Water supply whole-process energy-saving method based on water volume prediction
By constructing a multi-layer sensor model and combining real-time water level and pump characteristic curves, the system automatically schedules the start and stop of water pumps, solving the energy-saving problem of excessive water pressure in the water supply plant's pipeline network and achieving energy-saving effects throughout the entire process.
Patent Information
- Application Number
- CN202411142674.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2026-03-03
AI Technical Summary
Existing water supply plants struggle to effectively reduce energy consumption when the water pressure in the pipeline network is too high, especially without replacing equipment. How to combine pipeline pressure reduction with water volume prediction to achieve energy conservation throughout the entire process is an urgent problem to be solved.
By constructing a water supply prediction model based on multilayer sensors and combining real-time water level and pump characteristic curves, the system automatically schedules the start and stop of water pumps to ensure that the clear water tank maintains a high operating level, thereby achieving energy-saving effects in the pipe network.
It effectively reduces the additional energy consumption of the pipeline network for water transmission, saving an average of about 4.3% of energy consumption, and achieving energy saving throughout the entire water supply system.
Smart Images
Figure CN121593522A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of water supply systems, specifically relating to an energy-saving method for the entire water supply process based on water volume prediction. Background Technology
[0002] In response to national energy conservation and emission reduction policies, to reduce greenhouse gas emissions, combat global climate change, and adhere to the principles of sustainable development, many water supply plants have been moving towards energy-saving control in recent years. Currently, research on energy-saving control in water supply plants mainly focuses on two directions. One is to improve existing water supply facilities and equipment, replacing old pipes with higher-performance pipes to reduce water volume and energy loss in the system, while using more efficient and powerful pumps to improve the system's energy utilization efficiency. While this method can effectively achieve energy conservation and consumption reduction in water plants, it often requires significant investment and is difficult to popularize. The second approach involves establishing a water volume prediction model for the water plant, using this as a basis for intelligent dynamic control of pump operation. The online system automatically adjusts the start / stop of pumps or uses frequency conversion speed regulation based on predicted water volume to ensure efficient pump operation, thereby saving electricity and reducing energy consumption. Energy-saving scheduling methods based on water volume prediction can achieve good energy-saving effects with lower investment without replacing water plant equipment and are widely applicable to most water plants in my country. However, this method still has limitations; it cannot effectively address the situation where the overall water pressure in most pipe networks is significantly too high to meet the most unfavorable point water pressure requirements. Setting high water pressure in water supply networks not only accelerates equipment wear and tear but also increases energy consumption during water transmission. Therefore, how to organically combine network pressure reduction with water volume forecasting to achieve energy conservation and consumption reduction in the network has become an urgent problem to be solved. Summary of the Invention
[0003] To address the aforementioned shortcomings of existing technologies, this invention provides an energy-saving method for the entire water supply process based on water volume prediction. This method automatically schedules the start and stop of water pumps according to the predicted water volume and the pump characteristic curves of each pump, thereby achieving energy-saving effects on the pipeline network.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0005] An energy-saving method for the entire water supply process based on water volume prediction is proposed. The method obtains real-time water level and pump characteristic curves, and adjusts the operating conditions of the pump unit according to the water supply prediction model based on multilayer sensors, so as to keep the clear water tank running at a high level.
[0006] This method is based on a water volume analysis system, which includes a central processing unit (CPU), a water volume data storage unit, a real-time water level data storage unit, a real-time water level sensor, a pump characteristic curve storage unit, a pump operation command unit, and a pump unit. The real-time water level sensor collects real-time water level information from the clear water tank and transmits this information to the real-time water level data storage unit via a signal. The water volume data storage unit stores historical water volume data. The pump characteristic curve storage unit stores the characteristic curve data of each pump. The CPU communicates with the water volume data storage unit and the real-time water level data storage unit via signals. The system includes a storage device for reading historical water volume data and real-time water level data and comparing them to obtain comparison results; a central processing unit (CPU) connected to a pump characteristic curve memory via a signal for reading characteristic curve data of each pump and calculating matching values with the comparison results to obtain operating condition commands for each pump unit; a pump operation command unit connected to the CPU via a signal for receiving operating condition commands from each pump unit; and each pump unit connected to the pump operation command unit via a signal for receiving operating condition commands and operating the pump unit. The CPU has a built-in water supply prediction model program.
[0007] Specifically, the following steps are included:
[0008] S1: The real-time water level sensor transmits the collected real-time water level information of the clear water pool to the real-time water level data storage.
[0009] S2: The central processing unit reads the real-time water level information of the clear water tank from the real-time water level data storage at intervals with a set step size. The central processing unit then reads the historical water volume data from the water volume data storage and inputs the two into the water supply prediction model program for calculation to obtain the comparison result.
[0010] S3: The central processing unit reads the characteristic curve data of each water pump from the water pump characteristic curve memory, calculates the matching value with the comparison result, and obtains the operating condition command of each water pump unit.
[0011] S4: The central processing unit sends the operating condition instructions of each of the water pump units to the water pump operation instruction unit. The water pump operation instruction unit issues operating condition instructions and runs the water pump units. Each of the water pump units runs according to the operating condition instructions, thereby realizing the regulation of the water level in the clear water tank.
[0012] S5: Repeat S1 to S4 to achieve continuous dynamic adjustment of the water level in the clear water tank.
[0013] Furthermore, the comparison result is the overlap between the real-time water level information of the clear water tank and the historical water level data. If the overlap reaches 90% or more, the result is output. The real-time water level information of the clear water tank is the water level value of the most recently collected set step size.
[0014] Furthermore, the matching value is the time matching value for starting each of the water pumps, calculated based on the current water level of the clear water tank, the target water level of the clear water tank, and the characteristic curves of each of the water pumps.
[0015] Furthermore, it also includes step S6: storing the data from this operation into the water volume data storage device and the central processing unit to train the water volume prediction model.
[0016] Furthermore, the number of wavelengths is set to 8, and the step size is 15 minutes.
[0017] Furthermore, the water supply prediction model program is a data model based on a multilayer sensor, and its construction method is as follows:
[0018] 1) Construct a raw water volume prediction model based on a multilayer sensor;
[0019] 2) Input historical water volume data;
[0020] 3) Preprocess historical water volume data;
[0021] 4) Based on the preprocessed water volume data, determine the sequence length and prediction length of the model;
[0022] 5) Construct a feature dataset based on the preprocessed water volume data, and divide it into a training set and a test set to train and test the original model until the final target water volume prediction model is obtained.
[0023] The historical data mentioned in step 2) is the water volume value for more than 30 consecutive natural days, and the water volume value is the measured water volume value at 15-minute intervals.
[0024] The preprocessing described in step 3) includes handling outlier data points and handling missing data points. The handling of outlier data points involves: detecting outlier data points using box plots; and then processing these outlier data points by performing approximate value calculations using linear interpolation to obtain approximate data points. The detection method for outlier data points is based on a normal distribution, with the criterion being: |x-μ|>3σ; if so, it should be marked as an outlier, set to null, and filled using interpolation. Here, x represents the value of the data point to be detected, μ represents the average of all data points, and σ represents the standard deviation of the data, used to measure the dispersion of the data distribution.
[0025] The missing data points are handled by using interpolation to fill in the missing values.
[0026] Compared with existing technologies, the present invention has the following advantages:
[0027] 1. In view of the impact of pipeline water pressure on water plant energy consumption, this invention constructs a water supply volume prediction model for water plants based on multilayer sensors. This model is integrated into the online management system, and can automatically schedule the start and stop of water pumps based on the pump characteristic curves of each pump and the predicted water volume, thereby achieving energy-saving effects on the pipeline network.
[0028] 2. By monitoring and providing real-time feedback on the water level in the clear water tank online, and intelligently scheduling the operation of the water pumps, the clear water tank is kept at a high water level, thereby effectively reducing additional energy consumption for water delivery. Attached Figure Description
[0029] Figure 1 This is a schematic diagram of a water supply analysis system. Detailed Implementation
[0030] The present invention will be further described in detail below with reference to specific embodiments.
[0031] An energy-saving method for the entire water supply process based on water volume prediction is proposed. The method obtains real-time water level and pump characteristic curves, and adjusts the pump operating conditions according to a water volume prediction model based on a multilayer sensor to keep the clear water tank operating at a high level.
[0032] This method is implemented based on a water analysis system, such as... Figure 1 As shown, the water analysis system includes a central processing unit (CPU), a water volume data storage unit, a real-time water level data storage unit, a real-time water level sensor, a water pump characteristic curve storage unit, a water pump operation command unit, and a water pump unit. The real-time water level sensor is used to collect real-time water level information from the clear water tank and transmit this information to the real-time water level data storage unit via a signal. The water volume data storage unit stores historical water volume data. The water pump characteristic curve storage unit stores the characteristic curve data of each water pump. The CPU is connected to the water volume data storage unit and the real-time water level data storage unit via signals. The CPU is used to read historical water volume data and real-time water level data and compare them to obtain comparison results. The CPU is connected to a pump characteristic curve memory via a signal to read the characteristic curve data of each pump and calculate the matching value with the comparison results to obtain the operating condition instructions for each pump unit. The pump operation instruction unit is connected to the CPU via a signal to receive the operating condition instructions for each pump unit. Each pump unit is connected to the pump operation instruction unit via a signal to receive the operating condition instructions and operate the pump unit. The CPU has a built-in water volume prediction model program.
[0033] Specifically, the following steps are included:
[0034] S1: The real-time water level sensor transmits the collected real-time water level information of the clear water pool to the real-time water level data storage.
[0035] S2: The central processing unit reads the real-time water level information of the clear water tank from the real-time water level data storage at intervals with a set step size. The central processing unit then reads the historical water volume data from the water volume data storage and inputs the two into the water supply prediction model program for calculation to obtain the comparison result.
[0036] S3: The central processing unit reads the characteristic curve data of each water pump from the water pump characteristic curve memory, calculates the matching value with the comparison result, and obtains the operating condition command of each water pump.
[0037] S4: The central processing unit sends the operating condition instructions of each of the water pump units to the water pump operation instruction unit. The water pump operation instruction unit issues operating condition instructions and runs the water pump units. Each of the water pump units runs according to the operating condition instructions, thereby realizing the regulation of the water level in the clear water tank.
[0038] S5: Repeat S1 to S4 to achieve continuous dynamic adjustment of the water level in the clear water tank.
[0039] The water volume prediction model program is a data model based on a multilayer perceptron, and its construction method is as follows:
[0040] 1) Construct a raw water volume prediction model based on a multilayer sensor;
[0041] 2) Input historical water volume data;
[0042] 3) Preprocess historical water supply data;
[0043] 4) Based on the preprocessed water volume data, determine the sequence length and prediction length of the model;
[0044] 5) Construct a feature dataset based on the preprocessed water volume data, and divide it into a training set and a test set to train and test the original model until the final target water volume prediction model is obtained.
[0045] A Multilayer Perceptron (MLP) is a feedforward neural network architecture consisting of an input layer, one or more hidden layers, and an output layer. Within each layer, neurons receive the outputs of neurons in the previous layer, sum them using weighted averages, add a bias term, and then perform a nonlinear transformation through an activation function to generate the layer's output. This process is repeated in each layer until the final predicted output is produced. The key advantage of using MLPs as a water quantity prediction model lies in their nonlinear mapping capabilities and automatic feature learning. MLPs capture the complex relationship between input features and water quantity outputs through nonlinear activation functions in the hidden layers, while automatically extracting features, reducing reliance on manual feature engineering.
[0046] In this invention, the collected dataset consists of water volume data (Q) of the target pipeline network from 10:45:00 on January 11, 2024 to 00:00:00 on April 30, 2024, recorded every 15 minutes. As shown in Table 1, there are a total of 10,611 rows.
[0047] Table 1. Collection of Datasets
[0048]
[0049] In this invention, the step of preprocessing the feature data includes:
[0050] Outlier data points are detected using box plots on each factor dataset. Approximate values are then calculated using linear interpolation to process the outlier data points, resulting in approximate data points, which are then used to update the factor dataset. The outlier detection method is based on a normal distribution, and its criteria are as follows:
[0051] |x-μ|>3σ;
[0052] If it is an outlier, it should be marked as null and filled with an interpolation value.
[0053] For missing data points, interpolation is used to fill in the missing values. Specifically, the integrity of the dataset is checked, and it is found that some data points are missing, mainly due to instrument malfunction or offline status. Interpolation is used to fill in the missing values.
[0054] The supplemented pipeline water volume data series is subjected to time series testing to determine its stationarity, specifically an Augmented Dickey-Fuller (ADF) test. This test is performed using Python. The ADF results include the test statistic, critical value, and p-value. If the test statistic is less than the critical value and the p-value is less than the significance level (usually set to 0.05), the null hypothesis can be rejected, meaning the data does not have a unit root and is considered stationary. If the test statistic is greater than the critical value or the p-value is greater than the significance level, the null hypothesis cannot be rejected, meaning the data has a unit root and is considered non-stationary.
[0055] Then, based on the preprocessed water volume data, the optimal sequence length and prediction length of the model were explored. In the model test results, the accuracy of the initially constructed target water volume prediction model reached its best when the sequence length was 96 (i.e., 24 hours) and the prediction length was 8 (i.e., 2 hours). Therefore, the sequence length was determined to be 96 and the prediction length to be 8.
[0056] Based on the previously determined sequence length and prediction length, the dataset is uniformly divided into a 60% training set, a 20% validation set, and a 20% test set. Then, the original water volume prediction model based on multilayer perceptron is trained and tested. Normalization and denormalization operations are performed when inputting data into the model and outputting prediction results until a smart model that meets the requirements is obtained.
[0057] In this invention, the water volume prediction model uses a time step of 15 minutes to predict the water supply of the water plant in the next two hours. The data is then input into the demand pressure linear regression model to obtain the demand water pressure in the next two hours. The online system will add a safe water pressure as the target water pressure on the basis of the demand water pressure, and intelligently schedule the start and stop of the water pump based on the predicted water volume and the target water pressure to achieve energy saving and efficiency improvement.
[0058] The specific method for intelligently regulating the start and stop of the water pump is as follows:
[0059] The water volume prediction model uses a time step of 15 minutes to predict the water supply volume of the water plant in the next two hours, as shown in Table 2.
[0060] Table 2 Water Quantity Forecast
[0061]
[0062] The predicted data is input into a linear regression model of demand pressure to obtain the demand water pressure for the next two hours. The online system will then add a safe water pressure to the demand water pressure as the target water pressure. Subsequently, the system will select the optimal pump combination and operating frequency based on the pump characteristic curves.
[0063] Table 3 Comparison of energy consumption before and after intelligent scheduling optimization
[0064]
[0065] According to the experimental results shown in Table 3, after intelligent optimization scheduling, the energy consumption of the water pump in each time period decreased compared with that before optimization, with an average energy saving of about 4.3%. This shows that the energy-saving method for the whole process of water supply based on water volume prediction proposed in this application can effectively save energy and increase efficiency for the target pumping station.
[0066] The above embodiments of the present invention are merely illustrative examples and are not intended to limit the implementation of the invention. Those skilled in the art can make other variations and modifications based on the above description. It is impossible to exhaustively list all possible implementations here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.
Claims
1. A method for energy conservation throughout the entire water supply process based on water volume prediction, characterized in that, The system acquires real-time water level and pump characteristic curves of the clear water tank, and adjusts the operating conditions of the water intake and delivery pumps based on the water supply prediction model of the multi-layer sensor based on the water distribution network, so as to keep the clear water tank operating at a high level and save energy.
2. The energy-saving method for the entire water supply process based on water volume prediction according to claim 1, characterized in that, Based on an intelligent water volume prediction system, the system includes a central processing unit (CPU), a water volume data storage unit, a real-time water level data storage unit, a real-time water level sensor, a water pump characteristic curve storage unit, a water pump operation command unit, and a water pump unit. The real-time water level sensor collects real-time water level information from the clear water tank and transmits this information to the real-time water level data storage unit via a signal. The water volume data storage unit stores historical water volume data. The water pump characteristic curve storage unit stores the characteristic curve data of each water pump. The CPU communicates with the water volume data storage unit and the real-time water level data storage unit via signals. The central processing unit (CPU) is connected to a memory for reading historical water volume data and real-time water level data and comparing them to obtain comparison results. It is also connected to a pump characteristic curve memory for reading the characteristic curve data of each pump and calculating the matching value with the comparison results to obtain operating instructions for each pump. A pump operation instruction unit is connected to the CPU via a signal to receive operating instructions from each pump. Each pump unit is connected to the pump operation instruction unit via a signal to receive operating instructions and operate the pump. The CPU has a built-in intelligent water volume prediction model program. Specifically, the following steps are included: S1: The real-time water level sensor transmits the collected real-time water level information of the clear water pool to the real-time water level data storage. S2: The central processing unit reads the real-time water level information of the clear water tank from the real-time water level data storage at intervals with a set step size. The central processing unit then reads the historical water volume data from the water volume data storage and inputs the two into the water volume intelligent prediction model program for calculation to obtain the comparison result. S3: The central processing unit reads the characteristic curve data of each water pump from the water pump characteristic curve memory, calculates the matching value with the comparison result, and obtains the operating condition command of each water pump unit. S4: The central processing unit sends the operating condition instructions of each of the water pump units to the water pump operation instruction unit. The water pump operation instruction unit issues operating condition instructions and runs the water pump units. Each of the water pump units runs according to the operating condition instructions, thereby realizing the regulation of the water level in the clear water tank. S5: Repeat S1 to S4 to achieve continuous dynamic adjustment of the water level in the clear water tank.
3. The energy-saving method for the entire water supply process based on water volume prediction according to claim 2, characterized in that, The comparison result is the overlap between the real-time water level information of the clear water tank and the historical water level data. If the overlap reaches 90% or more, the result is output. The real-time water level information of the clear water tank is the water level value of the most recently collected set step size.
4. The energy-saving method for the entire water supply process based on water volume prediction according to claim 2, characterized in that, The matching value is the time matching value for starting each of the water pump units, calculated based on the current water level of the clear water tank, the target water level, and the characteristic curve of the water pump.
5. The energy-saving method for the entire water supply process based on water volume prediction according to claim 2, characterized in that, It also includes step S6: storing the data from this operation into the water volume data storage device and the central processing unit, and training the intelligent water volume prediction model program.
6. The energy-saving method for the entire water supply process based on water volume prediction according to claim 3, characterized in that, The number of step sizes is set to 8, and the step size interval is 15 minutes.
7. The energy-saving method for the entire water supply process based on water volume prediction according to any one of claims 2-6, characterized in that, The water volume prediction model program is a data model based on a multilayer perceptron, and its construction method is as follows: 1) Construct a raw water volume prediction model based on a multilayer sensor; 2) Input historical water volume data; 3) Preprocess historical water volume data; 4) Based on the preprocessed water volume data, determine the sequence length and prediction length of the model; 5) Construct a feature dataset based on the preprocessed water volume data, and divide it into a training set and a test set to train and test the original model until the final target water volume prediction model is obtained.
8. The energy-saving method for the entire water supply process based on water volume prediction according to claim 7, characterized in that, The historical data mentioned in step 2) refers to water volume values for more than 30 consecutive natural days, and these water volume values are measured at 15-minute intervals.
9. The energy-saving method for the entire water supply process based on water volume prediction according to claim 7, characterized in that, The preprocessing described in step 3) includes handling outlier data points and handling missing data points; The processing of outlier data points is as follows: Outlier data points are detected by box plotting the data of each factor; approximate values are calculated using linear interpolation to process the outlier data points and obtain approximate data points. The detection method for outlier data points is based on a normal distribution, with the criterion being: |x-μ|>3σ; if so, it should be marked as an outlier, set to null, and filled using interpolation; where x represents the value of the data point to be detected, μ represents the average of all data points, and σ represents the standard deviation of the data, used to measure the dispersion of the data distribution. The missing data points are handled by using interpolation to fill in the missing values.