Water intake pumping station energy consumption prediction and water supply safety early warning method and system based on intelligent algorithm

Abstract

The invention discloses a water intake pumping station energy consumption prediction and water supply safety early warning method and system based on an intelligent algorithm. The method comprises the following steps: constructing a pumping station energy consumption prediction model based on a multi-layer perceptron neural network and a water pump characteristic curve basis model by using pumping station historical operation data; establishing a data evaluation mechanism, evaluating the historical data volume and quality of the predicted pump combination, and comprehensively considering the influence of the problems of aging, cutting, spraying and the like of the water pump on the energy consumption prediction of the pump station in long-term operation of the pump station; a genetic algorithm is utilized to optimize weight threshold values of the multi-layer perceptron neural network and the water pump characteristic curve basis model in the pump station energy consumption prediction model, and the energy consumption of the water intake pump station is predicted; and combining a raw water pipeline pressure measuring point pressure model prediction result with a raw water pipeline pressure monitoring mechanism, establishing a water supply safety early warning model, and comprehensively ensuring the raw water supply safety. The pump station energy consumption prediction accuracy can be improved, and meanwhile the water supply safety of the long-distance raw water conveying pipeline is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of energy consumption prediction of pump stations and water supply safety, and particularly to a method and system for energy consumption prediction and water supply safety early warning of intake pump stations based on intelligent algorithms. Background Art

[0002] In modern urban water supply systems, intake pump stations, as key infrastructure, have a crucial impact on the stability and economy of the entire water supply network in terms of their energy consumption and operating efficiency. With the acceleration of the urbanization process and the increasing requirements for water resource management, how to effectively predict and manage the energy consumption of intake pump stations and ensure water supply safety has become an important challenge faced by urban water supply systems.

[0003] Currently, there are various methods for predicting the energy consumption of pump stations. Among them, polynomial fitting and neural network models are two relatively common ones. The polynomial fitting method is widely used for its simple and fast calculation. However, it has deficiencies in calculation accuracy in the non-efficient section and is difficult to accurately capture the complex non-linear relationships in the operation of intake pump stations. In addition, problems such as the aging of pump station pumps and the deviation of the actual working conditions of the pumps after cutting from the factory pump characteristic curves also limit the application of the polynomial fitting method in practice. The neural network model has strong non-linear fitting capabilities and the advantages of real-time update and iteration, and can better adapt to the changes in the operating conditions of intake pump stations. However, this model highly depends on the quality and quantity of data and is prone to overfitting problems. Especially when the data samples are insufficient or the data quality is poor, its prediction accuracy and reliability will be greatly affected.

[0004] At the same time, some water plants have problems of long-distance water supply due to the excessive distance between the water intake source and the actual location of the water plant. Compared with short-distance water supply, long-distance water supply faces more water supply safety risks. During the long-distance transportation process of the raw water pipeline, it is easily interfered by external factors such as pipeline leakage, water quality pollution, and water pressure fluctuations. These risks will not only affect the safety and stability of the entire water supply network, but also increase the water treatment cost and energy consumption. Therefore, for the problem of long-distance water supply, a more efficient and accurate water supply safety early warning method is needed to ensure the reliable operation of the water supply system. Summary of the Invention

[0005] Based on this, in view of the above technical problems, the present invention proposes a method and system for energy consumption prediction and water supply safety early warning of intake pump stations based on intelligent algorithms.

[0006] The first object of the present invention is to provide a method for energy consumption prediction and water supply safety early warning of intake pump stations based on intelligent algorithms.

[0007] The second object of the present invention is to provide a system for energy consumption prediction and water supply safety early warning of intake pump stations based on intelligent algorithms.

[0008] The first object of the present invention can be achieved by adopting the following technical solutions:

[0009] An energy consumption prediction and water supply safety warning method for a water intake pumping station based on intelligent algorithms, comprising the following steps:

[0010] Utilize the historical operation data of the pumping station to construct an energy consumption prediction model for the pumping station based on a multi-layer perceptron neural network model and a pump characteristic curve principle model;

[0011] Establish a data evaluation mechanism to evaluate the historical data volume and quality of the predicted pump combination, and establish a pump health assessment mechanism, comprehensively considering the impact of problems such as pump aging, cutting, and spraying on the energy consumption prediction of the pumping station during the long-term operation of the pumping station;

[0012] Optimize the weight thresholds of the multi-layer perceptron neural network and the pump characteristic curve principle model in the energy consumption prediction model of the pumping station using the genetic algorithm to predict the energy consumption of the water intake pumping station;

[0013] Utilize a long short-term memory neural network to capture the time series dynamic change law in the historical operation data of the pumping station and construct a pressure prediction model for the raw water pipeline pressure measuring point;

[0014] Combine the prediction results of the raw water pipeline pressure measuring point pressure model with the raw water pipeline pressure monitoring mechanism to establish a water supply safety warning model to comprehensively ensure the safety of raw water supply.

[0015] Furthermore, the multi-layer perceptron neural network model uses the backpropagation algorithm, consists of an input layer, multiple hidden layers, and an output layer, and during training, the error of the output layer is backpropagated to the hidden layer and the input layer to adjust the neural network weights and biases.

[0016] Furthermore, the pump characteristic curve principle model is fitted using a polynomial, and the polynomial fitting formula is:

[0017] P n = a0 + a1Q + a2Q 2 + a3Q 3 + …… + a m Q m (1)

[0018] Where: P represents the power consumption of the pump group in the water intake pumping station during water supply; Q represents the water supply volume in the water intake pumping station;

[0019] a0, a1, a2 …… a m Solve according to the least squares method to satisfy the normal equations:

[0020]

[0021]

[0022] Further, the pump station energy consumption prediction model P(Q) constructed by connecting the multi-layer perceptron neural network model f(Q) and the pump characteristic curve principle model g(Q) is as follows:

[0023] P(Q) = ω1f(Q) + ω2g(Q) (3)

[0024] Where: ω1 and ω2 are the weight thresholds of the multi-layer perceptron neural network model and the pump characteristic curve principle model respectively.

[0025] Further, for the energy efficiency calculation model for establishing the degree of pump aging, the calculation formula for the efficiency model after pump aging is as follows:

[0026]

[0027] Where: η′ i is the efficiency of the i-th pump considering aging; η i is the initial efficiency of the i-th pump; k i is the aging coefficient of the i-th pump; t i is the service life of the i-th pump.

[0028] Taking into account the effects of pump aging, spraying, and cutting on the pump usage efficiency, binary variables are used to regulate the cutting and spraying conditions of the pump. Let y i and z i represent (1 indicates that the corresponding operation has been performed, 0 indicates not performed). Then, the calculation formula for the pump efficiency model considering comprehensive factors is as follows:

[0029]

[0030] Where: η″ i is the efficiency of the i-th pump considering factors such as cutting, spraying, and aging; α is the coefficient for improving the pump efficiency by cutting; β is the coefficient for improving the pump efficiency by spraying. For the total pump efficiency model of the pump combination in the pump station, the weighted average method is used, and its weight is the flow rate of each pump. The calculation formula is as follows:

[0031]

[0032] Where: Q i is the flow rate of the i-th pump; n is the total number of pumps in the pump station.

[0033] Further, one of the constraint conditions for the genetic algorithm to optimize the weight thresholds ω1 and ω2 of the pump station energy consumption prediction P(Q) includes:

[0034] In the data preprocessing stage, data cleaning and data verification are carried out on the historical operation data of the water intake pump station to ensure the accuracy and integrity of the data. During the optimization process of the genetic algorithm, an evaluation mechanism for predicting the data volume and data quality of the pump combination is established. Based on the grade results obtained from the data evaluation mechanism, selective regulation is performed on the weight thresholds ω1 and ω2 of the pump station energy consumption prediction model P(Q). If the obtained evaluation grade result is excellent or good, the weight threshold of ω1 is selected for strengthening, and the weight threshold of ω2 is weakened correspondingly; on the contrary, when the obtained evaluation result is qualified or even unqualified, the weight threshold of ω2 is selected for strengthening, and the weight threshold of ω1 is weakened correspondingly.

[0035] Furthermore, the second constraint condition for the weight thresholds ω1 and ω2 of the genetic algorithm to optimize the pump station energy consumption prediction P(Q) includes:

[0036] The pump efficiency calculated based on the total pump efficiency model of the pump combination in the water intake pump station

[0037] η 总 is compared with the average energy efficiency of the actual pump combination in the water intake pump station to establish a pump health assessment mechanism. According to the feedback results, under the premise of ensuring the prediction accuracy of the pump station energy consumption prediction model, selective regulation is performed on the weight thresholds ω1 and ω2 of the pump station energy consumption prediction model P(Q). If the result of the obtained assessment mechanism is excellent or good, the weight threshold of ω1 is selected for strengthening, and the weight threshold of ω2 is weakened correspondingly; on the contrary, when the result of the obtained assessment mechanism is qualified or even unqualified, the weight threshold of ω2 is selected for strengthening, and the weight threshold of ω1 is weakened correspondingly.

[0038] Furthermore, the application of the genetic algorithm includes:

[0039] Initializing the population: Combining the data evaluation mechanism and the pump health assessment mechanism grade evaluation results in Constraint Condition 1 and Constraint Condition 2, if the weight threshold regulation conclusions obtained under the evaluation of Constraint 1 and Constraint 2 for the historical data of the pump combination to be predicted are consistent, then multiple groups of weight thresholds ω = (ω1, ω2) are randomly generated based on the constraint regulation requirements. If the regulation conclusions obtained from Constraint 1 and Constraint 2 are inconsistent, then the evaluation conclusion of Constraint 1 is given priority, and multiple groups of weight thresholds ω = (ω1, ω2) are randomly generated. And in either case, the relationship between the weight thresholds ω1 and ω2 satisfies 0 < ω1 + ω2 ≤ 2.

[0040] Fitness evaluation: Taking the prediction accuracy of the pump station energy consumption prediction model as the optimization goal, the mean square error (MSE) and the mean absolute error (MAE) are used as the fitness functions for the prediction accuracy of the pump station energy consumption prediction model to calculate the fitness of each group of weight thresholds to measure the quality of the solutions.

[0041] Operational selection: Based on the fitness values of each group of weight thresholds, select the group with a higher fitness value to enter the next iteration.

[0042] Crossover operation: Based on single-point crossover, randomly select two groups of weight thresholds, and exchange ω1 and ω2 pairwise to generate two new groups of weight thresholds.

[0043] Mutation operation: Based on Gaussian mutation, add a random value drawn from a Gaussian distribution to introduce mutation, and introduce the idea of non-uniform mutation. The above Gaussian mutation amplitude gradually decreases with the increase of the number of iterations.

[0044] Fitness evaluation and update: Evaluate the fitness values of the generated offspring weight thresholds in each round, and screen and eliminate the offspring with low fitness to ensure the stability of the population size.

[0045] Termination condition judgment: Continue to apply the genetic algorithm to the offspring weight threshold groups, and perform multiple iterations until the fitness value of the offspring converges to a certain stable value or reaches the maximum preset number of iterations.

[0046] Furthermore, the fitness value is calculated as follows:

[0047]

[0048] where: ∈ is a very small positive integer to prevent the denominator from being zero, and its influence on the calculation of the fitness value can be ignored.

[0049] Furthermore, the Gaussian mutation formula is as follows:

[0050] ω′ i =ω i +N(α, β) (9)

[0051] where: in N(α, β), α represents the mean of ω i and β is the Gaussian mutation amplitude.

[0052] Furthermore, use the long short-term memory neural network to capture the hourly periodic characteristics in the preprocessed historical operation data of the pumping station, determine the optimal window length and step out, improve the ability of the model to capture the long-term dependence of time series data, and the trained model is used to predict the pressure value of the pressure measuring point on the raw water pipeline section.

[0053] Furthermore, perform statistical analysis on the historical data of the pumping station, calculate the average pressure of each pressure measuring point on the raw water pipeline standard deviation σ, and combine the characteristic curve of the raw water pipeline to determine the theoretical pressure value of each pressure measuring point, and set the preset safety valve value of each pressure measuring point on the raw water pipeline section based on this.

[0054] Further, based on the relationship between the predicted pressure value at the pressure measurement point and the preset safety valve value, a multi-dimensional early warning response mechanism is established, which is divided into three levels of early warning modes according to different situations: [[ID=HID=2]]

[0055] Level 1 early warning: When the predicted pressure threshold is about to exceed the preset initial safety threshold a warning text message is triggered to notify and remind the on-duty personnel to enter the alert state and pay attention to the dynamic changes of the subsequent predicted pressure values at the pressure measurement points;

[0056] Level 2 early warning: When the predicted pressure value at the pressure measurement point continuously exceeds the preset safety valve value for 30 minutes, an alarm instruction is triggered and the on-duty personnel are dispatched to the site to check the situation;

[0057] Level 3 early warning: When the predicted pressure value at the pressure measurement point continuously exceeds the preset safety valve value for a long time and the incoming plant flow rate significantly decreases (>12%), the on-duty personnel are immediately dispatched to the site to check, and the pump station dispatching group and the on-site emergency repair group enter the alert state and wait for orders at any time.

[0058] The second object of the present invention can be achieved by adopting the following technical solutions:

[0059] A water intake pump station energy consumption prediction and water supply safety early warning system based on an intelligent algorithm, the system includes:

[0060] A data collection and preprocessing layer, which involves data collection, preprocessing, and storage. The data sources include real-time sensor reception and historical data obtained from the database. Data preprocessing involves steps such as cleaning, denoising, normalization, and verification. At the same time, the collected data has a temporary storage function;

[0061] A data transmission layer, which involves data transmission protocols and data security. The transmission protocols include 4G / 5G networks, Wi-Fi, and LoRa protocols, and data security is ensured through encrypted transmission;

[0062] A model construction and model management layer, which involves multiple models, including a data evaluation model, a pump health assessment model, a pump station energy consumption prediction model, a genetic algorithm optimization model, a pressure measurement point pressure monitoring model, and a water supply safety early warning model;

[0063] An application display and user interaction layer, which involves outputting model prediction results, visual presentation, and user interaction. The result output interface includes pump station energy consumption prediction and pressure measurement point pressure prediction. The visual interface includes an energy consumption curve graph, a pressure curve graph, and an early warning status graph. The interaction interface provides functions such as parameter adjustment, historical data query, and early warning information notification.

[0064] The present invention has the following beneficial effects compared with the prior art:

[0065] The energy consumption prediction model of the pump station constructed based on the multi-layer perceptron neural network and the basic principle model of the pump characteristic curve effectively solves the problems existing in the existing methods for predicting the energy consumption of the pump station, improves the accuracy of the energy consumption prediction of the pump station, greatly reduces the problem that the prediction accuracy of the model highly depends on the quantity and quality of data, and at the same time effectively solves the influence of the aging of the pump during use, regular maintenance spraying, cutting and other operations on the prediction accuracy of the pump station energy consumption. At the same time, aiming at the problem of long-distance water supply in the raw water pipeline section, the ability of the long short-term memory neural network to capture long-term dependencies is used to more efficiently and accurately predict the pressure value of the pressure measuring point in the raw water pipeline section, establish a more perfect water supply safety warning mechanism, and ensure the safety of raw water supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a schematic flow chart of the energy consumption prediction and water supply safety warning method for the raw water pump station based on the intelligent algorithm in Embodiment 1 of the present invention;

[0067] Figure 2 It is a schematic flow chart of the evaluation process of the data evaluation model in Embodiment 1 of the present invention;

[0068] Figure 3 It is a schematic flow chart of the evaluation process of the pump health assessment model in Embodiment 1 of the present invention;

[0069] Figure 4 It is a schematic flow chart of the genetic algorithm for optimizing the energy consumption prediction model of the raw water pump station in Embodiment 1 of the present invention;

[0070] Figure 5 It is a schematic flow chart of the water supply safety warning mechanism for the raw water pipeline section in Embodiment 1 of the present invention;

[0071] Figure 6 It is an architecture diagram of the energy consumption prediction and water supply safety warning system for the raw water pump station based on the intelligent algorithm in Embodiment 2 of the present invention. SPECIFIC IMPLEMENTATION METHOD

[0073] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0074] Embodiment 1:

[0075] As Figure 1 shown, this embodiment provides an energy consumption prediction and water supply safety warning method for the raw water pump station based on the intelligent algorithm, including the following steps:

[0076] S01. Based on the historical operation data of the pumping station, an energy consumption prediction model for the pumping station is constructed using a multi-layer perceptron neural network model and a pump characteristic curve principle model.

[0077] In this embodiment, the energy consumption prediction model P(Q) of the pumping station constructed by connecting the multi-layer perceptron neural network model f(Q) and the pump characteristic curve principle model g(Q) is as follows:

[0078] P(Q) = ω1f(Q) + ω2g(Q) (3)

[0079] Where: ω1 and ω2 are the weight thresholds of the multi-layer perceptron neural network model and the pump characteristic curve principle model respectively.

[0080] Specifically, the multi-layer perceptron neural network model uses the backpropagation algorithm, consists of an input layer, multiple hidden layers and an output layer, and adopts an improved Sigmiod activation function to overcome the problem of easy disappearance of the deep network gradient, improve the model's ability to capture deep features, and during training, the error of the output layer is propagated back to the hidden layer and the input layer to adjust the neural network weights and biases. The parameters used in the model training process include: the opening and closing conditions of the pump combination in the water intake pumping station, the water intake volume, the power consumption of production, the output head of the pump combination, etc.

[0081] It can be understood that the water intake pumping station has a large amount of historical operation data, which can be used for the training of the multi-layer perceptron neural network model, improve the prediction accuracy of the model, and be used later to construct the energy consumption prediction model of the water intake pumping station.

[0082] The improved Sigmiod activation function will be named the Adaptive-Sigmiod function, and the function expression is as follows:

[0083]

[0084] Where: a > 1, the steepness of the curve is controlled by adjusting a to enhance the gradient amplitude; b > 0, by introducing the linear term bx, the output saturation is avoided.

[0085] Specifically, the pump characteristic curve principle model uses a polynomial for fitting, and the polynomial fitting formula is:

[0086] P n = a0 + a1Q + a2Q 2 + a3Q 3 + …… + a m Q m (1)

[0087] Where: P represents the power consumption of water supply when the pump group in the water intake pumping station is working; Q represents the water supply volume of the water intake pumping station.

[0088] a0, a1. a2……a m Solve according to the least squares method to satisfy the normal equations:

[0089]

[0090] S02. Establish a data evaluation mechanism to evaluate the historical data volume and quality of the predicted pump combination, and establish a pump health evaluation mechanism, comprehensively considering the impacts of pump aging, cutting, spraying, etc. on the pump station energy consumption prediction during the long-term operation of the pump station.

[0091] In this embodiment, an energy efficiency calculation model for the degree of pump aging is established, and the calculation formula for the efficiency model after pump aging is as follows:

[0092]

[0093] Where: η′ i is the efficiency of the i-th pump considering aging; η i is the initial efficiency of the i-th pump; k i is the aging coefficient of the i-th pump; t i is the service life of the i-th pump.

[0094] Comprehensively considering the impacts of pump aging, spraying, and cutting on the pump usage efficiency, use binary to control the variables of pump cutting and spraying conditions, and use y i , z i to represent (1 means the corresponding operation has been carried out, 0 means not carried out). Then, the calculation formula for the pump efficiency model under comprehensive factor consideration is as follows:

[0095]

[0096] Where: α is the coefficient for improving the pump efficiency by cutting; β is the coefficient for improving the pump efficiency by spraying.

[0097] Take the total pump efficiency model of the pump combination in the pump station and use the weighted average method, with its weight being the flow rate of each pump. The calculation formula is as follows:

[0098]

[0099] Where: Q i is the flow rate of the i-th pump; n is the total number of pumps in the pump station.

[0100] As Figure 2 shown, in this embodiment, using the real-time data of the pump station received by the sensor and the historical data in the database, comprehensively evaluate the data quantity and quality of different pump combinations, and comprehensively obtain the evaluation result. This evaluation result will be updated regularly and recorded on file.

[0101] AsFigure 3 As shown, in this embodiment, real-time data of the water intake pump station received by sensors and historical data in the database are also used. Based on the established water pump health evaluation mechanism, the health of the water pump is evaluated, and the evaluation results are comprehensively obtained. Information such as water pump cutting, spraying, and aging is regularly reminded to be updated, and the water pump health evaluation results are updated accordingly based on the updated results and recorded on file.

[0102] S03. Use the genetic algorithm to optimize the weight thresholds of the multi-layer perceptron neural network and the water pump characteristic curve theory model in the pump station energy consumption prediction model to predict the energy consumption of the water intake pump station.

[0103] In this embodiment, the evaluation results of the pump station data evaluation mechanism and the water pump health evaluation mechanism are used as the first constraint condition and the second constraint condition for optimizing the pump station energy consumption prediction model by the genetic algorithm, so as to regulate and generate the initial weight threshold data group.

[0104] Specifically, in the data preprocessing stage, data cleaning and data verification are performed on the historical operation data of the water intake pump station to ensure the accuracy and integrity of the data. During the optimization process of the genetic algorithm, an evaluation mechanism for the predicted pump combination data volume and data quality is established. Based on the grade results obtained from the data evaluation mechanism, selective regulation is performed on the weight thresholds ω1 and ω2 of the pump station energy consumption prediction model P(Q). If the obtained evaluation grade result is excellent or good, the weight threshold of ω1 is selected to be strengthened, and the weight threshold of ω2 is weakened correspondingly; on the contrary, when the obtained evaluation result is qualified or even unqualified, the weight threshold of ω2 is selected to be strengthened, and the weight threshold of ω1 is weakened correspondingly.

[0105] Specifically, based on the water pump efficiency η calculated by the total water pump efficiency model of the pump combination of the water intake pump station 总 and the average energy efficiency of the actual pump combination in the water intake pump station a water pump health evaluation mechanism is established for comparison. According to the feedback results, under the premise of ensuring the prediction accuracy of the pump station energy consumption prediction model, selective regulation is performed on the weight thresholds ω1 and ω2 of the pump station energy consumption prediction model P(Q). If the obtained evaluation mechanism result is excellent or good, the weight threshold of ω1 is selected to be strengthened, and the weight threshold of ω2 is weakened correspondingly; on the contrary, when the obtained evaluation mechanism result is qualified or even unqualified, the weight threshold of ω2 is selected to be strengthened, and the weight threshold of ω1 is weakened correspondingly.

[0106] As Figure 4 shown, in this embodiment, the application of the genetic algorithm to optimize the weight thresholds ω1 and ω2 of the water intake pump station energy consumption prediction model is as follows:

[0107] Initializing the population: Combining the data evaluation mechanisms in Constraint 1 and Constraint 2 and the evaluation results of the pump health assessment mechanism level, if the weight threshold regulation conclusions obtained under the evaluation of Constraint 1 and Constraint 2 of the pump combination historical data to be predicted are consistent, then multiple groups of weight thresholds ω = (ω1, ω2) are randomly generated based on the constraint regulation requirements. If the regulation conclusions obtained from Constraint 1 and Constraint 2 are inconsistent, then the evaluation conclusion of Constraint 1 is given priority, and multiple groups of weight thresholds ω = (ω1, ω2) are randomly generated. And in either case, the relationship between the weight thresholds ω1 and ω2 satisfies 0 < ω1 + ω2 ≤ 2.

[0108] Fitness evaluation: Taking the prediction accuracy of the pump station energy consumption prediction model as the optimization goal, using the mean square error (MSE) and the mean absolute error (MAE) as the fitness functions for the prediction accuracy of the pump station energy consumption prediction model, calculate the fitness of each group of weight thresholds to measure the quality of the solutions.

[0109] Operational selection: According to the fitness values of each group of weight thresholds, select the groups with higher fitness values to enter the next iteration.

[0110] Crossover operation: Based on single-point crossover, randomly select two groups of weight thresholds, and exchange ω1 and ω2 pairwise to generate two new groups of weight thresholds.

[0111] Mutation operation: Based on Gaussian mutation, add random values drawn from a Gaussian distribution to introduce mutations, and introduce the idea of non-uniform mutation. The above Gaussian mutation amplitude gradually decreases as the number of iterations increases.

[0112] Fitness evaluation and update: Evaluate the fitness values of the offspring weight thresholds generated in each round, and screen and eliminate the offspring with low fitness to ensure the stability of the population size.

[0113] Termination condition judgment: Continue to apply the genetic algorithm to the offspring weight threshold groups, and perform multiple iterations until the fitness values of the offspring converge to a certain stable value or reach the maximum preset number of iterations.

[0114] Specifically, the fitness value is calculated as follows:

[0115]

[0116] where: ∈ is a very small positive integer to prevent the denominator from being zero, and its influence on the calculation of the fitness value can be ignored.

[0117] Specifically, the Gaussian mutation formula is as follows:

[0118] ω′ i =ω i +N(α, β) (9)

[0119] wherein: in N(α, β), α represents the ω i mean value, and β is the Gaussian mutation amplitude.

[0120] S04. Use a long short-term memory neural network to capture the time series dynamic change law in the historical operation data of the pumping station, and construct a pressure prediction model for the pressure measuring points of the raw water pipeline.

[0121] In this embodiment, a long short-term memory neural network is used to capture the hourly periodic characteristics of data such as the water supply volume of the water intake pumping station, the output head of the water intake pumping station, and the pressure value of the pressure measuring point of the raw water pipeline section in the preprocessed historical operation data of the pumping station, determine the optimal window length and step out, improve the ability of the model to capture long-term dependencies in time series data, and the trained model is used to predict the pressure value of the pressure measuring point of the raw water pipeline section.

[0122] Specifically, the construction process of the long short-term memory neural network model mainly includes four core parts: the forgetting gate, the input gate, the output gate, and the cell state (memory state).

[0123] The forgetting gate selectively forgets some information in the cell state by receiving the current input x t and the hidden state h at the previous time step t-1 , and its forgetting ratio is determined by the value output by the Sigmoid function (a value between 0 and 1).

[0124] The calculation formula of the forgetting gate is as follows:

[0125] f t = σ(ω f ·[h t-1 , x t +b f )

[0126] where: f t is the output of the forgetting gate; σ represents the Sigmoid function; ω f is the weight of the forgetting gate; b f is the bias of the forgetting gate.

[0127] The specific formula of the Sigmoid function is as follows:

[0128]

[0129] The input gate enters new information into the cell state. The input gate is mainly regulated by the tanh layer and the Sigmoid layer. Among them, the value output by the Sigmoid layer (a value between 0 and 1) determines the information update ratio of the cell state, that is, reflects the importance of each input feature. The tanh layer generates new information to be added to the cell state and determines which information is finally entered into the cell state for effect.

[0130] The calculation formula of the input gate is as follows:

[0131] i t = σ(ω i ·[h t-1 , x t +b i )

[0132]

[0133] Where: i t is the output of the input gate; is the candidate memory cell; ω i , b i are the weight matrix and bias term of the input gate; ω c , b c are the weight matrix and bias term of the candidate memory cell.

[0134] The specific formula of the tanh function is as follows:

[0135]

[0136] The cell state is a key part of the long short-term memory neural network. The forget gate forgets part of the cell state information, and the input gate inputs new information. Both jointly act on the update of the cell state.

[0137] The cell state update formula is as follows:

[0138]

[0139] Where: C t is the cell state at the current moment; C t-1 is the cell state at the previous moment.

[0140] The output gate takes some information in the cell state as the output at the current moment. It is also jointly regulated by the tanh layer and the Sigmoid layer. The sigmoid layer determines the proportion of the output, and the tanh layer processes the cell state and then multiplies it by the output of the sigmoid layer to obtain the final output.

[0141] The calculation formula of the output gate is as follows:

[0142] o t = σ(ω o ·[h t-1 , x t +b o )

[0143] Where: o t is the output of the output gate; ω o is the weight matrix of the output gate; b o is the bias term of the output gate.

[0144] The further output gate outputs the cell state C at the current moment t Performs information processing to obtain a new hidden state h t , and the calculation formula for the hidden state update process is as follows:

[0145] h t = o t ·tanh(C t )

[0146] The current cell state C t is responsible for storing long-term information and is the backbone of information transmission, while the new hidden state h in the output gate t is responsible for transmitting short-term information. Finally, the output gate transmits C t and h t to the next time window

[0147] It can be understood that the water intake pumping station has the condition of real-time data acquisition, and its historical operation data volume is huge. The long short-term memory neural network model can effectively capture the long-term dependence relationship therein and accurately predict the pressure value of the pressure measuring point in the raw water pipeline section

[0148] S05. Combine the predicted result of the raw water pipeline pressure measuring point pressure model with the raw water pipeline pressure monitoring mechanism to establish a water supply safety early warning model to comprehensively ensure the safety of raw water supply

[0149] In this embodiment, through the statistical analysis of the historical data of the pumping station, calculate the average pressure of each pressure measuring point in the raw water pipeline standard deviation σ, and combine the characteristic curve of the raw water pipeline to determine the theoretical pressure value of each pressure measuring point, and set the preset safety valve value of each pressure measuring point in the raw water pipeline section based on this

[0150] As Figure 5 shown, based on the relationship between the predicted pressure value of the pressure measuring point and the preset safety valve value, establish a multi-dimensional early warning response mechanism, which is divided into three-level early warning modes according to different situations:

[0151] Level 1 early warning: When the predicted pressure threshold is about to exceed the preset initial safety threshold , trigger an early warning text message notification to remind the duty personnel to enter the alert state and pay attention to the dynamic change of the subsequent predicted pressure value of the pressure measuring point

[0152] Level 2 early warning: When the predicted pressure value of the pressure measuring point continuously exceeds the preset safety valve value for 30 minutes, trigger an alarm instruction, and the duty personnel go to the site to check the situation

[0153] Level 3 Early Warning: When the pressure value at the predicted pressure measurement point continuously exceeds the preset safety valve value for a long time and the incoming plant flow rate significantly decreases (>12%), the on-duty personnel shall immediately go to the site for investigation, and the pump station dispatching group and the on-site emergency repair group shall enter a state of alert and await orders at any time.

[0154] Embodiment 2:

[0155] As Figure 6 shown, this embodiment provides an energy consumption prediction and water supply safety early warning system for a water intake pump station based on intelligent algorithms. The system mainly includes the following major modules:

[0156] Data acquisition and preprocessing layer, which involves data acquisition, preprocessing, and storage. The data sources include real-time sensor reception and historical data obtained from the database. Data preprocessing involves steps such as cleaning, denoising, normalization, and verification. At the same time, the acquired data has a temporary storage function;

[0157] Data transmission layer, which involves data transmission protocols and data security. The transmission protocols include 4G / 5G networks, Wi-Fi, and LoRa protocols. Data security is ensured through encrypted transmission;

[0158] Model construction and model management layer, which involves multiple models, including data evaluation model, pump health assessment model, pump station energy consumption prediction model, genetic algorithm optimization model, pressure measurement point pressure monitoring model, and water supply safety early warning model;

[0159] Application display and user interaction layer, which involves output of model prediction results, visual presentation, and user interaction. The result output interface includes pump station energy consumption prediction and pressure measurement point pressure prediction. The visual interface includes energy consumption curve graph, pressure curve graph, and early warning status graph. The interaction interface provides functions such as parameter adjustment, historical data query, and early warning information notification.

Claims

1. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm, characterized in that, It includes the following steps: S01: Construct a pump station energy consumption prediction model based on the multi-layer perceptron neural network model and the pump characteristic curve theory model using the historical operation data of the pump station; S02: Establish a data evaluation mechanism to evaluate the quantity and quality of the historical data of the predicted pump combination; establish a pump health evaluation mechanism, and comprehensively consider the impacts of pump aging, cutting, and spraying problems during the long-term operation of the pump station on the pump station energy consumption prediction; S03: Use the genetic algorithm to optimize the weight thresholds of the multi-layer perceptron neural network and the pump characteristic curve theory model in the pump station energy consumption prediction model, and predict the energy consumption of the water intake pump station; S04: Use the long short-term memory neural network to capture the time series dynamic change law in the historical operation data of the pump station, and construct a pressure prediction model for the pressure measurement points of the raw water pipeline; S05: Combine the prediction results of the pressure model of the pressure measurement points of the raw water pipeline with the raw water pipeline pressure monitoring mechanism to establish a water supply safety early warning model, and comprehensively ensure the safety of the raw water supply.

2. The method for predicting energy consumption of a water intake pumping station and early warning of water supply safety based on an intelligent algorithm according to claim 1, wherein, The multi-layer perceptron neural network model uses the backpropagation algorithm and consists of an input layer, multiple hidden layers, and an output layer. During training, the error of the output layer is backpropagated to the hidden layer and the input layer to adjust the weights and biases of the neural network.

3. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, characterized in that The pump characteristic curve theory model is fitted using a polynomial, and the polynomial fitting formula is: P n = a0 + a1Q + a2Q 2 + a3Q 3 + …… + a m Q m (1) Where: P represents the power consumption of the water intake pump station during water supply; Q represents the water supply volume of the water intake pump station. a0, a1, a2... a m Solve according to the least squares method, satisfying the normal equations:

4. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, characterized in that, The pump station energy consumption prediction model P(Q) constructed by connecting the multi-layer perceptron neural network model f(Q) and the pump characteristic curve theory model g(Q) is as follows: P(Q) = ω1f(Q) + ω2g(Q) (3) Where: ω1 and ω2 are the weight thresholds of the multi-layer perceptron neural network model and the pump characteristic curve theory model respectively.

5. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, characterized in that, Establish an energy efficiency calculation model for the degree of pump aging. The efficiency model calculation formula after pump aging is as follows: where: η′ i is the efficiency of the i-th pump considering aging; η i is the initial efficiency of the i-th pump; k i is the aging coefficient of the i-th pump; t i is the service life of the i-th pump; If the impacts of pump aging, spraying, and cutting on the pump usage efficiency are considered comprehensively, and binary is used to control the variables of pump cutting and spraying conditions, with y i and z i representing (1 indicates that the corresponding operation has been carried out, 0 indicates that it has not), then the calculation formula for the pump efficiency model considering comprehensive factors is as follows: in: is the efficiency of the i-th water pump after taking into account factors such as cutting, spraying, and aging; α is the improvement coefficient of cutting on the water pump efficiency; β is the improvement coefficient of spraying on the water pump efficiency. The total water pump efficiency model of the pump combination of the water pump station uses the weighted average method, and its weight is the flow rate of each water pump. The calculation formula is as follows: Where: Q i is the flow rate of the i-th pump; n is the total number of pumps in the water intake pumping station.

6. The method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, wherein, One of the constraint conditions for the genetic algorithm to optimize the weight thresholds ω1 and ω2 of the pump station energy consumption prediction P(Q) includes: During the data preprocessing stage, clean and verify the historical operation data of the water intake pump station to ensure the accuracy and integrity of the data. During the genetic algorithm optimization process, establish an evaluation mechanism for the quantity and quality of the predicted pump combination data. Based on the grade results obtained from the data evaluation mechanism, selectively regulate the weight thresholds ω1 and ω2 of the pump station energy consumption prediction model P(Q). If the obtained evaluation grade results are excellent or good, select to strengthen the weight threshold of ω1 and correspondingly weaken the weight threshold of ω2; on the contrary, when the obtained evaluation results are qualified or even unqualified, select to strengthen the weight threshold of ω2 and correspondingly weaken the weight threshold of ω1.

7. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 5, characterized in that, Another constraint condition for the genetic algorithm to optimize the weight thresholds ω1 and ω2 of the pump station energy consumption prediction P(Q) includes: The pump efficiency η calculated based on the total pump efficiency model of the pump combination in the water intake pump station 总 is compared with the average energy efficiency of the actual pump combination in the water intake pump station to establish a pump health assessment mechanism. According to the feedback results, under the optimization goal of ensuring the prediction accuracy of the energy consumption prediction model of the water intake pump station, the weight thresholds ω1 and ω2 of the pump station energy consumption prediction model P(Q) are selectively regulated. If the result of the obtained assessment mechanism is excellent or good, then the weight threshold of ω1 is strengthened and the weight threshold of ω2 is weakened correspondingly; on the contrary, when the result of the obtained assessment mechanism is qualified or even unqualified, then the weight threshold of ω2 is strengthened and the weight threshold of ω1 is weakened correspondingly.

8. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, characterized in that, The application of the genetic algorithm includes: Initialization of population: Combining the data evaluation mechanisms in Constraint 1 and Constraint 2 and the evaluation results of the pump health assessment mechanism level, if the weight threshold adjustment conclusions obtained under the evaluation of Constraint 1 and Constraint 2 of the pump combination historical data to be predicted are consistent, then multiple groups of weight thresholds ω=(ω1, ω2) are randomly generated based on the constraint adjustment requirements. If the adjustment conclusions obtained from Constraint 1 and Constraint 2 are inconsistent, then the evaluation conclusion of Constraint 1 is given priority, and multiple groups of weight thresholds ω=(ω1, ω2) are randomly generated. And in either case, the relationship between the weight thresholds ω1 and ω2 satisfies 0<ω1 + ω2 ≤ 2; Fitness evaluation: Taking the prediction accuracy of the pumping station energy consumption prediction model as the optimization goal, using the mean square error (MSE) and the mean absolute error (MAE) as the fitness functions for the prediction accuracy of the pumping station energy consumption prediction model, calculate the fitness of each group of weight thresholds to measure the quality of the solution; Operational selection: According to the fitness values of each group of weight thresholds, select the group with a higher fitness value to enter the next round of iteration; Crossover operation: Based on single-point crossover, randomly select two groups of weight thresholds, and exchange ω1 and ω2 pairwise to generate two new groups of weight thresholds; Mutation operation: Based on Gaussian mutation, add random values drawn from a Gaussian distribution to introduce mutation, and introduce the idea of non-uniform mutation. The above Gaussian mutation amplitude gradually decreases with the increase of the number of iterations; Fitness evaluation and update: Evaluate the fitness values of the offspring weight thresholds generated in each round, and screen and eliminate the offspring with low fitness to ensure the stability of the population size; Termination condition judgment: Continue to apply the genetic algorithm to the offspring weight threshold groups, and iterate multiple times until the fitness value of the offspring converges to a certain stable value or reaches the maximum preset number of iterations; 9. A method for predicting energy consumption of a water intake pump station and warning of water supply safety based on an intelligent algorithm according to claim 8, characterized in that, The fitness value is calculated as follows: Where: ∈ is a very small positive integer to prevent the denominator from being zero, and its influence on the calculation of the fitness value can be ignored; 10. The method for energy consumption prediction and water supply safety early warning of a water intake pump station based on an intelligent algorithm according to claim 8, characterized in that: The Gaussian mutation formula is as follows: ω′ i = ω i + N(α, β) (9) Wherein: in N(α, β), α represents the mean value of ω i and β is the Gaussian mutation amplitude.

11. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, characterized in that Use a long short-term memory neural network to capture the hourly periodic characteristics in the preprocessed historical operation data of the pumping station, determine the optimal window length and step out, improve the ability of the model to capture long-term dependencies in time series data, and the trained model is used to predict the pressure value at the pressure measurement point of the raw water pipe section; 12. The method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, characterized in that, Statistical analysis of historical data of the pumping station to calculate the average pressure of each pressure measuring point in the raw water pipeline The standard deviation σ is calculated and combined with the raw water pipeline characteristic curve to determine the theoretical pressure value of each pressure measuring point. Based on this, the preset safety valve value of each pressure measuring point in the raw water pipeline section is set.

13. A method for predicting the energy consumption of a water intake pumping station and warning of water supply safety based on an intelligent algorithm according to claim 1, characterized in that, Based on the relationship between the predicted pressure value at the pressure measurement point and the preset safety valve threshold, establish a multi-dimensional early warning response mechanism, which is divided into three levels of early warning modes according to different situations: Level 1 Warning: When the predicted pressure threshold is about to exceed the preset initial safety threshold A warning SMS notification is triggered to remind the on-duty personnel to enter a state of alert and pay attention to the dynamic changes in the pressure values at the subsequent predicted pressure measurement points; Secondary warning: When the predicted pressure value at the pressure measurement point continuously exceeds the preset safety valve threshold for 30 minutes, trigger an alarm command, and the duty personnel go to the site to check the situation; Tertiary warning: When the predicted pressure value at the pressure measurement point continuously exceeds the preset safety valve threshold for a long time and the incoming flow rate decreases significantly (>12%), the duty personnel immediately go to the site to check, and the pumping station dispatching group and the on-site emergency repair group enter a state of alert and wait for orders at any time; 14. A water intake pumping station energy consumption prediction and water supply safety early warning system based on intelligent algorithms, characterized in that, It includes the following major modules: Data acquisition and preprocessing layer, which involves data acquisition, preprocessing and storage. The data sources include real-time reception by sensors and obtaining historical data from the database. Data preprocessing involves steps such as cleaning, denoising, normalization and verification. At the same time, the acquired data has a temporary storage function; The data transmission layer involves data transmission protocols and data security. The transmission protocols include 4G / 5G networks, Wi-Fi, and LoRa protocols, and data security is ensured through encrypted transmission; The model construction and model management layer involves multiple models, including a data evaluation model, a pump health assessment model, a pumping station energy consumption prediction model, a genetic algorithm optimization model, a piezometric point pressure monitoring model, and a water supply safety warning model; The application display and user interaction layer involves the output of model prediction results, visual presentation, and user interaction. The result output interface includes pumping station energy consumption prediction and piezometric point pressure prediction. The visual interface includes energy consumption curve graphs, pressure curve graphs, and warning status graphs. The interaction interface provides functions such as parameter adjustment, historical data query, and warning information notification.

Images