Pump station water supply amount prediction method for optimized operation of pump station

By using water consumption prediction methods based on historical data and real-time temperature in pump station management, combined with daily temperature difference impact factor and VMD neural network module, the problems of low operating efficiency and waste of energy are solved, and accurate prediction of water consumption is achieved The optimization of operational strategies has improved the stability of water supply services and intelligent management.

CN119090077AInactive Publication Date: 2024-12-06CHONGQING UNIV

Patent Information

Application Number
CN202411208367.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing pump station management model is based on manual experience and lacks consideration of dynamic changes in water consumption, resulting in low operating efficiency, waste of energy consumption, and inaccurate water consumption prediction, making it difficult to optimize pump station operation.

Method used

The water consumption prediction method based on historical data and real-time temperature is adopted, and the water consumption prediction model is constructed by calculating the daily temperature difference influence factor, using VMD and neural network modules to make predictions, and the pump station operation strategy is adjusted in real time.

Benefits of technology

It improves the accuracy of water consumption prediction, realizes optimization of pump station operation, reduces energy consumption and costs, improves the stability and reliability of water supply services, and promotes the intelligence and refinement of pump station management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090077A_ABST
    Figure CN119090077A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of water pump station regulation and control, and particularly relates to a pump station water supply prediction method for pump station optimized operation, which comprises the following steps: S1, acquiring historical data including water consumption and temperature of a time sequence; calculating a daily temperature difference influence factor of the corresponding time sequence according to the temperature data; s2, constructing a water consumption prediction model which is used for predicting the water consumption of the time sequence in the future X hours according to the water consumption, the temperature and the daily temperature difference influence factors; s3, taking the water consumption, temperature and daily temperature difference influence factor data of the time sequence as training data to train the water consumption prediction model; s4, acquiring actual water consumption and temperature, and calculating a real-time daily temperature difference influence factor; performing real-time water consumption prediction through the trained water consumption prediction model; and S5, adjusting and optimizing the operation of the pump station according to the water consumption prediction result obtained in the step S4. According to the method, the precision of the predicted water consumption can be improved, so that the operation of the pump station is effectively optimized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of water pump station control, and in particular relates to a method for predicting the water supply volume of a pump station for optimizing the operation of the pump station. Background Art

[0002] The water supply system is an important part of urban infrastructure, and the water pump is the key in the water supply system. The energy consumed by its operation accounts for the main part of the entire system. However, most pumping stations in my country still have the problem of energy waste. According to relevant industry statistics, the operating efficiency of urban water pump units is low, currently generally between 50% and 75%. As the main energy consumption link in the urban water supply system, improving its operating efficiency has an important impact on achieving the industry's energy-saving goals. Therefore, studying the optimization operation strategy of water pumps, improving the operating efficiency of pump stations and reducing operating energy consumption will help achieve the city's green and low-carbon development goals.

[0003] At present, my country's water supply pumping stations and booster pumping stations generally adopt a management model based on manual experience, mainly for the safe operation of the pipeline network. This model has a certain degree of blindness in management and decision-making, resulting in many pumping stations with low operating efficiency, pressure waste, and energy waste. The main reason for this problem is that urban water consumption is a dynamic process that changes with the seasons and water use periods, so the demand for water flow and pressure of the corresponding pumping station is also dynamically changing; however, the experience-based operation and management model often lacks consideration of such variable factors. For example, a typical operation mode based on manual experience is: first consider the control mode of constant pumping station outlet pressure under the premise of ensuring water supply at the most unfavorable point of the pipeline network during peak hours. The operation of the water supply system in this mode will not only generate a lot of energy waste, but also cause the problem of excessive pressure in the pipeline network during low-peak water use periods, increasing the risk of pipeline bursts.

[0004] As more and more water companies in my country have established information platforms for water supply systems and collected a large amount of water supply information data, this has made it possible to adopt a data model to drive the operation and management of pump stations. To optimize the management of pump stations, a major premise is that it is necessary to accurately predict the real-time water consumption. If the water consumption prediction is inaccurate, the optimization of water pump management will not achieve the goal. However, the accuracy of directly using historical water consumption data to predict water consumption is not accurate. Therefore, the current operation and management of pump stations still has a lot of room for improvement.

[0005] In summary, how to improve the accuracy of predicted water consumption and effectively optimize the operation of pumping stations has become a problem that needs to be solved urgently. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for predicting water supply of a pumping station for optimizing the operation of the pumping station, which can improve the accuracy of the predicted water consumption, thereby effectively optimizing the operation of the pumping station.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A method for predicting water supply of a pumping station for optimizing operation of a pumping station comprises the following steps:

[0009] S1. Obtain historical data; the historical data includes water consumption and temperature in a time series; and calculate the daily temperature difference impact factor of the corresponding time series according to the temperature data according to a preset calculation method;

[0010] S2. Construct a water consumption prediction model to predict the water consumption of the time series in the next X hours based on the water consumption, temperature and daily temperature difference influencing factors;

[0011] S3, using the time series water consumption, temperature and daily temperature difference influencing factor data of S1 as training data, to train the water consumption prediction model constructed by S2;

[0012] S4. Obtain the actual water consumption and temperature, and calculate the real-time daily temperature difference influencing factor according to the preset method; then perform real-time water consumption prediction by completing the trained water consumption prediction model;

[0013] S5. Adjust and optimize the operation of the pump station based on the water consumption prediction results obtained in S4.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] 1. Compared with the conventional water consumption forecasting based on historical water consumption data, this method adds the influencing factors of temperature and daily temperature difference when predicting water consumption.

[0016] Temperature difference is an important factor affecting water consumption (such as changes in residents' water consumption due to seasonal changes and differences in weather conditions). Its inclusion in the model significantly enhances the environmental adaptability of the model. However, through analysis, the applicant found that the impact of temperature on water consumption shows a certain lag, and as the lag time increases, the correlation coefficient tends to gradually decrease. Therefore, this method creatively designs a daily temperature difference influencing factor. By combining the temperature and daily temperature difference influencing factor data set by this method with historical water consumption data, the accuracy of the predicted water consumption can be effectively improved compared with conventional technologies.

[0017] 2. By acquiring water consumption and temperature data in real time, calculating the temperature difference influencing factors instantly, and using the trained model for real-time prediction, the pump station can respond quickly to changes in demand and adjust the operation strategy in a timely manner, such as adjusting the number of water pumps turned on and the water supply pressure, so as to achieve the purpose of energy saving and consumption reduction and improve water supply efficiency.

[0018] 3. Adjustment and optimization of pump station operation based on prediction results can avoid over-allocation or under-allocation of resources, reduce energy consumption and costs, and improve the stability and reliability of water supply services. At the same time, it also provides a scientific basis for the management and decision-making of pump stations, and promotes the intelligent and refined operation and management of pump stations.

[0019] In summary, this method can improve the accuracy of predicted water consumption, thereby effectively optimizing the operation of the pump station.

[0020] Preferably, the preset calculation method for calculating the daily temperature difference impact factor is:

[0021]

[0022] q' n-1 =q n-1 +509.13·α·(T2-T1);

[0023]

[0024] In the formula, is the daily temperature difference influencing factor; q n-1 is the measured water consumption on the n-1th day; T2 is the average temperature on the n-1th day; T1 is the average temperature of the previous N days; T 25℃ Indicates 25℃.

[0025] With this setting, by calculating the daily temperature difference impact factor, the monthly average temperature in different seasons can be linked to the daily temperature difference, thereby building a connection between the predicted water volume and temperature changes, making the prediction model more accurate when dealing with temperature changes. The daily temperature difference impact factor can effectively characterize the hysteresis characteristics of temperature, thereby ensuring the accuracy of subsequent water consumption forecasts.

[0026] Preferably, the water consumption prediction model includes a VMD module and a neural network module;

[0027] The working process of the water consumption prediction model includes: the VMD module decomposes the input time series data into several imf components, and sends them to the neural network module to obtain several corresponding prediction components; then, the prediction components are superimposed to obtain the water consumption prediction result.

[0028] Such a setting can, 1. improve the prediction accuracy. The VMD module decomposes the input time series data into several intrinsic mode function (IMF) components, which usually have different frequency characteristics and time scales, and can more carefully describe the dynamic changes of the original data. This decomposition process helps to remove noise and redundant information in the data, allowing the subsequent neural network module to focus more on learning the key features in the data, thereby improving the accuracy and stability of the prediction.

[0029] 2. It can enhance the adaptability of the model. Since the VMD module can adaptively decompose complex time series data into multiple simple components, the neural network module can use more targeted structures and parameters when processing each component. This divide-and-conquer strategy not only reduces the complexity of the model, but also enhances the model's ability to adapt to changes in different types of data, allowing the model to maintain a high prediction performance when facing complex and changing water demand.

[0030] 3. It can improve computing efficiency. By decomposing the original data into multiple IMF components and predicting them separately, and then superimposing the prediction results to obtain the final water consumption prediction results, this parallel processing method can significantly improve computing efficiency. Compared with directly predicting the entire time series data, the divide-and-conquer strategy reduces the computational burden of the neural network module, allowing the model to respond to real-time data faster and give prediction results.

[0031] Preferably, the neural network module is an LSTM neural network model or a BP neural network model.

[0032] Preferably, when the water consumption prediction model is working, each imf component is normalized before being sent to the neural network module.

[0033] Preferably, when the water consumption prediction model is working, after obtaining the corresponding prediction components, each prediction component is also denormalized and then superimposed.

[0034] Since different IMF components may have different dimensions and numerical ranges, directly feeding these components into the neural network module may cause gradient vanishing or gradient exploding problems during model training, thus affecting training efficiency and stability. Normalization can eliminate this dimensional difference, making all components numerically at the same order of magnitude, which helps the neural network converge to the optimal solution faster. In addition, normalization not only helps model training, but also improves the prediction accuracy of the model. Because neural networks usually perform better when processing data with similar values, normalized data is closer to the ideal input range of the neural network, allowing the model to more accurately capture the features and patterns in the data, thereby giving more accurate prediction results.

[0035] Preferably, in S1, the acquired historical data is also preprocessed, and the preprocessing includes outlier processing and missing value filling.

[0036] Outliers and missing values ​​are common problems in data. They may be caused by measurement errors, equipment failures, missing records, etc. If these inaccurate data are used directly for model training, they will seriously affect the prediction accuracy and reliability of the model. Through outlier processing and missing value filling, these inaccurate data points can be corrected or replaced, thereby improving the overall data quality.

[0037] Preferably, 5≤X≤24.

[0038] Since the correlation between temperature and the previous day is the largest, such a setting can ensure the accuracy of the predicted water consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0040] Figure 1 A flowchart of the method is shown in FIG.

[0041] Figure 2 This is a schematic diagram of the decomposition results of VMD summer monthly water consumption in Example 1;

[0042] Figure 3 This is a schematic diagram of the decomposition results of VMD winter monthly water consumption in Example 1;

[0043] Figure 4 This is a schematic diagram of the research idea of ​​the water supply prediction model in Example 2;

[0044] Figure 5 This is a schematic diagram of sample data of hourly water consumption in Example 2;

[0045] Figure 6 This is a schematic diagram of the water consumption prediction results of the BP single model in Example 2;

[0046] Figure 7 This is a schematic diagram of the water consumption prediction results of the LSTM single model in Example 2;

[0047] Figure 8 Schematic diagram of the water consumption prediction results of the VMD-BP combined model in Example 2

[0048] Fig. 9 This is a schematic diagram of the water consumption prediction results of the VMD-LSTM combined model in Example 2;

[0049] Fig.10 This is a schematic diagram showing the comparison of the summer water consumption prediction performance of the prediction model in Example 2;

[0050] Fig.11 This is a schematic diagram comparing the winter water consumption prediction performance of the prediction model in Example 2. DETAILED DESCRIPTION

[0051] The following is a further detailed description through specific implementation methods:

[0052] Embodiment 1

[0053] like Figure 1 As shown, this embodiment discloses a method for predicting the water supply of a pumping station for optimizing the operation of the pumping station, comprising the following steps:

[0054] S1. Obtain historical data; the historical data includes water consumption and temperature in a time series; and calculate the daily temperature difference impact factor of the corresponding time series according to the temperature data in accordance with a preset calculation method.

[0055] In specific implementation, the preset calculation method for calculating the daily temperature difference impact factor is:

[0056]

[0057] q' n-1 =q n-1 +509.13·α·(T2-T1);

[0058]

[0059] In the formula, is the daily temperature difference influencing factor; q n-1 is the measured water consumption on the n-1th day; T2 is the average temperature on the n-1th day; T1 is the average temperature of the previous N days; T 25℃ Indicates 25℃.

[0060] With this setting, by calculating the daily temperature difference impact factor, the monthly average temperature in different seasons can be linked to the daily temperature difference, so as to build a connection between the predicted water volume and temperature changes, making the prediction model more accurate in dealing with temperature changes. The average temperature of the N days before the current day can be obtained by averaging the temperature data of the previous N days recorded in real time by obtaining the meteorological temperature big data through the network; the N days here are preferably the number of days in a certain time period to ensure the stability of the average temperature data as much as possible. For example, the value range of N days can be 15 to 60 days. According to actual needs, select the specific N day value, such as 15 days, 30 days, 45 days or 60 days. The hysteresis characteristics of temperature can be effectively characterized by the daily temperature difference impact factor. Thereby, the accuracy of subsequent water consumption forecast can be guaranteed.

[0061] S2. Construct a water consumption prediction model to predict the water consumption in the next X hours of time series based on water consumption, temperature and daily temperature difference influencing factors. Among them, 5≤X≤24. Since the correlation of temperature is the largest with the previous day, such a setting can ensure the accuracy of the predicted water consumption.

[0062] In specific implementation, the water consumption prediction model includes a VMD module and a neural network module;

[0063] The working process of the water consumption prediction model includes: the VMD module decomposes the input time series data into several imf components, and sends them to the neural network module to obtain several corresponding prediction components; then, the prediction components are superimposed to obtain the water consumption prediction result.

[0064] Among them, before sending each IMF component into the neural network module, each IMF component is also normalized. After obtaining the corresponding several prediction components, each prediction component is also denormalized and then superimposed. Since different IMF components may have different dimensions and numerical ranges, directly sending these components into the neural network module may cause gradient vanishing or gradient explosion problems during model training, thereby affecting training efficiency and stability. Normalization can eliminate this dimensional difference, so that all components are at the same numerical level, which helps the neural network converge to the optimal solution faster. In addition, normalization not only helps model training, but also improves the prediction accuracy of the model. Because neural networks usually perform better when processing data with similar values, the normalized data is closer to the ideal input range of the neural network, allowing the model to more accurately capture the characteristics and laws in the data, thereby giving more accurate prediction results.

[0065] In this way, the prediction accuracy can be improved. The VMD module decomposes the input time series data into several intrinsic mode function (IMF) components, which usually have different frequency characteristics and time scales, and can more carefully characterize the dynamic changes of the original data. This decomposition process helps to remove noise and redundant information in the data, so that the subsequent neural network module can focus more on learning the key features in the data, thereby improving the accuracy and stability of the prediction. The adaptability of the model can also be enhanced. Since the VMD module can adaptively decompose complex time series data into multiple simple components, the neural network module can adopt more targeted structures and parameters when processing each component. This divide-and-conquer strategy not only reduces the complexity of the model, but also enhances the model's ability to adapt to changes in different types of data, so that the model can maintain a high prediction performance when facing complex and changing water demand. And it can improve computational efficiency. By decomposing the original data into multiple IMF components and predicting them separately, and then superimposing the prediction results to obtain the final water consumption prediction result, this parallel processing method can significantly improve computational efficiency. Compared with directly predicting the entire time series data, the divide-and-conquer strategy reduces the computational burden of the neural network module, allowing the model to respond to real-time data more quickly and give prediction results.

[0066] Among them, the neural network module is an LSTM neural network model or a BP neural network model.

[0067] BP neural network, also known as the back propagation model of error, is a mature model that has been widely used. It trains neural networks based on the gradient descent algorithm and uses it to adjust the weights and biases of the neural network to minimize the error between the predicted output and the actual output. The essence of its learning and training process is an error correction algorithm, including two processes: forward propagation of information and back propagation of errors. The basic BP neural network structure includes an input layer, at least one hidden layer, and an output layer. The main feature of the network structure is that all layers are fully connected. Each layer of neurons is connected to the next layer of neurons in a certain way, but the neurons in the same layer are independent of each other. Each layer of neurons only receives information from the previous layer of neurons.

[0068] The LSTM model has made improvements on the structure of the RNN model, avoiding the problem of vanishing gradients in the RNN model and significantly improving the prediction performance in some long-distance dependent tasks. The LSTM neural network model has a good ability to handle vanishing gradients and long-term dependence on learning sequences. The internal processing units are more refined, allowing it to retain valuable information and forget other information to a certain extent, thereby achieving long-term memory.

[0069] The idea of ​​VMD (Variational Mode Decomposition) is to regard the signal to be decomposed as composed of different intrinsic mode functions (IMFs). In order to avoid modal aliasing during the signal decomposition process, VMD adopts a completely non-recursive modal decomposition method instead of the recursive solution idea used in traditional signal decomposition algorithms. Compared with traditional methods, VMD has the advantages of non-recursive solution and autonomous selection of the number of modes. This design enables VMD to decompose signals more effectively while avoiding the influence of aliasing problems. Therefore, the VMD method can decompose complex signals into a series of modal components of different frequencies, each of which has specific frequency and amplitude characteristics. Its main idea is to find the modal components of the signal through an iterative optimization process.

[0070] Variational Mode Decomposition (VMD) is an adaptive and completely non-recursive signal decomposition method. VMD can adapt to the characteristics of the signal to accurately decompose the signal and extract useful information while minimizing the effects of noise and aliasing, which makes VMD a powerful signal processing tool.

[0071] In the specific implementation, the balance parameter of the VMD decomposition method is set to 2000. The two time series of summer water consumption and winter water consumption are respectively introduced into the VMD signal decomposition algorithm. After decomposition, 6 modes are finally obtained for each. The results are as follows: Figure 2 and Figure 3 As shown in the figure, it can be seen that after the two sets of original data sequences are decomposed, each of them obtains 6 sets of IMFs with different frequencies. Subsequently, these six sets of data sequences need to be normalized and then imported into the neural network model again, so that the model can learn and train the characteristics of each signal, and obtain the prediction results respectively. Then, these result sequences are denormalized and added to obtain the final prediction result.

[0072] S3, using the time series water consumption, temperature and daily temperature difference influencing factor data of S1 as training data, to train the water consumption prediction model constructed by S2.

[0073] In the specific implementation, in order to ensure the validity of the training data, data preprocessing is also performed before obtaining the training data. The preprocessing includes outlier processing and missing value filling. Outliers and missing values ​​are common problems in data. They may be caused by measurement errors, equipment failures, record omissions, etc. If these inaccurate data are directly used for model training, they will seriously affect the prediction accuracy and reliability of the model. Through outlier processing and missing value filling, these inaccurate data points can be corrected or replaced, thereby improving the overall data quality.

[0074] In specific implementation, the preprocessing process is as follows.

[0075] 1) Data detection and processing

[0076] The 3δ (3 times the standard deviation) detection method is a method based on the normal distribution, where δ represents the standard deviation and μ represents the mean. The idea of ​​this method is to assume that the data follows a normal distribution. In this case, the probability that the data falls within the range of (μ-3δ, μ+3δ) is 0.9974. If the data point value in the sample exceeds this range, it does not conform to the normal distribution and will be judged as an outlier due to anomalies, disturbances or noise. Generally speaking, the ±3δ error can be regarded as the limit error of the data, and the probability of falling outside this interval is only 0.27%. In a limited number of data collections, the probability of normal water consumption data falling outside this range is very small. Therefore, the ±3δ criterion is widely used because of its simple theory and reliable verification results. It is suitable for finding outliers in a large amount of instantaneous water supply flow data.

[0077] 2) Methods for processing data jump outliers

[0078] For abnormal values ​​of jump flow in individual instantaneous data, the "next and next" filling method in statistics can be used to handle it. That is, use linear interpolation to estimate the missing data. Assuming that the flow data is missing at time point t, it is necessary to find the time points t1 and t2 of the two non-abnormal flow values ​​closest to t, and the corresponding flow values ​​are Q1 and Q2 respectively. Then use the following formula to calculate the missing data Q:

[0079]

[0080] Through the above method, missing data can be estimated by known traffic data to fill in outliers.

[0081] 3) Missing value filling method

[0082] For the missing continuous time series data that may be caused by equipment failure or regular maintenance, since the water consumption data series has periodic and seasonal characteristics, when processing this part of the missing data, in order to ensure that the data characteristics are not destroyed, we cannot rely solely on linear interpolation. We need to use more complex methods to fill in the missing values ​​to more accurately reflect the actual changes in water consumption.

[0083] Historical data is used to reveal the periodic characteristics of the water volume series, and the impact of random disturbances is reduced by analyzing the deviation ratio between historical data and missing data. Specifically, a total of 96 data of hourly flow values ​​for 4 consecutive days are taken as a whole, assuming that data from the nth to the (n+kth)th data points are missing. First, by analyzing historical data, the periodic characteristics of the water volume series, such as daily peaks and valleys, and other periodic changes, can be identified. Then, the average value of the corresponding time period in the historical data (from the nth to the (n+kth)th) is calculated and used as the estimated value of the real-time data. The deviation ratio of the historical data in the same time period as the missing data is determined by calculating the deviation ratio of the average value of the corresponding time period in the historical data and the actual real-time data.

[0084]

[0085] Finally, the estimated value of the real-time data is adjusted with the average value of the corresponding time period in the historical data, and the estimated value of the real-time data is corrected according to the calculated deviation ratio to obtain a more accurate filling value.

[0086]

[0087]

[0088] In the formula, q i represents the flow value of the i-th day containing missing data; q x is the flow value on the xth day with complete data; Q n is the supplementary value of the missing flow at the nth moment; Q xn is the flow value at the nth moment on the xth day.

[0089] S4. Obtain the actual water consumption and temperature, and calculate the real-time daily temperature difference influencing factor according to the preset method; then perform real-time water consumption prediction by completing the trained water consumption prediction model;

[0090] S5. Adjust and optimize the operation of the pump station based on the water consumption prediction results obtained in S4.

[0091] Compared with the conventional water consumption prediction based on historical water consumption data, this method adds temperature and daily temperature difference influencing factors when predicting water consumption. Temperature difference is an important factor affecting water consumption (such as changes in residents' water consumption due to seasonal changes and differences in cold and warm weather), and its inclusion in the model significantly enhances the environmental adaptability of the model. However, through analysis, the applicant found that the impact of temperature on water consumption shows a certain lag, and as the lag time increases, the correlation coefficient tends to gradually decrease. Therefore, this method creatively designs a daily temperature difference influencing factor. By combining the temperature and daily temperature difference influencing factor data set by this method with historical water consumption data, the accuracy of the predicted water consumption can be effectively improved compared with conventional technologies.

[0092] By acquiring water consumption and temperature data in real time, calculating the temperature difference influencing factors in real time, and using the trained model for real-time prediction, the pump station can quickly respond to changes in demand and adjust the operation strategy in a timely manner, such as adjusting the number of water pumps turned on and the water supply pressure, so as to achieve energy saving and consumption reduction and improve water supply efficiency. In addition, the operation adjustment and optimization of the pump station based on the prediction results can avoid over-allocation or under-allocation of resources, reduce energy consumption and costs, and improve the stability and reliability of water supply services. At the same time, it also provides a scientific basis for the management and decision-making of the pump station, and promotes the intelligent and refined operation and management of the pump station.

[0093] This method can improve the accuracy of predicted water consumption, thereby effectively optimizing the operation of the pump station.

[0094] Embodiment 2

[0095] In order to better illustrate the accuracy of the water consumption prediction model constructed in this method, the following control experiment is used to illustrate it.

[0096] Test model experimental framework

[0097] As a pre-arrangement of the pump station optimization strategy, the water supply forecasting performance has an important impact on the effect of the optimization scheme. Therefore, the BP model and LSTM model, which are mature in the field of water supply forecasting, are selected as comparative research objects. The VMD modal decomposition is combined with the traditional forecasting model to construct a combined forecasting model, and the seasonal influence factor is added. The purpose is to explore and select a combined forecasting model with high prediction accuracy and good seasonal stability from different dimensions. The main research ideas are as follows: Figure 4 shown.

[0098] The comparative experiment selected the hourly water consumption data of the six months of summer and winter, counting 30 days before the day with the highest water consumption, as the original data source. The water consumption curve after preprocessing the data samples is as follows Figure 5 shown.

[0099] The water consumption data of the previous 29 days were designated as the model training set for learning and training of the prediction model; the water consumption data of the highest day were used as the test set of data samples to evaluate the prediction performance of the model. The number of water consumption data samples in summer and winter were 769 and 744 respectively. The basic situation of the experimental data is shown in Table 1.

[0100] Table 1

[0101]

[0102] It can be seen from Table 1 that the difference between the maximum and minimum water consumption in summer and winter is 1090.69 and 797.99 respectively, indicating that the water consumption fluctuates greatly within the 30-day time scale. The difference between the average water consumption in summer and winter is 210.8, indicating that the water consumption in summer and winter is quite different. In order to obtain a more accurate prediction result, it is necessary to further process the sample data using the normalization method, and then use the neural network model to predict the water consumption.

[0103] Single prediction model experiment (BP / LSTM)

[0104] In this experiment, the BP model and LSTM model are set to perform single-model prediction on the data, so as to obtain the single-model water consumption prediction results to explore the improvement effect of the combined VMD method on the single-model prediction performance.

[0105] 1) BP single model prediction

[0106] The final hourly prediction effect of the BP neural network single model on the highest daily water consumption in summer and winter is as follows: Figure 6 shown.

[0107] 2) LSTM single model prediction

[0108] The final hourly prediction effect of the LSTM neural network single model on the highest daily water consumption in summer and winter is as follows Figure 7 shown.

[0109] 3) Analysis of single model prediction results

[0110] Using the evaluation index table in Section 2.1.3 above, combined with the prediction results of the maximum daily water consumption of the BP model and the LSTM model, the prediction accuracy of the two single models was evaluated. The evaluation index calculation results are shown in Table 2:

[0111] Table 2

[0112]

[0113] From the above prediction results, we can see that the MAPE values ​​of a single BP model or a single LSTM model are in the range of 11% to 13%; the prediction accuracy differences in different seasons are 2.2% and 0.5% respectively, and the single LSTM model performs better than the BP model in terms of seasonal stability. The two single models are "better" prediction models in the prediction performance evaluation, and neither of them reaches the MAPE value <10%, the performance evaluation standard of "accurate" prediction.

[0114] Combined prediction model experiment (VMD-BP / LSTM)

[0115] 1) VMD-BP model

[0116] Using the above method, the original data is decomposed by VMD and then predicted. Different BP models and LSTM models are set to be combined respectively, and the prediction effect of each combination model is tested. The final hourly prediction results of the VMD-BP model for the highest daily water consumption in summer and winter are as follows: Figure 8 shown.

[0117] 2) VMD-LSTM model

[0118] The final hourly prediction effect of VMD-LSTM combined model on the highest daily water consumption in summer and winter is as follows: Fig. 9 shown.

[0119] Analysis of combined model prediction results

[0120] The evaluation indicators of the maximum daily water consumption prediction results of the VMD-BP combination model and the VMD-LSTM combination model are summarized as shown in Table 3:

[0121] Table 3

[0122]

[0123] It can be seen from the prediction results that the prediction accuracy of the combined model decomposed by the VMD method is significantly improved compared with the single model. Compared with the BP single model, the prediction accuracy of the VMD-BP combined model is improved by about 2% on average; compared with the LSTM single model, the prediction accuracy of the VMD-LSTM combined model is more significantly improved, with an average increase of about 6%, and the MAPE values ​​in summer and winter are both less than 10%, reaching the evaluation standard of the "accurate" prediction model. In terms of seasonal stability: the accuracy difference of the VMD-BP combined model is 2.6%, which is slightly lower than the stability of the BP single model by 0.4%; the accuracy difference of the VMD-LSTM combined model is 0.1%, which is slightly higher than the stability of the LSTM single model by 0.4%.

[0124] Analysis and comparison of experimental results

[0125] Both LSTM model and BP model are widely used in the field of short-term water consumption forecasting. This method uses VMD mode decomposition method to combine and improve the above two forecasting models, explore the space where the combined model can improve the accuracy compared with the single model prediction, and explore the stability of the prediction performance of each model under seasonal temperature differences. The evaluation indicators of the prediction results of each model in the comprehensive comparison experiment are summarized as shown in Table 4.

[0126] Table 4

[0127]

[0128] Plot the data in the table into a columnar contrast Fig.10 and Fig.11 shown.

[0129] Through comparative data analysis of the charts, it can be seen that in terms of the prediction performance of the single model, there is not much difference in the prediction accuracy between the BP model and the LSTM model. In terms of seasonal stability, the performance of the LSTM model is slightly better than that of the BP model; the prediction performance of the combined model is significantly better than that of the single model, among which the VMD-LSTM model has the best prediction performance, and its prediction results can reach the "accurate" evaluation standard regardless of summer or winter. In terms of seasonal stability, the VMD-LSTM model is still slightly better than the VMD-BP model, but this advantage has not increased or decreased significantly due to the addition of the VMD decomposition method.

[0130] In general, the prediction accuracy of the model is improved by the VMD method, and the accuracy improvement effect of the VMD-LSTM model is significantly better than that of the VMD-BP model. At the same time, it is also found that the VMD method has not made much improvement in seasonal stability. The difference in model types may be the main factor affecting stability. The comparative experimental results show that the LSTM model is significantly better than the BP model in terms of prediction accuracy stability in different seasons. It can be considered that VMD-LSTM is a combined prediction model that is more suitable for providing advance water volume prediction data for water pump optimization operation strategy.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting water supply of a pumping station for optimizing operation of a pumping station, characterized in that: The following steps are involved: S1. Obtain historical data; the historical data includes water consumption and temperature in a time series; and calculate the daily temperature difference impact factor of the corresponding time series according to the temperature data according to a preset calculation method; S2. Construct a water consumption prediction model to predict the water consumption of the time series in the next X hours based on the water consumption, temperature and daily temperature difference influencing factors; S3, using the time series water consumption, temperature and daily temperature difference influencing factor data of S1 as training data, to train the water consumption prediction model constructed by S2; S4. Obtain the actual water consumption and temperature, and calculate the real-time daily temperature difference influencing factor according to a preset method; Then, the trained water consumption prediction model is used to make real-time water consumption predictions. S5. Adjust and optimize the operation of the pump station based on the water consumption prediction results obtained in S4.

2. The method for predicting water supply of a pumping station for optimizing operation of a pumping station according to claim 1, characterized in that: The preset calculation method for calculating the daily temperature difference impact factor is: what, n-1 =q n-1 +509.13·α·(T2-T1); In the formula, is the daily temperature difference influencing factor; q n-1 is the measured water consumption on the n-1th day; T2 is the average temperature on the n-1th day; T1 is the average temperature of the previous N days; T 25℃ Indicates 25℃.

3. The method for predicting water supply of a pumping station for optimizing operation of a pumping station according to claim 1, characterized in that: The water consumption prediction model includes VMD module and neural network module; The working process of the water consumption prediction model includes: the VMD module decomposes the input time series data into several imf components, and then sends them to the neural network module to obtain the corresponding several prediction components; Afterwards, each prediction component is superimposed to obtain the water consumption prediction result.

4. The method for predicting water supply of a pumping station for optimizing operation of a pumping station according to claim 3, characterized in that: The neural network module is an LSTM neural network model or a BP neural network model.

5. The method for predicting water supply of a pumping station for optimizing operation of a pumping station according to claim 3, characterized in that: When the water consumption prediction model is working, each imf component is normalized before being sent to the neural network module.

6. The method for predicting water supply of a pumping station for optimizing operation of a pumping station according to claim 5, characterized in that: When the water consumption prediction model is working, after obtaining the corresponding prediction components, each prediction component is also denormalized and then superimposed.

7. The method for predicting water supply of a pumping station for optimizing operation of a pumping station according to claim 1, characterized in that: In S1, the acquired historical data is also preprocessed, and the preprocessing includes outlier processing and missing value filling.

8. The method for predicting water supply of a pumping station for optimizing operation of a pumping station according to claim 1, characterized in that: 5≤X≤24。

Citation Information

Patent Citations

  • River flow forecasting method

    CN108647807A

  • GRU neural network-based residential water consumption prediction system and prediction method

    CN113128754A

  • VMD-GA-BiLSTM-based monthly precipitation prediction method

    CN116973990A

Cited By

  • Intelligent pipe network pressure control method and system

    CN120386399A