Intelligent consumption reduction system and method based on active decision supply and demand balance model

Abstract

The invention belongs to the technical field of intelligent control of a water supply network, and particularly discloses an intelligent consumption reduction system and method based on an active decision supply and demand balance model, and the system comprises a pressure data processing module which is used for cleaning historical pressure data and extracting the mean value and variance characteristics of pressure loss; the pressure loss prediction module is used for predicting a future pressure loss mean value and variance by adopting a neural network model, and identifying a water consumption peak period according to the mean value and variance; and the water outlet pressure game decision module is used for establishing a dynamic game model between a pump room and a user, the pump room aims at minimizing the water outlet pressure, and the user complaint behavior is triggered by the tail end pressure according to probability distribution. And the system calculates the optimal water outlet pressure value capable of realizing maximum energy conservation on the premise of ensuring the water use satisfaction of the user by solving game equilibrium. According to the invention, refined and intelligent regulation and control of water supply pressure can be realized, the energy-saving effect is obvious, and meanwhile, the water supply stability in the peak period is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent control of water supply network, and particularly relates to an intelligent energy-saving system and method based on an active decision-making supply-demand balance model. BACKGROUND

[0002] The municipal water supply system is a key infrastructure for ensuring urban water supply. However, the traditional water supply system has the problem of high energy consumption in operation. The root cause lies in that the pump house usually adopts a fixed pressure or a simple empirical pressure control strategy. Under this strategy, the outlet pressure of the pump house remains constant or is only roughly adjusted in segments within 24 hours a day. Whether in the early morning or in the morning and evening, the system supplies water at the same or similar high pressure regardless of the low water demand in the valley period or the high water demand in the peak period.

[0003] This control mode has significant defects: in periods of low water demand, maintaining high pressure operation causes a large amount of energy waste, directly increasing the operating cost of the water supply enterprise; long-term maintenance of unnecessary high pressure increases the operating load of the pipe network and valves and other accessories, accelerates equipment aging, and even increases the risk of pipe network leakage; it is difficult to find a dynamic and refined balance point between the two conflicting goals of "energy saving and consumption reduction" and "guaranteeing user water satisfaction". Simply reducing the pressure may lead to user complaints during peak periods, while maintaining high pressure will inevitably lead to high energy consumption.

[0004] Therefore, how to adjust the outlet pressure of the pump house in real time and dynamically according to the actual water demand of the user, to maximize energy saving and consumption reduction under the premise of guaranteeing user experience, is a technical problem to be solved in the field. SUMMARY

[0005] The present application aims to solve the problems in the background art, and provides an intelligent energy-saving system and method based on an active decision-making supply-demand balance model, which can accurately predict changes in user water demand and calculate the optimal outlet pressure through a game model based on this, thereby achieving significant energy saving and consumption reduction of the water supply system without reducing user water satisfaction.

[0006] Technical solution: The intelligent consumption reduction system based on the active decision supply-demand balance model comprises a pressure data processing module, which is configured to acquire historical data of pump house outlet water pressure and user terminal pressure, perform outlier processing, and calculate historical time series data of pressure loss, wherein the historical time series data of pressure loss comprises a pressure loss mean value representing user water average load and a pressure loss variance representing water fluctuation intensity; a pressure loss prediction module, which is configured to predict the pressure loss mean value and the pressure loss variance in a future predetermined time period based on historical time series data of the pressure loss mean value and the pressure loss variance through a pre-trained neural network model, and identify a water use peak period according to the predicted pressure loss variance; and an outlet water pressure game decision module, which is configured to establish a dynamic game model between the pump house and the user, wherein the pump house takes minimizing the outlet water pressure as a target, and the user's complaint behavior is triggered based on the received terminal pressure according to a preset complaint probability distribution, an optimal outlet water pressure value for balancing energy saving and consumption reduction and user water satisfaction is determined by solving the game model, and the value is output to control the pump house.

[0007] Further improve the above technical scheme, the pressure data processing module adopts quartile method for outlier processing, specifically: in the preset translation time window, calculate the first quartile Q1 and the third quartile Q3, identify the data points exceeding the range of Q1-1.5*IQR and Q3+1.5*IQR as outliers, and fill them by using forward filling method, wherein IQR is the interquartile range.

[0008] Further, the pressure loss prediction module comprises two independent neural network models, which are respectively: a first LSTM neural network model for predicting future mean value according to historical data of pressure loss mean value; and a second LSTM neural network model for predicting future variance according to historical data of pressure loss variance. Further, the first LSTM neural network model and the second LSTM neural network model are both long short-term memory neural network models, which comprise an input layer for receiving historical pressure loss data, an LSTM layer comprising a plurality of LSTM units, and a fully connected output layer using a linear activation function, and are trained with mean square error as a loss function.

[0009] Further, the pressure loss prediction module identifying the water use peak period comprises: determining one or more local maximum values as peak reference points in the predicted future pressure loss variance; multiplying the values of the peak reference points by a preset threshold coefficient to generate a dynamic threshold; and marking a period in which the predicted pressure loss variance exceeds the dynamic threshold as a water use peak period.

[0010] ​Further, the water outlet pressure game decision module is configured to model the pressure loss as a random variable subject to a normal distribution of the predicted mean and variance, and to calculate the total complaints by Monte Carlo simulation of a plurality of simulated users, wherein the simulation includes randomly generating a pressure loss value for each simulated user from the normal distribution.

[0011] Further, in the Monte Carlo simulation, the complaint behavior of each simulated user is modeled by a Sigmoid function that includes a shift parameter activated only during the identified peak water usage period to increase the end pressure threshold that triggers a complaint during the peak period.

[0012] Further, the total complaints is obtained by Bernoulli trials of the complaint event of each simulated user and summation.

[0013] Further, the water outlet pressure game decision module solves the game model by an iterative approach, which includes increasing the water outlet pressure according to the following formula if the calculated user complaints based on the current water outlet pressure and predicted pressure loss is greater than zero: ; wherein, is a preset adjustment step size, is the total complaints; and repeating the process until the calculated complaints is zero or the water outlet pressure reaches a preset upper limit.

[0014] The intelligent energy reduction method of the intelligent energy reduction system based on the active decision supply-demand balance model includes the following steps: A data processing step: obtaining historical data of pump house water outlet pressure and user end pressure, performing outlier processing, and calculating historical time series data of pressure loss, which includes pressure loss mean representing user water average load and pressure loss variance representing water fluctuation intensity; A loss prediction step: based on the historical time series data of the pressure loss mean and variance, predicting the pressure loss mean and variance in a future predetermined period of time through a pre-trained neural network model, and identifying the peak water usage period according to the predicted pressure loss variance; A game decision and control step: establishing a dynamic game model between the pump house and the user, wherein the pump house aims to minimize the water outlet pressure, and the complaint behavior of the user is triggered based on the end pressure received according to a preset complaint probability distribution, by solving the game model, an optimal water outlet pressure value is obtained, which balances energy saving and user water satisfaction, and the pump house is controlled according to the value.

[0015] Advantages: Compared with the prior art, the advantages of the present application are that: the present application can accurately predict future water load and fluctuation intensity by learning historical water mode through LSTM neural network, so that the system changes from passive response to active prediction, realizing forward-looking intelligent regulation; according to the prediction result, the outlet pressure is dynamically adjusted, the pipe network pressure is significantly reduced in the late night, intermission and other low water valley periods, and sufficient supply is ensured in the peak period, compared with the traditional constant pressure control strategy, the energy saving effect is remarkable; the user complaint prediction mechanism based on game theory and probability model quantifies the abstract "user satisfaction" into a calculable complaint probability, ensuring that the user's water experience is effectively guaranteed while saving energy and reducing consumption; the best balance point between energy consumption and satisfaction is automatically found through game decision, and "zero complaint" is taken as the game equilibrium target, which can automatically maintain or improve to safe pressure during the water peak period, ensuring that the user's water experience is not affected, and can adapt to seasonal or long-term changes in user water habits, with high intelligence and robustness. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is the architecture diagram of the intelligent energy saving system based on the active decision supply-demand balance model in the embodiment of the present application.

[0017] Figure 2 is a comparison diagram of the true value and the predicted value of the pressure loss variance in the embodiment of the present application.

[0018] Figure 3 is a distribution diagram of the peak period and the ordinary period identified based on the pressure loss variance in the embodiment of the present application.

[0019] Figure 4 is a comparison diagram of the original fixed outlet water pressure and the dynamic outlet water pressure optimized by the present application in the embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions of the present application will be described in detail below with the help of the drawings, but the protection scope of the present application is not limited to the described embodiments.

[0021] Embodiment 1: The intelligent energy saving system based on the active decision supply-demand balance model provided in the embodiment is shown in Figure 1 The overall architecture includes a pressure data processing module, a pressure loss prediction module and an outlet pressure game decision module, which together constitute an intelligent control system.

[0022] 1. Pressure data processing module This module is the data basis of the system, responsible for cleaning and feature extraction of the original data. The outlet pressure and the end pressure , according to the translation window, the window size is 10 min. The maximum threshold value and the minimum threshold value are calculated for each window of data using the quartile method ; Among them, is the first quartile (25% percentile), is the third quartile (75% percentile), . Values exceeding the threshold are considered outliers, that is Or . The outliers are marked as missing values, and the forward filling method is used to fill the outliers: ; Among them, is the nearest legal data time point before the missing value.

[0023] For the identified outliers, the forward filling method is used to fill them, that is, the value of the nearest legal data point before the outlier is used to replace it. After data cleaning, the pressure loss per second is calculated according to the cleaned outlet pressure and the end pressure , and the pressure loss value reflects the user's water consumption and pressure loss caused by the water delivery process at that time, and the specific calculation formula is: ; The pressure loss is regarded as a random variable, and the window size is set to 1 min. The mean and variance of the pressure loss are calculated. The mean of the pressure loss reflects the average water consumption load per minute, and the specific calculation formula is: ; Among them, is the number of data points per minute, which is generally . The variance of the pressure loss reflects the intensity of the water consumption fluctuation per minute, and the calculation formula is: ; Finally, the module outputs the historical time series data of the mean and variance of the pressure loss.

[0024] 2、Pressure loss prediction module Pressure loss prediction consists of two core steps: neural network prediction and peak period identification. Neural networks are established for the mean and variance of the pressure loss for prediction.

[0025] (1) Neural network prediction The mean and variance of pressure loss are constructed by LSTM neural network respectively. The historical data is divided into multiple data segments by sliding window, and each segment has 11 data. These data segments constitute the data set of LSTM neural network. For each prediction point, the input layer of LSTM neural network is the pressure loss data of the past 10 min at the corresponding time of the previous day. The model will use the pressure loss data of the past 10 min, i.e. 10 time steps, to predict the pressure loss at this prediction point. The LSTM layer contains 20 LSTM units, and the result of the last time step is output as the predicted value. A fully connected layer with a single neuron is added at the top of the model, using a linear activation function to map the features extracted by the LSTM layer to the final predicted value. The model uses mean squared error as the loss function, which can effectively measure the deviation between the predicted value and the actual value. The specific formula is as follows: ;

[0026] The mean squared error (MSE) is used as the loss function for training until the model can accurately reproduce the fluctuation rules of historical data. Referring to Figure 2 , the figure shows the high fitting effect of the predicted value (orange point) and the true value (green point) of the pressure loss variance.

[0027] After training the neural network with pressure loss data, the previous day's data is used as input to predict the pressure loss data using the trained neural network, and the predicted values of the mean and variance of the pressure loss of the next day are obtained. The specific formula is as follows: ; ;

[0028] (2) Peak period identification and smoothing Peak detection is performed on the predicted data. The larger value of the two local maximum points is considered as the peak value of water use in a day, denoted as . The range exceeding a certain peak value is considered as the peak period, and the peak label is generated to generate a dynamic threshold value. The specific formula is as follows: ; ​​​​​​​​​​. Any time period whose variance exceeds this dynamic threshold is marked as a peak water usage time period by the system, with reference to Figure 3 The figure clearly shows the distribution of the identified peak time periods (red dots) and normal time periods (blue dots).

[0029] 3. Water outlet pressure game decision module Through the dynamic game model between the pump house and the user, the water outlet pressure is reduced while ensuring the user's water usage satisfaction. The specific process includes the following steps: (1) Game environment The pump house can supply water at a certain water outlet pressure. The pump house pressure adjustment range is set as . Among them, is the original water outlet pressure before optimization, to ensure the water usage experience of residents, and is set as the safety lower limit of the system to prevent the water pressure from dropping too sharply and increase the risk of system complaints. Based on the pressure loss mean and variance output by the pressure loss prediction module, a random pressure loss model is constructed, and the pressure loss follows a normal distribution, reflecting the statistical characteristics of the pressure loss in the prediction period. The specific form is as follows: ; The end pressure reaching the user after the pressure loss is: ; The user can decide whether to make a complaint according to the end pressure. The smaller the end pressure, the higher the probability of user complaints. The sigmoid function is used to depict the distribution function of the user's complaint probability, and the specific expression is as follows: ; Among them, represents the user's complaint, is an indicator function of peak time period, is a shift parameter of the peak time period distribution function, usually taking 0.05, to avoid the time of all-day peak water usage period being too long. The introduction of the shift parameter of the peak time period ensures the water usage experience of the user in the peak time period.

[0030] (2) Pressure loss and user complaints In each round of game, a Monte Carlo simulation is performed to generate pressure loss for virtual users. The pressure loss is a random variable following a normal distribution: ; The pressure loss is truncated to ensure its reasonableness, . According to the current pump house water outlet pressure and pressure loss , the end pressure arriving at each user is calculated and the complaint probability of the user : ; ; A complaint event is generated by Bernoulli trial, and the random variable represents whether the user complains, and obeys 0-1 distribution, . The total number of complaints of the user has the following expression: ; At each decision-making moment, the final water outlet pressure is decided through game, and the time for completing a complete game equilibrium calculation is within 1 minute, which can meet the regulation and control requirements of the water supply system.

[0031] (3) Pressure regulation and equilibrium judgment When there are user complaints, , the pump house increases the water outlet pressure in proportion: ; Wherein, is the pressure regulation step, and small step regulation can obtain smaller water outlet pressure while meeting the water demand of the user. Repeat the process until the game reaches the equilibrium condition of zero complaints, that is, , or the water outlet pressure reaches the preset upper limit Pmax, that is, . At this time, the water outlet pressure obtained is the optimal solution of the current period, which is the lowest energy consumption pressure under the premise of ensuring user satisfaction.

[0032] As shown in Figure 4 , after the optimization of the system, the water outlet pressure (red dot) is significantly lower than the original fixed pressure (blue dot) in the low valley period such as midnight and noon, realizing deep energy saving; and in the morning and evening peak period, the pressure can quickly rise to a safe level, ensuring the centralized water demand.

[0033] As described above, although the present application has been shown and described with reference to specific preferred embodiments, it is not to be construed as being in any way limited to the details shown and described. Various modifications and changes can be made in form and detail without departing from the spirit and scope of the application as defined in the appended claims.

Claims

1. An intelligent consumption reduction system based on an active decision-making supply and demand balance model, characterized in that, Comprising: a pressure data processing module configured to obtain historical data of pump house outlet water pressure and user end pressure, perform outlier processing, and calculate historical time series data of pressure loss, the historical time series data of pressure loss including pressure loss mean value representing user water average load and pressure loss variance representing water fluctuation intensity; a pressure loss prediction module configured to predict pressure loss mean value and pressure loss variance in a future predetermined time period based on historical time series data of pressure loss mean value and pressure loss variance, and identify water peak time period according to the predicted pressure loss variance through a pre-trained neural network model; an outlet water pressure game decision module configured to establish a dynamic game model between the pump house and the user, wherein the pump house aims to minimize outlet water pressure, and the user's complaint behavior is triggered based on the end pressure received according to a preset complaint probability distribution, and an optimal outlet water pressure value for balancing energy saving and user water satisfaction is determined by solving the game model, and the value is output to control the pump house. 2.The intelligent consumption reduction system based on the active decision-making supply-demand balance model of claim 1, wherein, The pressure data processing module adopts quartile method for outlier processing, specifically: in a preset translation time window, the first quartile Q1 and the third quartile Q3 are calculated, data points exceeding the range of Q1-1.5*IQR and Q3+1.5*IQR are identified as outliers, and forward filling method is used for filling, wherein IQR is interquartile range. range of Q1-1.5*IQR and Q3+1.5*IQR are identified as outliers, and forward filling method is used for filling, wherein IQR is interquartile range. 3.The intelligent consumption reduction system based on the active decision-making supply-demand balance model of claim 1, wherein, The pressure loss prediction module includes two independent neural network models, which are respectively: a first LSTM neural network model for predicting future mean value according to historical data of pressure loss mean value; and a second LSTM neural network model for predicting future variance according to historical data of pressure loss variance.

4. The intelligent consumption reduction system based on active decision-making supply and demand balance model according to claim 3, characterized in that, The first LSTM neural network model and the second LSTM neural network model are both long short-term memory neural network models, which include an input layer for receiving historical pressure loss data, an LSTM layer including a plurality of LSTM units, and a fully connected output layer using a linear activation function, and are trained with mean square error as a loss function. 5.The intelligent consumption reduction system based on the active decision-making supply-demand balance model of claim 1, wherein, The pressure loss prediction module identifying the water peak time period includes: determining one or more local maximum values as peak reference points in the predicted future pressure loss variance; multiplying the values of the peak reference points by a preset threshold coefficient to generate a dynamic threshold; and marking a period when the predicted pressure loss variance exceeds the dynamic threshold as a water peak time period. 6.The intelligent consumption reduction system based on the active decision-making supply-demand balance model of claim 1, wherein, The outlet water pressure game decision module is configured to: model the pressure loss as a random variable subject to a normal distribution of the predicted mean value and variance; and calculate the total number of complaints by Monte Carlo simulation of a plurality of simulated users, wherein the simulation includes randomly generating pressure loss values for each simulated user from the normal distribution.

7. The intelligent consumption reduction system based on active decision-making supply and demand balance model according to claim 6, characterized in that, In the Monte Carlo simulation, the complaint behavior of each simulated user is modeled by a Sigmoid function, which includes a translation parameter activated only in the identified water peak time period, for increasing the end pressure threshold triggering complaints during the peak period. 8.The intelligent consumption reduction system based on the active decision-making supply-demand balance model of claim 7, wherein, The total number of complaints is obtained by performing Bernoulli trials on the complaint events of each simulated user and summing them up. 9.The intelligent consumption reduction system based on the active decision-making supply-demand balance model of claim 1, wherein, The water outlet pressure game decision module solves the game model by an iterative method, which includes: if the number of user complaints calculated based on the current water outlet pressure and the predicted pressure loss is greater than zero, the water outlet pressure is increased according to the following formula: ; wherein, is a preset adjustment step, is the total number of complaints; and the process is repeated until the calculated number of complaints is zero or the water outlet pressure reaches a preset upper limit. 10.The intelligent consumption reduction method based on the active decision-making supply-demand balance model of the intelligent consumption reduction system according to claim 1, characterized in that, Comprising the following steps: Data processing step: Obtain historical data of outlet pressure of pump house and user end pressure, process outliers, and calculate historical time series data of pressure loss, which includes pressure loss mean value representing user water average load and pressure loss variance representing water fluctuation intensity; Loss prediction step: Based on historical time series data of pressure loss mean value and variance, predict pressure loss mean value and variance in a predetermined future period through a pre-trained neural network model, and identify water peak period according to predicted pressure loss variance; Game decision and control step: Establish a dynamic game model between pump house and user, in which pump house aims to minimize outlet pressure, and user's complaint behavior is triggered based on received end pressure according to preset complaint probability distribution, solve the game model to determine an optimal outlet pressure value for balancing energy saving and user water satisfaction, and control the pump house according to the value.

Images