals Geographical location or pipe network topological location; S01 also includes: mapping data of different sampling frequencies to a uniform time step , for high-frequency data Perform aggregation, which is defined as follows: in, is the set of original timestamps falling into the tth unified time window, which represents the set of original sampling points corresponding to the tth time step; To unify scheduling or preset time steps, is the original high-frequency data, is the unified time step data after aggregation.

6. The multi-objective water supply scheduling method based on big data drive according to claim 5, characterized in that: S02 includes: The low-rank matrix completion algorithm is used to fill the missing values ​​of historical water use data in multi-source data, and a historical water use matrix is ​​constructed based on the historical water use data. , for the set with missing positions , using nuclear norm minimization, the function is defined as follows: in, is the matrix to be reconstructed, is a projection operator that only retains the observed position and sets the rest to zero; is the Frobenius norm, is the nuclear norm, which is the sum of the singular values, is the regularization trade-off coefficient; The statistical threshold method based on median filter residual and / or multivariate anomaly detection based on isolation forest are used to identify outliers in multi-source data, and a sliding median filter is constructed for each sequence. , and its residual is defined as follows Interquartile range (IQR) threshold using the statistical threshold method Identify outliers, or give outlier scores using multivariate anomaly detection with isolation forests , when the threshold is exceeded , then the value is judged to be abnormal; in, For water supply goals The measured water consumption at historical time t is, is the window width The median filter, 、 are the first and third quartiles of the residual distribution, is the empirical coefficient, and its value ranges from 1.5 to 3; 、 are the anomaly score value and score threshold of the isolation forest algorithm respectively; After identifying the outliers, the outliers are repaired by local linear interpolation to complete the preprocessing of multi-source data, and then the data feature vector is constructed for subsequent prediction. , which is expressed as follows: in, = 1, 2, ... N , which is the water supply target index, 、 Water supply targets In historical time 、 The measured water consumption, is the lag order; are temperature and rainfall respectively; is a date indicator variable; For water supply goals Water supply priority label; is the hour of time step t; Generate a spatiotemporal correlation map based on the hydraulic distance between water supply targets, historical water use correlation, and pressure coupling relationship , where the weight function is constructed as follows: It is normalized and defined as follows: in, = 1, 2, ... N , which is the water supply target index, j is the node index, is the traversal node index during normalized summation, is the hydraulic distance of the shortest path in the pipe network, is the Pearson correlation coefficient of historical water supply demand, is the pressure coupling, which is the sensitivity of pressure changes to changes in demand at another node, 、 、 is a non-negative combination weight, and + + =1, is the final adjacency matrix after normalization, is the unnormalized comprehensive weight, It is the comprehensive weight corresponding to the traversal node index during normalized summation.

7. The multi-objective water supply scheduling method based on big data drive according to claim 6, characterized in that: S03 includes: Based on the pre-processed multi-source data, combined with time series, meteorological and / or water supply target priority as input features, the water demand of each water supply target in the future time period is predicted by the graph spatiotemporal prediction model to obtain the first prediction result, wherein the graph spatiotemporal prediction model uses the most recent window length Tensor input , which contains all the node features of the water supply target and is processed using K-layer graph convolution and temporal structure. The graph convolution function is defined as follows: in, For the The graph convolution operation of the layer, For the l The input feature representation of the layer, is the order of the Chebyshev polynomial, For the Tier k Order filter parameters, is a Chebyshev polynomial, 、 They are graph Laplacian operator and scaled Laplacian operator in graph convolution respectively; for The maximum eigenvalue of is the adjacency matrix, represents the symmetric normalized adjacency matrix; The temporal convolution function is defined as follows: in, For the The temporal convolution operation of the layer, 、 are the temporal convolution kernel and bias respectively, For nodes In the The input feature time series of the layer, is the activation function; The prediction output function is defined as follows: in, The spatiotemporal prediction model for water supply targets The first prediction result obtained by predicting water demand in the future time period is is the adjacency matrix, The length of the most recent window The tensor input contains all the node features of the water supply target, is the prediction step length; The function definition of the seasonal statistical model is as follows: in, is a lag operator, satisfying ; 、 、 、 are non-seasonal and seasonal AR / MA polynomials; 、 is the difference order, For seasonal cycles, is the residual white noise, For water supply goals Water consumption at time t; After performing differential detrending and deseasonalization through seasonal statistical models, we can obtain the water supply target. The corresponding steady-state series of water consumption ,in, is the lag operator, 、 is the difference order, For seasonal cycles, For water supply goals Water consumption at time t; In single-step prediction, the prediction step size , for the stationary series , SARIMA degenerates into ARMA( ), the prediction function is defined as follows: in, is the predicted value of the next step of the stationary sequence in single-step prediction, , which is the equivalent AR total order, , which is the total order of the equivalent MA; is the non-seasonal AR order, is the seasonal AR order, is the non-seasonal MA order, is the seasonal MA order, For seasonal cycles, 、 are the autoregressive coefficient and the moving average coefficient, which are obtained from the polynomial expansion coefficients. For stationary series In the The value of the lag order, is the residual white noise in the j The estimated value at the order lag; When performing a single-step prediction, the prediction output of the second prediction result is defined as follows: in, is the second prediction result during single-step prediction, Indicates the inverse transformation of the difference and seasonal difference. is the predicted value of the next step of the stationary sequence in single-step prediction; In multi-step forecasting, , use the predicted value of the previous step to replace the unknown true value, and then perform recursion to obtain the first h Predicted value at step time , ; The objective function of the gradient boosting regression model is defined as follows: in, Indicates the m regression trees, is the number of trees, is the squared error loss, For the m Number of leaves, For the j The predicted weight of the leaf, 、 is the regularization parameter; For water supply goals Water consumption at time t; For water supply goals Data feature vector at time t; When making a single-step prediction based on the gradient boosting regression model, it is assumed that M After rounds of boosting, the model is a collection of trees , then the prediction function is defined as follows: in, , which is the water supply target In time t The data feature vector of = 1, 2, ... N , which is the water supply target index, 、 Water supply targets In historical time 、 The measured water consumption, is the lag order; are temperature and rainfall respectively; is a date indicator variable; For water supply goals Water supply priority label; For the hour at time step t, each tree is a set of leaf weights With the split condition, when predicting, by putting Falling from top to bottom to a leaf , and then output the weight of the leaf ; is the third prediction result during single-step prediction; In multi-step prediction, As the next feature Continue calling Functions, loops H Steps, get multi-step prediction values , , which is set as the third prediction result; Obtain the first prediction result predicted by the graph spatiotemporal prediction model , the second forecast result predicted by the seasonal statistical model And the third prediction result predicted by the gradient boosting regression model Finally, the predicted water demand of each water supply target in the future time period is obtained through error adaptive weighted integration; The error adaptive weighted integration includes calculating the most recent calibration window W The mean absolute percentage error (MAPE) of each model is defined as follows: The ensemble weights are defined as follows: in, , ; , For the model In the Recent Calibration window W Mean absolute percentage error MAPE within; The fusion prediction function is defined as follows: in, N is the number of models involved in the prediction, For the model j Water supply targets The prediction result at time t is, For water supply goals The water consumption at time t, For water supply goals Across time steps h When the water demand is predicted by integrating the first prediction result, the second prediction result and the third prediction result; 、 、 is the fusion weight of the first prediction result, the second prediction result, and the third prediction result; After obtaining the predicted water demand, a demand uncertainty scenario set and / or prediction confidence interval is constructed based on the residual sample, where the residual sample formula is defined as follows: in, is the prediction residual, Water supply targets In time t Actual and forecast water consumption; The scenario construction through bootstrapping is defined as follows: in, S =1, 2…… S ; S is the number of scenes, is the scene bias bootstrapped from the residual set; is the predicted water consumption after scenario correction; For water supply goals In time Forecasted water consumption; If the distribution is approximately normal, the confidence interval is defined as follows: in, 、 are the lower and upper confidence limits, respectively. is the quantile of the normal distribution, is the estimate of the residual standard deviation, For water supply goals In time Forecasted water consumption.

8. The multi-objective water supply scheduling method based on big data drive according to claim 7 is characterized in that: S04 includes: According to the preset priorities of multiple water supply targets, the corresponding margin is configured for the predicted water demand , whose function is defined as follows: in, For water supply goals The safety factor of the corresponding water supply priority label is greater than 0, and the higher the priority, the larger the value. For water supply goals In time Forecasted water consumption; By solving the source-water supply target linear allocation model, the initial water supply allocation plan corresponding to multiple water supply targets is obtained. ; Among them, the source-water supply target linear allocation model is defined as follows: Assuming water source set , the decision is , then the objective function is as follows: It includes the following constraints: in, For water source k Towards the goal exist Water supply allocation for each time period; For water source k Unit water supply cost; For water source k exist Available water supply capacity during the period, For water supply goals The corresponding margin.

9. The multi-objective water supply scheduling method based on big data drive according to claim 8, characterized in that: In S05, a multi-objective optimization function is constructed with one or more costs including electricity consumption cost, water source cost, and pump start-up and shutdown cost as the objective to be minimized; S05 includes: Within the preset rolling optimization time domain, according to the rolling optimization window Constructing decision variables ; Rolling optimization window ; Decision variables ; in, is the number of optimization window steps, which represents the time span of the optimization window. is the execution time corresponding to the decision; For the moment The planned supply of the pump at the time is in the range of [0, ], For water supply goals Predicted water consumption at time t; For the moment The water shortage of the water source is [0, ], For water supply goals The tolerable water shortage ratio corresponding to the water supply priority is [0, 1]; is the kth water source at time The water intake range is [0, ], For water source k At the moment Available water supply capacity, For pipe section e At the moment of traffic, For nodes At the moment The node pressure, For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump; For the moment The water storage capacity of the water source; Taking multiple costs as the objectives to be minimized, a multi-objective optimization function is constructed, which is defined as follows: in, is the energy cost, is the water production cost, Punishment for lack of water, is the start-stop cost, are target weight coefficients respectively, and their cumulative sum is 1; In energy costs In the calculation, the number of pumps is p , and their corresponding numbers are 1, 2... p , For the moment The electricity price during the period, For pumps p efficiency, For pumps p At the moment The lift, is the time step; The cost of water production In the calculation, the amount of water source is k , and their corresponding numbers are 1, 2... k , For water source k The unit water supply cost, is the kth water source at time The water intake range is [0, ], For water source k At the moment Available water supply capacity; Punishment for lack of water Calculating, is the water shortage penalty coefficient, For the moment the amount of water shortage; Start-stop costs Calculating, For pumps p Start-up and shutdown costs; For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump; In the default decision variables Under these conditions, it satisfies the following multiple constraints: (1) Constraints on demand and water shortage: in, , ; For water supply goals At the moment Forecasted water consumption; For the moment The planned supply of the pump at the time is in the range of [0, ]; For the moment the amount of water shortage; (2) Priority water shortage upper limit constraint: in, For the moment the amount of water shortage; For water supply goals The tolerable water shortage ratio corresponding to the water supply priority is [0, 1]. For water supply goals At the moment Forecasted water consumption; (3) Node traffic balance constraints in, 、 are the collections of inflow and outflow pipe sections respectively; is the set of water sources connected to the node, is the water supply target set for taking water at the corresponding node; For pipe section e At the moment of traffic, is the kth water source at time The amount of water taken, For the moment The planned pump supply volume at the time (4) Dynamic constraints on water volume in water storage tanks in, For the moment The amount of water stored in the water source, 、 are the upper and lower limits of the water storage volume respectively; (5) Pump constraints and start-stop logic in, For pumps p of traffic, For pumps p At the moment The switch status of the pump is 1 for starting the pump or 0 for stopping the pump; For pumps p Rated maximum flow rate; (6) Pressure and pipe section hydraulic approximation constraints Using linear constraints, let in, , is the linearized friction coefficient, which is obtained from the derivative of the reference flow point; 、 Node n Maximum safety pressure and minimum pressure; For pipe section e At the moment of traffic, 、 Node 、 m At the moment Nodal pressure; (7) Water source capacity constraints in, For water source k At the moment Available water supply capacity, is the kth water source at time water intake; Based on constraints (1)-constraint (7), the decision variables are minimized to obtain the scheduling solution, which is defined as: .

10. A multi-objective water supply scheduling system driven by big data, characterized in that: It includes: A data collection unit is used to collect location information, historical water consumption data, water supply network operation data, meteorological environment data, and water supply rule parameters of multiple water supply targets within the water supply scheduling area, and to construct a multi-source data set corresponding to the multiple water supply targets; The data preprocessing unit is used to fill missing values, identify and repair outliers in multi-source data according to preset conditions to complete the preprocessing of multi-source data, and then generate a spatiotemporal correlation diagram based on the hydraulic distance between the water supply targets, the historical water consumption correlation and the pressure coupling relationship; A water use prediction unit is used to predict the water demand of each water supply target in the future time period based on preprocessed multi-source data, combined with time series, meteorological and / or water supply target priorities as input features, through a graph spatiotemporal prediction model to obtain a first prediction result, and then combine the seasonal statistical model and the gradient boosting regression model to output a second prediction result and a third prediction result respectively, and then obtain the predicted water demand of each water supply target in the future time period through error adaptive weighted integration, and construct a demand uncertainty scenario set and / or prediction confidence interval based on the residual sample; A water supply planning unit is used to allocate corresponding margins to the predicted water demands corresponding to the preset priorities of multiple water supply targets, and then solve the source-water supply target linear allocation model to obtain an initial water supply allocation plan corresponding to the multiple water supply targets; The scheduling optimization unit is used to construct a multi-objective optimization function within a preset rolling optimization time domain, using multiple costs as the objectives to be minimized. Then, the function is solved by combining preset decision variables and satisfying multiple preset constraints to obtain a scheduling solution. A water supply execution unit is used to execute the scheduling plan according to a preset rolling cycle to supply water to multiple water supply targets within the water supply scheduling area; The water supply monitoring unit is used to monitor the water demand of multiple water supply targets in real time, and then calculate the deviation rate between the two in combination with the predicted water demand. When the deviation rate exceeds the trigger threshold corresponding to the water supply target priority, the scheduling optimization unit is called to re-optimize the scheduling plan for the remaining period of the preset rolling cycle, obtain the updated scheduling decision and send it to the water supply execution unit for execution.

Images