A model algorithm-based multi-water plant parallel operation optimization scheduling method

Abstract

The application discloses a kind of based on model algorithm's multiple water plant parallel operation optimization scheduling method, first, the water quantity comprehensive prediction model of city water supply system is established and the combination prediction method of adaptive filtering prediction and BP neural network prediction is used to solve model, then, the water supply system hydraulic model is established and FMGA is used to solve model automatic checking problem;Finally, two-stage optimization scheduling model is established and AE is used to solve optimization scheduling problem, first, the best water supply and water supply pressure of each water plant are solved by establishing primary scheduling model, and then the best combination mode of pump house pump is determined by establishing secondary model, so as to reduce the energy consumption of city water supply system, improve the water supply efficiency of multiple water plant parallel operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of water supply, and specifically refers to an optimal scheduling method for parallel operation of multiple water plants based on a model algorithm. Background Art

[0002] The water supply system is an important part of urban infrastructure. At present, the water supply scheduling management in many cities in China is still manual scheduling based on experience. Although this traditional scheduling method can meet the urban water supply demand to a certain extent, for a complex water supply system with parallel water supply from multiple water plants, the water supply scheduling method based on manual experience is difficult to adapt to the change of water volume and reasonably distribute water volume. The water pumps in the water supply system are difficult to continuously operate in the efficient range, resulting in low working efficiency of the water pumps, high water supply costs and large energy consumption.

[0003] With the acceleration of the urbanization process and the continuous promotion of the integration of urban and rural water supply, the scale of the water supply system is getting larger and the complexity is getting higher. Continuing to use the traditional scheduling method will not meet the needs of the current ecological civilization construction and the development of the information society. Existing optimal scheduling for parallel water supply from multiple water plants does not consider the factors of energy conservation and consumption reduction. Facing the growing demand of the urban water transmission system for low carbon emissions, it is necessary to establish a new optimal scheduling method to reduce carbon emissions. Summary of the Invention

[0004] The purpose of the present invention is to provide an optimal scheduling method for parallel operation of multiple water plants based on a model algorithm, which can reduce the energy consumption of the water supply system and improve the water supply efficiency of the parallel operation of multiple water plants.

[0005] To achieve the above purpose, an optimal scheduling method for parallel water supply from multiple water plants based on a model algorithm of the present invention is based on an online detection water use data platform - SCADA system and a water supply network information system recording system - GIS system, and includes the following steps:

[0006] S1, predicting the urban water consumption;

[0007] S2, establishing and automatically checking the water supply system model;

[0008] S3, solving the two-level optimal scheduling model by an algorithm to determine the optimal pump group scheme.

[0009] As a further scheme of the present invention: the step S1 includes the following steps:

[0010] S11, collecting, sorting and analyzing the water use data on the SCADA system;

[0011] S12, predicting the water consumption by using the adaptive filtering prediction method;

[0012] S13. Use the BP neural network prediction method to predict the water consumption;

[0013] S14. Combine the adaptive filtering prediction and the BP neural network prediction with weights to establish a prediction model;

[0014] S15. Check and evaluate the established model.

[0015] As a further solution of the present invention: The step S2 includes the following steps:

[0016] S21. Collect the node attribute information of the pipe network segments, the pipe network operation data, the pump operation data, the detection data of the pipe network detection points and the measured data on the pipe network GIS system; The measured data includes pipe friction test, pump performance test, valve performance test, water use pattern test and pressure flow test;

[0017] S22. Input the collected data into the modeling software to establish a pipeline topology model;

[0018] S23. Establish the basic equations for pipe network calculation and conduct preliminary hydraulic calculation of the pipe network;

[0019] S24. Use FMGA to automatically check the established model. FMGA is a combination of the factorization machine FM and the genetic algorithm GA;

[0020] S25. Establish a pipe network leakage model for dynamic simulation, combine it with the SCADA system to locate the leakage and determine the influence range.

[0021] As a further solution of the present invention: The step S24 includes the following steps:

[0022] S241. Check the water output flow and pressure of each water plant;

[0023] S242. Adjust the pump characteristic curve, the friction coefficient and the flow coefficient of the main pipe;

[0024] S243. Check the pressure and flow of the monitoring points in the pipe network;

[0025] S244. Use FMGA to automatically check the pipe network system model.

[0026] As a further solution of the present invention: The step S3 includes the following steps:

[0027] S31. Establish a two-level optimal scheduling model;

[0028] S32. Solve the model with the alpha evolutionary algorithm;

[0029] S33. Determine the pump combination mode and the optimal scheduling scheme;

[0030] As a further solution of the present invention: Step S31 includes:

[0031] S311. The primary optimization scheduling objective function and constraint conditions;

[0032] (1) The objective function is that the total operating cost of multi-plant water supply can be decomposed into water intake and water conveyance costs, water treatment costs of water plants, and secondary pumping station lifting costs;

[0033] (2) The constraint conditions refer to the mathematical expressions that control or restrict the operation of the system, including pipe network hydraulic balance, water supply volume of water plants, water supply pressure of secondary pumping stations, and user water pressure requirements;

[0034] S312. The secondary optimization scheduling objective function and constraint conditions;

[0035] (1) The objective function is the overall operating power of the pumping station;

[0036] (2) The constraint conditions are the pump speed ratio and the pump discharge flow;

[0037] S313. Establish an optimization mathematical model, convert the multi-objectives to be controlled in the optimization scheduling into a single-objective problem, and solve the model by establishing the objective function and constraint conditions.

[0038] As a further solution of the present invention: Step S32 includes:

[0039] S321. Coding design:

[0040] First level: Water volume and water pressure distribution, real number coding [r1, r2,... r n , satisfying ∑r i = 1;

[0041] Second level: Pump scheduling, binary coding [s1, s2,..., s m , where s i = 1 indicates that pump i is operating;

[0042] S322. Fitness function design:

[0043] First-level fitness:

[0044] In the formula: c i —— Unit water supply cost of water plant i;

[0045] r i —— Real number coding of water volume and water pressure distribution;

[0046] Q 总 —— Total water demand;

[0047] C i—— Penalty term, such as insufficient water supply;

[0048] The second - level fitness:

[0049] In the formula: p j —— The power of pump j;

[0050] η j —— The efficiency of pump j;

[0051] S j,t —— The state of pump j at time period t;

[0052] The comprehensive fitness: f 总 = ω1f max + ω2f min ;

[0053] In the formula, ω1 is the weight value emphasizing cost, and ω2 is the weight value emphasizing energy consumption;

[0054] S323. Population initialization:

[0055] The first level: Randomly generate a real - number vector that satisfies ∑r i = 1; ​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​

[0067] Phase 1: Fix the secondary pump scheduling and optimize the water source allocation;

[0068] Water supply capacity constraint: (Water supply capacity constraint);

[0069] Phase 2: Fix the primary water source allocation and optimize the pump scheduling;

[0070] Constraints: (Grid load limit);

[0071] Iteration: Repeat Phase 1 and Phase 2 until the fitness converges;

[0072] S327. Elite retention and population update:

[0073] Retain the top 5% of the water source allocation schemes and the corresponding pump scheduling for each generation, and replace the 20% of the individuals with the lowest fitness in the old population with new individuals;

[0074] S327: Termination condition:

[0075] The number of iterations reaches 200 generations, the fitness change is less than 1% for 10 consecutive generations, and the water supply reliability reaches over 95%.

[0076] As a further solution of the present invention: The step S33 includes:

[0077] S331. Fitting the pump characteristic curve;

[0078] S332. Comprehensive analysis of the pump operation efficiency;

[0079] S333. Solving the optimization scheduling model;

[0080] S334. Energy-saving analysis of the pump operation conditions and energy-saving analysis of the scheduling scheme.

[0081] Compared with the prior art, the beneficial effects of the present invention are as follows: From the perspectives of model theory and model application, the developed water supply system simulation model of the present invention is more applicable to actual engineering. Through the water consumption prediction model, the water supply network hydraulic model, and the two-stage optimization scheduling model, the water volume can be reasonably allocated, saving water resources; the pump can continuously operate in the high-efficiency range, solving the problems of high water supply cost and large energy consumption. Brief description of the drawings

[0082] Figure 1 It is the flow chart of the multi-water plant parallel operation optimization scheduling method based on the model algorithm of the present invention.

[0083] Figure 2 It is the flow chart of the multi-water plant parallel water supply water consumption prediction of the present invention.

[0084] Figure 3 It is the flow chart for modeling the hydraulic model of parallel water supply from multiple water plants in the present invention. Specific embodiments

[0085] The present invention will be further described below with reference to the accompanying drawings.

[0086] As Figure 1 shown, an optimized scheduling method for parallel operation of multiple water plants based on a model algorithm. This method is based on an online water consumption data detection platform - the SCADA system and a water supply network information system recording system - the GIS system, and specifically includes the following steps:

[0087] S1. Predict the urban water consumption;

[0088] S2. Establish and automatically check the water supply system model;

[0089] S3. Solve the two-level optimized scheduling model by an algorithm to determine the optimized pump group scheme.

[0090] Step S1 includes the following steps:

[0091] S11. Collect, sort out and analyze the water consumption data on the SCADA system;

[0092] S12. Predict the water consumption by using the adaptive filtering prediction method, as Figure 2 shown;

[0093] When predicting the water consumption by the adaptive filtering method, first select a certain number of time series and determine the number and values of weights. In this example, the water consumption for 24 hours will be predicted, that is, there are 24 weights, and the value of each initial weight is 1 / 24; then use the 24 weights and the water consumption data for 24 hours on the first day to calculate the water consumption for the 25th hour, and calculate the 25th weight according to the weight adjustment formula and the predicted value and the true value obtained; then use the latter 24 weights and the new 24 water consumption data from the 2nd hour to the 25th hour to calculate the water consumption for the 26th hour, and so on. Finally, use the water consumption for 168 hours to continuously calculate new weights, set the number of iterations until the 24 weights basically tend to be stable, and use the latest set of weights to calculate the hourly water consumption for 24 hours on the 8th day. The adaptive filtering prediction model is:

[0094]

[0095] In the formula: y’ t+1 ——The predicted value for the (t + 1)th hour;

[0096] ω i ——The weight of the observed value for the (t - i + 1)th hour;

[0097] yt-i+1 —— Observation value at the (t - i + 1)-th hour;

[0098] N—— The number of weights, i = 1, 2,..., N, t = N, N + 1,..., n, where n is the number of sequence data.

[0099] The weight adjustment formula is:

[0100] ω’ i = ω i + 2ke t+1 y t-i+1

[0101] In the formula: ω’ i —— The adjusted i-th weight;

[0102] ω i —— The i-th weight before adjustment.

[0103] First, initialize the weights, usually set to equal values, ω i = 1 / 7. Then calculate the predicted values for each period based on the current weights, and calculate the error: e t = y t+1 - y’ t+1 , and update the weights: Finally, repeat the iteration until the error converges.

[0104] According to the model formula of the above adaptive filtering method, using the water consumption of a certain city for 7 days, that is, 168 hours, as the training data, predict the water consumption for 24 hours on the eighth day. Use Fortran to write a program for prediction, and the prediction results are shown in Table 1 below:

[0105] Table 1. Adaptive Filtering Prediction Result Table

[0106]

[0107]

[0108] It can be seen from the table that the prediction result has a large error and the prediction effect is not good.

[0109] S13, use the BP neural network prediction method to predict the water consumption;

[0110] When using a BP neural network, the factors affecting water consumption need to be decomposed, quantified, and then added to the model to predict water consumption. In short-term water consumption prediction, natural factors include the impact of temperature changes and climate dryness on the hourly water consumption of the city on the day, while social factors include water usage habits during holidays. When predicting hourly water consumption, three main influencing factors are selected, namely temperature, sunny or cloudy weather, and date. Before writing the Matlab function, the influencing factors including temperature, sunny or cloudy weather, and date need to be quantified. The quantification mechanism for the three influencing factors is to quantify the influencing factors by observing the actual impact of the daily temperature, sunny or cloudy conditions, and weekdays on the hourly water consumption in historical data. Among them, the temperature quantization value is quantified according to the proportion of water consumption in the corresponding temperature range in historical data; the sunny or cloudy quantization value is quantified based on the common sense that water consumption is more on sunny days and less on rainy days; the week quantization value is quantified according to the ratio of the average water consumption in the past week to the water consumption on Monday. The quantization results are shown in Table 2 below:

[0111] Table 2. Quantification Result Table

[0112] 0-5℃ 0.15～0.25 Sunny 2.25 One 1.00 5-10℃ 0.25～0.35 Sunny turning to cloudy 2 Two 1.017 10-15℃ 0.35～0.45 Cloudy 1.75 Three 1.01 15-20℃ 0.45～0.55 Cloudy turning to overcast 1.5 Four 1.014 20-25℃ 0.55～0.65 Overcast 1.25 Five 1.032 25-30℃ 0.65～0.75 Light rain 1 Six 1.036 30-35℃ 0.75～0.85 Moderate rain 0.75 Seven 1.040 35-40℃ 0.85～0.95 Heavy rain 0.5

[0113] The BP neural network prediction model is as follows:

[0114]

[0115] O pi = f(net pj )

[0116] f(x) = 1 / (1 + e -x

[0117] In the formula: net pj —— The total input of unit j;

[0118] W ij —— The weight between neurons i and j;

[0119] O pi —— The input of unit i;

[0120] f —— Activation function;

[0121] N —— Number of processing units;

[0122] P —— Number of training samples.

[0123] The BP neural network also has the function of automatically adjusting the weight W p according to the prediction error E ij , and its formula is:

[0124]

[0125] W ij (t + 1)= w ij (t)+ ηδ pj o pi

[0126] Where: δ pj —— Training error;

[0127] t —— Number of learning times;

[0128] η —— Learning factor.

[0129] According to the above model formula, a Matlab function program can be written to perform model solution calculations. The BP network structure has 3 input units, 2 hidden layers and 1 output unit. The 3 input units are holiday situation, sunny or cloudy situation and temperature situation respectively, and the output unit is the hourly water consumption; the first hidden layer has 7 nodes, and the second hidden layer has 5 nodes. The network learning rate is 0.01, the momentum factor is 0.08, the network training times is 40,000 times, the training error and target error during synthesis are 0.001 and 0.00005 respectively, and the synthesis threshold is 0.01. The activation function of the hidden layer is Sigmoid, the activation function of the output layer is purelin, the network training function is traingdm, and the network learning function is learngdm. Use the data of the first 6 days as training samples, the data of the 7th day as test samples, and predict the data of the 8th day.

[0130] First, prepare the water consumption data after normalization processing for the first 6 days as the data for model training; then use the prepared data to train the model, adjust the weights of the model until the weights are stable, and use the water consumption data of the 7th day to test the model; finally, use the trained network to predict the hourly water consumption on the 8th day. Write a Matlab program according to the above principle, and the specific prediction results are shown in Table 3 below:

[0131] Table 3. Prediction results of BP neural network

[0132]

[0133] It can be seen from the table that the prediction result error is large and the prediction effect is not good.

[0134] S14, perform weighted combination of adaptive filtering prediction and BP neural network prediction to establish a prediction model;

[0135] By establishing a combined weight coefficient optimization model with the sum of squared relative prediction errors as the objective function and the combined weights as decision variables, the optimal weights of the combined model are solved. The established combined weight coefficient optimization model is:

[0136]

[0137] The constraint conditions are as follows:

[0138] In the formula, y--Combined prediction result

[0139] w i --Weight of the i-th single prediction result

[0140] y i --The i-th single prediction result

[0141] n--Number of single prediction models

[0142] When using the combined model to predict the urban hourly water consumption, when combining the single model prediction results, the combined weight coefficient period is 24 hours. Therefore, taking 24 combined weight coefficients as a period, finally 24 groups of optimal combined weight coefficients in one period are obtained. On this basis, the single models are combined.

[0143] In the process of solving the objective function, the weight coefficient can be solved by the Lagrange method for solving the maximum and minimum values of multivariate functions, that is, a Matlab function can be written to solve the weight coefficient. Taking 7 o'clock and 8 o'clock as examples, the weight coefficients are 0.2453 and 0.7547, 0.6593 and 0.3407 respectively. The specific coefficients, combined prediction results and corresponding prediction errors are as follows in the table:

[0144] Table 4. Combined prediction result table

[0145]

[0146] It can be seen from the table that the prediction result error is controlled below 15%, and the prediction effect is good.

[0147] S15. Check and evaluate the established model:

[0148] The calculation formula of the model evaluation index is as follows:

[0149] (1) Mean absolute error:

[0150] (2) Mean relative error:

[0151] (3) Precision variance: A t =1 - |f t - q t | / q t

[0152]

[0153] (4) Model effectiveness: S = E(A t )(1 - σ(At )

[0154] Where:

[0155] n——Prediction time;

[0156] f t ——Water consumption predicted at the t-th hour in the model;

[0157] q t ——Actual water consumption at the t-th hour;

[0158] The MAE of the combined prediction model is 1.3185, the MAPE is 4.49, the precision variance is 0.0422, and the model effectiveness is 0.9148. The results show that the combined prediction model has better accuracy and stability.

[0159] Step S2 includes the following steps:

[0160] S21, Collect the node attribute information of pipeline segments in the pipe network GIS system, the pipe network operation data, the pump operation data, the detection data of pipe network detection points, and the measured data; among which the measured data includes pipeline friction test, pump performance test, valve performance test, water use pattern test, and pressure flow test.

[0161] S22, Input the collected data into the modeling software to establish a pipeline topology model, as Figure 3 shown;

[0162] S23, Establish the basic equations for pipe network calculation and conduct preliminary hydraulic calculation of the pipe network;

[0163] S24, Automatically check the established model.

[0164] 1. The first step in modeling is to collect pipe network data and input the pipe network data in the GIS system into the modeling software. The modeling software generally has a dedicated docking module and does not require manual input.

[0165] 2. The pipe network topology structure after preliminary data import cannot be directly used for hydraulic calculation. First, simplify and organize the node pipeline segments.

[0166] 3. Establish the pipe network hydraulic calculation equations, including: continuity equation, pressure drop equation, energy equation, and pump characteristic curve.

[0167] (1) Node flow equation set

[0168] Any node j in the pipe network should conform to the law of conservation of mass, ensuring that the total inflow of all nodes is equal to the total outflow of all nodes.联立管网中所有节点的流量方程得到节点流量方程组。节点流量方程可以表示为:

[0169]

[0170] Among them, q is the flow rate of pipe segment i, Q is the flow rate of node j, s is the associated set of node j, N is the total number of nodes in the pipe network, represents the directed summation of the pipe segments in the associated set of node j, the pipe segment flow rate flowing out of the node is positive, and the flow rate flowing into the node is negative.

[0171] (2) Pipe segment pressure drop equations

[0172] The head difference between the two ends of any pipe segment i in the pipe network is equal to the pressure drop of this pipe segment. By联立the pressure drop equations of all pipe segments in the pipe network, the pipe segment pressure drop equations can be obtained. It can be expressed as:

[0173]

[0174] Among them, F i and T i are respectively the starting node number and the ending node number of pipe segment i, and are respectively the head of the starting node and the head of the ending node of pipe segment i, h i is the pressure drop of pipe segment i, and M is the total number of pipe segments in the pipe network.

[0175] (3) Loop energy equations

[0176] In the pipe network model, the closed pipe segments form loops. It is stipulated that the values of the flow rate and the head loss in the loop are positive in the clockwise direction and negative in the counterclockwise direction. Then, the sum of the head loss values of all loops should be 0. By联立the energy equations of all loops in the pipe network, the loop energy equations can be obtained. Loop energy equation:

[0177]

[0178] Among them, k represents the numbers of each loop in the pipe network, and i represents the numbers of each pipe segment in the pipe network.

[0179] 4. Conduct pipe network operating condition calculations. After the calculations, the node water pressure, pipe segment flow rate, head loss, pipe network water quality, the outlet pressure and flow rate of each water treatment plant, and other pipe network operating condition information are obtained.

[0180] (1) Head loss calculation

[0181] The head includes the elevation head, the pressure head, and the velocity head. Since the velocity head is much smaller than the elevation head and the pressure head, the velocity head can be ignored in the simplified calculation. The formula for the head is:

[0182]

[0183] The head loss is usually calculated by the Chezy formula. For full flow in circular pipes, the Darcy-Weisbach formula is generally used for calculation. To perform more accurate hydraulic calculations under different conditions and flow regimes, the Colebrook-White formula, Hazen-Williams formula, Manning formula, and Pavlovsky formula are commonly used.

[0184] For the pipe head loss caused by pipe wall friction, there are the following three calculation formulas:

[0185] ① Darcy-Weisbach formula

[0186] The Darcy-Weisbach formula and the Chezy formula are the most basic and classic pipe and channel hydraulic calculation formulas, applicable to all flow regimes. The Darcy-Weisbach formula is:

[0187]

[0188] Among them, h is the head loss along the path, with the unit of m;

[0189] v is the average flow velocity of the cross-section of the flowing water, with the unit of m / s;

[0190] C is the Chezy coefficient;

[0191] D is the diameter of the pipe section, with the unit of m;

[0192] g is the acceleration due to gravity, with the unit of m / s 2 ;

[0193] l is the pipe length, with the unit of m;

[0194] λ is the friction coefficient along the path. Substituting into the Darcy-Weisbach formula gives the formula:

[0195]

[0196] Among them, R is the hydraulic radius of the cross-section of the flowing water, with the unit of m.

[0197] ② Hazen-Williams formula

[0198] The Hazen-Williams formula is usually used for hydraulic calculations in water supply pipe networks. The formula is:

[0199]

[0200] Among them, q is the flow rate of the pipe section, with the unit of m 3 / s, and C is the Hazen-Williams coefficient.

[0201] ③ Manning formula

[0202] The Manning formula is applicable to hydraulic calculations of open channels and rough non-full pipe flow pipes. The formula is:

[0203]

[0204] Among them, n is the Manning roughness coefficient.

[0205] In most cases, the hydraulic calculation considers the average flow velocity of the pipeline in the cross-section, and the calculation formula is:

[0206]

[0207] Among them, v is the average flow velocity; S is the cross-sectional area of the pipeline; D is the pipeline diameter; q is the pipeline flow rate.

[0208] 5. Automatic calibration of the pipe network model. The automatic calibration of the hydraulic model belongs to the parameter optimization problem. The main task is to automatically adjust a series of model parameters so that the model calculation value is in good agreement with the measured value. The mathematical description is:

[0209] Solve: X = (f, m, s); Minimize: the error F(X). [[ID=2|0]]

[0210] Among them, f is the resistance adjustment coefficient of the pipeline; m is the water consumption adjustment coefficient of the node at time t; s is the operating state of the component (pipeline, valve, pump) at time t. Generally, the genetic algorithm is used for optimization and solution.

[0211] FMGA is a combination of the factorization machine (FM) and the genetic algorithm (GA), which uses the genetic algorithm to optimize the hyperparameters of the factorization machine. FM captures the interaction relationship between features through a low-dimensional embedding vector, and solves the problems of dimensionality disaster and sparsity. Its core formula is:

[0212]

[0213] GA is a global optimization algorithm that simulates the natural selection process (crossover, mutation, selection) to find the optimal solution. Combining the two algorithms can achieve complex parameter tuning. When there are many hyperparameters of FM (such as multiple groups of embedding dimensions), GA can automatically explore the optimal combination to achieve the purpose of automatic calibration. Use FMGA to optimize the embedding dimension k and regularization coefficient λ of FM:

[0214] Population initialization: Randomly generate multiple groups of (k, λ) parameter combinations.

[0215] Fitness evaluation: Train the FM model for each group of parameters and calculate the AUC value on the validation set.

[0216] Genetic operation: Select individuals with high AUC, cross to generate new parameters, and randomly mutate some parameters.

[0217] Iterative optimization: Repeat the steps of fitness evaluation and genetic operation, and retain the optimal parameter combination. [[ID=|44]]

[0218] S25. Establish a pipeline network leakage model for dynamic simulation, combine it with the SCADA system for leakage location, and determine the influence range.

[0219] Step S24 includes the following steps:

[0220] S241. Check the water output flow rate and pressure of each water plant.

[0221] S242. Adjust the pump characteristic curve, the friction coefficient and the flow coefficient of the main pipeline.

[0222] S243. Check the pressure and flow rate of the monitoring points in the pipeline network.

[0223] S244. Automatically check the pipeline network system model using FMGA.

[0224] Step S3 includes the following steps:

[0225] S31. Establish a two-level optimal scheduling model;

[0226] S32. Solve the model using the alpha evolutionary algorithm;

[0227] S33. Determine the pump combination method and the optimal scheduling plan.

[0228] Step S31 includes:

[0229] S311. The first-level optimal scheduling objective function and constraint conditions;

[0230] (1) The objective function is that the total operation cost of multi-plant water supply can be decomposed into the water intake and water conveyance cost, the water treatment cost of the water plant, and the secondary pump house lifting cost.

[0231] The specific formula is:

[0232]

[0233] In the formula, C i1 is the water intake and water conveyance cost of water plant i; C i2 is the water treatment cost of water plant i; C i3 [[ID=5L]]is the secondary pump house lifting cost of water plant i.

[0234] (2) The constraint conditions refer to the mathematical expressions that control or restrict the operation of the system, including the pipeline network hydraulic balance, the water supply volume of the water plant, the water supply pressure of the secondary pump house, and the water pressure requirements of users. The specific formula is:

[0235] Pipeline network hydraulic balance

[0236] For each node j

[0237] For each pipe segment i

[0238] where, q k --- the flow rate of pipe segment k;

[0239] Q j --- the water consumption of node j;

[0240] --- the total water heads of the upstream and downstream nodes of pipe segment l;

[0241] h i --- the head loss of pipe segment l;

[0242] The water supply of the water plant Q i :

[0243] For each water plant i, Q min ≤Q i ≤Q max ;

[0244] The water supply pressure H of the secondary pumping station i :

[0245] For each water plant i, the secondary pumping station H imin ≤H i ≤H imax ;

[0246] The water pressure of users H j :

[0247] H jmin ≤H j ≤H jmax .

[0248] S312. Secondary optimal scheduling objective function and constraint conditions;

[0249] (1) The objective function is the overall operating power of the pumping station;

[0250] The power target of the constant-speed pump:

[0251] The power target of the variable-speed pump:

[0252] where, Q--- the water flow rate of the water pump;

[0253] Q i --- the water flow rate of the i-th water pump;

[0254] S i --- the flow rate speed ratio of the i-th water pump;

[0255] w i --- the water pump startup decision variable, 0 means not put into operation, 1 means put into operation;

[0256] l---Number of constant-speed pumps;

[0257] n-l---Number of variable-speed pumps;

[0258] a i0 、a i1 、a i2 ---Fitting parameters of the flow-power curve of the i-th pump.

[0259] (2) The constraint conditions are the speed ratio of the pump and the discharge flow of the pump.

[0260] Speed ratio constraint: S imin ≤S i ≤S imax ;

[0261] Total flow constraint: H1 = H2 =.... = H e

[0262] In the formula, H e ---Optimal pressure of the pumping station obtained in the primary optimization dispatch;

[0263] Q e ---Optimal water supply of the pumping station obtained in the primary optimization dispatch;

[0264] Single-pump flow constraint: Q imin ≤Q i ≤Q imax .

[0265] S313. Establish an optimization mathematical model, convert the multi-objectives to be controlled in the optimization dispatch into a single-objective problem, and solve the model by establishing the objective function and constraint conditions.

[0266] Primary dispatch mathematical model:

[0267]

[0268] Secondary dispatch mathematical model:

[0269]

[0270] In step S32, the Alpha evolution (AE) algorithm is a novel evolutionary algorithm that updates the solution using an alpha operator with adaptive basis vectors and random and adaptive step sizes. First, candidate solutions are selected to construct an evolutionary matrix. The population state is estimated through diagonal or weighted operations on the evolutionary matrix. To enhance the relevance of the estimation in each generation, the estimation results from two evolutionary paths are accumulated to achieve adaptive basis vectors. Second, composite differential operations are used to construct an adaptive step size to estimate the problem gradient, which accelerates the convergence of acoustic emissions. Finally, the attenuation factor alpha adaptively adjusts the generated random step size according to the search space to balance exploration and exploitation. Specifically, the following steps are included:

[0271] S321. Coding Design:

[0272] Level 1: Water volume and pressure distribution, real number encoding [r1,r2,…r] n ], satisfying ∑r i =1.

[0273] Level 2: Pump scheduling, binary encoding [s1, s2, ..., s m ], where s i =1 indicates that pump i is running.

[0274] S322. Fitness Function Design:

[0275] First level fitness:

[0276] In the formula: c i —Unit water supply cost of water plant i;

[0277] r i —Real number encoding of water volume and pressure distribution;

[0278] Q 总 —Total water demand;

[0279] C i —Penalties, such as insufficient water supply;

[0280] Second-level fitness:

[0281] In the formula: p j —The power of water pump j;

[0282] η j —The efficiency of water pump j;

[0283] S j,t —The state of water pump j in time period t;

[0284] Overall fitness: f 总 =ω1fmax +ω2f min ;

[0285] In the formula, ω1 is the weight value that focuses on cost, and ω2 is the weight value that focuses on energy consumption.

[0286] S323. Population Initialization:

[0287] Level 1: Randomly generate samples satisfying ∑r i A real vector of numbers equal to 1;

[0288] Level 2: For each water source allocation scheme, generate a water pump schedule that satisfies the water balance.

[0289] S324. Adaptive selection strategy:

[0290] Tournament selection based on congestion: T decreases from 100 to 0.1 with each iteration to avoid premature convergence.

[0291] S325. Crossover and Mutation:

[0292] Level 1: Real Number Encoding

[0293] Crossover: Polynomial crossover (preserving ∑r) i =1);

[0294] Variation: Gaussian variation, step size σ = 0.1 × (r max -r min );

[0295] Level 2: Binary Encoding

[0296] Crossover: Single-point crossover (exchanging the start / stop sequences of two individual water pumps);

[0297] Mutation: Bit-by-bit flip (probability P) m =0.05);

[0298] S326. Two-stage alternating optimization:

[0299] Phase 1: Fix the scheduling of the second-level water pumps and optimize water resource allocation;

[0300] Water supply capacity constraints: (Water supply capacity constraints);

[0301] Phase Two: Fix the primary water source allocation and optimize pump scheduling;

[0302] constraint: (Grid load limitation);

[0303] Iteration: Repeat phase one and phase two until the fitness converges.

[0304] S327. Elite Preservation and Population Renewal:

[0305] Each generation retains the top 5% of the water allocation schemes and corresponding water pump scheduling, replacing the 20% of individuals with the lowest fitness in the old population with new individuals.

[0306] S327: Termination Condition:

[0307] The number of iterations reaches 200 generations, the fitness change is less than 1% for 10 consecutive generations, and the water supply reliability reaches more than 95% (preset threshold).

[0308] Step S33 includes the following steps:

[0309] S331. Pump characteristic curve fitting;

[0310] S332. Comprehensive analysis of water pump operating efficiency;

[0311] S333. Solving the optimal scheduling model;

[0312] S334. Energy-saving analysis of water pump operation conditions and energy-saving analysis of dispatching schemes.

[0313] Based on the water consumption prediction results in step S1, the AE algorithm is used to solve the established first-level optimization scheduling model. Taking three water plants as an example, namely water plant A, water plant B, and water plant C, the optimized outflow rate and pressure of each water plant are shown in Table 5 below:

[0314] Table 5. Optimized Dispatch Schemes for Each Water Plant

[0315]

[0316] The effluent flow rate and pressure of the water treatment plant were solved based on the first-level optimal scheduling model. Then, the AE algorithm was used to solve the established second-level optimal scheduling model. The daily operating status of the pumps in water treatment plants A, B, and C after optimization is shown in Table 6 below:

[0317] Table 6. Optimized Pump Unit Start-up Methods

[0318]

[0319] Table 7 shows the energy-saving analysis of the pumps in water plants A, B, and C after secondary optimization. As can be seen, the water supply volume of each pump in the pumping station differs from that before optimization, and the water supply tasks have been re-divided. Compared with experience-based scheduling, the optimized pumping station operation plan can save approximately 7724 kWh of electricity per day, with an energy efficiency of approximately 20.13%, demonstrating good economic benefits.

[0320] Table 7. Energy Saving Analysis of Each Water Plant After Optimized Scheduling

[0321]

Claims

1. A method for optimized scheduling of parallel operation of multiple water plants based on model algorithms, characterized in that, This method, based on the online water usage data platform—SCADA system and the water supply network information system recording system—GIS system, includes the following steps: S1, predicts urban water consumption; S2, Water supply system model establishment and automatic verification; S3, the algorithm solves the two-level optimization scheduling model to determine the optimal scheme for the pump group; Step S1 includes the following steps: S11, collect, organize and analyze water usage data from the SCADA system; S12 uses an adaptive filtration prediction method to predict water consumption; S13, uses the BP neural network prediction method to predict water consumption; S14, a prediction model is established by weighting and combining adaptive filtering prediction and BP neural network prediction; S15, Establish a good model for verification and evaluation; Step S2 includes the following steps: S21, collect the attribute information of pipeline segment nodes, pipeline operation data, pump operation data, pipeline inspection point detection data and measured data from the pipeline GIS system; among which the measured data includes pipeline friction test, water pump performance test, valve performance test, water use mode test and pressure and flow test; S22, Input the collected data into the modeling software to build a pipeline topology model; S23, Establish the basic equations for pipeline network calculation and perform preliminary hydraulic calculations for the pipeline network; S24. The established model is automatically verified using FMGA, which is a combination of factorization machine (FM) and genetic algorithm (GA). S25. Establish a pipeline leakage model for dynamic simulation, and combine it with the SCADA system to locate the leakage and determine the scope of impact. Step S3 includes the following steps: S31, Establish a two-level optimized scheduling model; S32, the model is solved using the Alpha Evolutionary Algorithm; S33, Determine the pump combination method and optimize the scheduling scheme; Step S32 includes: S321. Coding Design: Level 1: Water volume and pressure distribution, real number encoding ,satisfy ; Level 2: Pump scheduling, binary encoding ,in This indicates that pump i is running; S322. Fitness Function Design: First level fitness: ; In the formula: —Unit water supply cost of water plant i; —Real number encoding of water volume and pressure distribution; —Total water demand; —Penalty items; Second-level fitness: ; In the formula: —The power of water pump j; —The efficiency of water pump j; —The state of water pump j in time period t; Overall adaptability: ; In the formula: It is a weighted value that emphasizes cost. It is a weighted value that emphasizes energy consumption; S323. Population Initialization: Level 1: Randomly generated to satisfy A real vector; Level 2: For each water source allocation scheme, generate a water pump schedule that satisfies the water balance. S324. Adaptive selection strategy: Tournament selection based on congestion: T decreases from 100 to 0.1 with each iteration to avoid premature convergence; S325. Crossover and Mutation: Level 1: Real Number Encoding Crossover: Polynomial crossover, maintaining ; Mutation: Gaussian mutation, step size ; Level 2: Binary Encoding Crossover: Single-point crossover, swapping the start / stop sequences of two individual water pumps; Mutation: Bit-by-bit flip, probability ; S326. Two-stage alternating optimization: Phase 1: Fix the scheduling of the second-level water pumps and optimize water resource allocation; Water supply capacity constraints: ; Phase Two: Fix the primary water source allocation and optimize pump scheduling; constraint: ; Iteration: Repeat phase one and phase two until the fitness converges; S327. Elite Preservation and Population Renewal: Each generation retains the top 5% of the water allocation schemes and corresponding water pump scheduling, and replaces the 20% of individuals with the lowest fitness in the old population with new individuals; S327: Termination Condition: The number of iterations reached 200 generations, the fitness change was less than 1% for 10 consecutive generations, and the water supply reliability reached over 95%.

2. The method for optimizing the scheduling of parallel operation of multiple water plants based on model algorithms according to claim 1, characterized in that, Step S24 includes the following steps: S241. Verify the outflow rate and pressure of each water plant; S242. Adjust the pump characteristic curve, the friction coefficient and flow coefficient of the main pipe; S243. Verify the pressure and flow rate at monitoring points in the pipeline network; S244. Use FMGA to automatically verify the pipeline system model.

3. The method for optimizing the parallel operation and scheduling of multiple water plants based on a model algorithm as described in claim 1, characterized in that, Step S31 includes: S311. First-level optimization scheduling objective function and constraints; (1) The objective function is to decompose the total operating cost of multi-plant water supply into water intake and transmission costs, water plant water production costs and secondary pump station boosting costs; (2) Constraints refer to the mathematical expressions that control or restrict the operation of the system, including the hydraulic balance of the pipeline network, the water supply of the water plant, the water supply pressure of the secondary pumping station, and the water pressure requirements of users; S312. Objective function and constraints for second-level optimization scheduling; (1) The objective function is the overall operating power of the pumping station; (2) The constraints are the pump speed ratio and the pump flow rate; S313. Establish an optimization mathematical model, transform the multiple objectives that need to be controlled in the optimization scheduling into a single objective problem, and solve it by establishing a model with objective function and constraints.

4. The method for optimizing the parallel operation and scheduling of multiple water plants based on a model algorithm as described in claim 1, characterized in that, Step S33 includes: S331. Pump characteristic curve fitting; S332. Comprehensive analysis of water pump operating efficiency; S333. Solving the optimal scheduling model; S334. Energy-saving analysis of water pump operation conditions and energy-saving analysis of dispatching schemes.

Images