Water pump energy-saving operation method based on artificial intelligence

Abstract

The invention relates to the technical field of water pump energy saving, and discloses a water pump energy-saving operation method based on artificial intelligence. Water pump operation data in an area are obtained in real time, after data preprocessing is carried out, a water pump performance prediction model is constructed by adopting a PatchTST-based deep learning algorithm, and meanwhile, a water pump energy consumption regulation and control model is constructed in combination with a graph neural network-based deep learning algorithm. And dynamically adjusting a water pump operation strategy by combining output results of the two models. The method can accurately predict the future performance of the water pump, fully considers the mutual influence and operation characteristics of the water pumps, achieves the dynamic optimization of the operation strategy of the water pump, and remarkably improves the energy-saving effect. In addition, the method further comprises the steps of continuously monitoring the running state and the energy consumption condition of the water pump, feeding back real-time monitoring data to the model for optimization, and ensuring continuous improvement of the prediction precision and the regulation and control effect. And green and sustainable development of an urban water supply system is effectively promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of water pump energy saving, and in particular to an energy-saving operation method of a water pump based on artificial intelligence. Background Art

[0002] In the current urban water supply field, water pumps play an irreplaceable role as key equipment. However, the operation mode of traditional water pumps often lacks flexibility and is difficult to dynamically adjust according to actual needs, resulting in high energy consumption and low efficiency. In addition, the operation status of water pumps is easily affected by many factors, such as ambient temperature, water pump flow, pressure, speed, etc. Changes in these factors may affect the energy consumption and performance of water pumps.

[0003] At present, although some water pump energy-saving technologies and methods have been proposed, most of these methods are based on simple rule control or empirical models, which are difficult to accurately capture the complex changes in the operating status of the water pump and cannot achieve refined energy-saving regulation. Therefore, there is a need for an energy-saving method that can comprehensively consider multiple factors and realize real-time monitoring and intelligent regulation of the operating status of the water pump. Summary of the invention

[0004] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide an energy-saving operation method for a water pump based on artificial intelligence, so as to solve the problems in the prior art of high energy consumption, low efficiency and difficulty in dynamic adjustment of water pumps according to actual needs.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an energy-saving operation method of a water pump based on artificial intelligence, the method comprising: Step S100: acquiring the water pump operation data in the area in real time; Step S200: preprocessing the water pump operation data; Step S300: using a deep learning algorithm based on PatchTST, combined with pre-processed historical water pump operation data and historical water pump performance data, to build a water pump performance prediction model; Step S400: using a deep learning algorithm based on a graph neural network, combining pre-processed historical water pump operation data, historical water pump energy consumption data, and historical regional water flow demand data, to build a water pump energy consumption control model; Step S500: dynamically adjust the water pump operation strategy of the area in combination with the output results of the water pump performance prediction model and the water pump energy consumption control model.

[0006] Preferably, in a possible implementation manner of the first aspect, the water pump operation data includes water pump position coordinates, water pump flow, water pump pressure, water pump speed, water pump motor current, water pump motor voltage and water pump ambient temperature.

[0007] Preferably, in a possible implementation manner of the first aspect, the data preprocessing includes data cleaning, data denoising and data normalization.

[0008] Preferably, in a possible implementation of the first aspect, the historical water pump operation data includes water pump position coordinates, historical water pump flow, historical water pump pressure, historical water pump speed, historical water pump motor current, historical water pump motor voltage, historical water pump ambient temperature and timestamp.

[0009] Preferably, in a possible implementation manner of the first aspect, the structure of the water pump performance prediction model includes: The first input layer is used to input pre-processed historical water pump operation data and historical water pump performance data, wherein the historical water pump performance data includes historical water pump flow, historical water pump head, historical water pump power and historical water pump efficiency; The PatchTST feature extraction layer is used to split the input data into multiple time series subsequences and perform feature extraction on each subsequence to capture the local and global features of the time series data; The self-attention mechanism layer assigns dynamic weights to each subsequence through the self-attention mechanism to obtain the long-term dependencies of time series data; The first output layer receives the output of the PatchTST feature extraction layer and the self-attention mechanism layer, and outputs the pump flow prediction, pump head prediction, pump power prediction, and pump efficiency prediction within a preset time in the future.

[0010] Preferably, in a possible implementation manner of the first aspect, the structure of the water pump energy consumption control model includes: The second input layer is used to input the historical water pump operation data, historical water pump energy consumption data and historical regional water flow demand data of each water pump after preprocessing; The graph neural network layer constructs each node and edge of the graph neural network. The nodes are water pumps, and the edges are the associations between water pumps. The graph neural network is used to extract the features of the nodes and edges in the water pump network, capturing the mutual influence between water pumps and their operating characteristics. Graph attention mechanism layer, which introduces the attention mechanism and assigns different weights to each node in the graph; The second output layer receives the output of the graph attention mechanism layer and outputs the results of the water pump energy consumption control model, including the speed and flow rate of each water pump in the graph.

[0011] Preferably, in a possible implementation manner of the first aspect, step S500 specifically includes: Input the real-time water pump operation data into the water pump performance prediction model to predict the water pump flow, water pump head, water pump power and water pump efficiency within a preset time in the future; Input the real-time water flow demand data into the water pump energy consumption control model, and output the speed and flow of each water pump in the area that meets the regional water flow demand; Formulate a pump operation strategy based on the existing speed and flow of each pump in the area; According to the water pump operation strategy, the speed and flow rate of each water pump are adjusted in real time to achieve energy-saving operation of the water pump.

[0012] Preferably, in a possible implementation of the first aspect, the method also includes continuously monitoring the operating status and energy consumption of the water pump, and feeding back the real-time monitoring data to the water pump performance prediction model and the water pump energy consumption control model to optimize the model prediction accuracy and control effect.

[0013] The beneficial effects of the present invention are as follows: the water pump performance prediction model constructed by the deep learning algorithm based on PatchTST can accurately capture the complex changes in water pump operation data, predict future water pump performance, and provide a scientific basis for energy-saving regulation. Combined with the energy consumption control model of the graph neural network, the mutual influence and operating characteristics between water pumps are fully considered to achieve dynamic optimization of water pump operation strategies and significantly improve energy-saving effects. Finally, continuous data feedback and model optimization mechanisms ensure the continuous improvement of model prediction accuracy and regulation effects, and achieve long-term efficient energy saving of water pump operation. In summary, the present invention effectively promotes the green and sustainable development of urban water supply systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 A flow chart of a method for energy-saving operation of a water pump based on artificial intelligence is provided for this application. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] Embodiment 1: Figure 1 As shown, the present invention provides an energy-saving operation method of a water pump based on artificial intelligence, the method comprising: Step S100: Acquire the water pump operation data in the area in real time.

[0018] Specifically, the water pump operation data includes the water pump position coordinates, water pump flow, water pump pressure, water pump speed, water pump motor current, water pump motor voltage and water pump ambient temperature.

[0019] In this embodiment, the water pump position coordinates are obtained by the GPS positioning system to obtain the geographical coordinates of the water pump, which are used to determine the geographical location of the water pump. The water pump flow rate uses an electromagnetic flowmeter to measure the water flow rate of the water pump, reflecting the water supply capacity of the water pump. The water pump pressure uses a pressure sensor installed on the water outlet pipe to measure the pressure of the water pump outlet, reflecting the pressure of the water pump outlet.

[0020] The water pump speed obtains the water pump speed information through the speed sensor installed on the motor shaft, reflecting the running speed of the water pump. The water pump motor current and water pump motor voltage use the current sensor and voltage sensor connected to the power line of the motor to measure the current value and voltage value of the motor when it is working, reflecting the running status and energy consumption of the motor. The water pump ambient temperature uses the temperature sensor installed on the water pump housing to measure the temperature of the environment around the water pump, reflecting the temperature of the environment around the water pump.

[0021] Step S200: pre-processing the water pump operation data.

[0022] Specifically, data preprocessing includes data cleaning, data denoising, and data normalization. Data cleaning refers to checking and processing errors, duplications, missing values, and other issues in the data set to ensure the accuracy and completeness of the data. When obtaining water pump operation data, data anomalies may occur due to sensor failures, data transmission errors, and other reasons.

[0023] In this embodiment, first, the water pump operation data in the area is obtained in real time and stored in a DataFrame named "df". The DataFrame contains multiple fields, including "water pump location coordinates", "water pump flow", "water pump pressure", etc.

[0024] In the data cleaning step, we take the processing of missing values ​​and invalid values ​​in the "Water Pump Flow" field as an example. Specifically, the pd.to_numeric function is used to convert the values ​​in the "Water Pump Flow" field to numeric types. For invalid values ​​that cannot be converted to numeric values, the pd.to_numeric function will convert them to NaN. Then, the fillna function is used to replace the NaN values ​​with the mean of all non-NaN values ​​in the "Water Pump Flow" field to fill the missing values.

[0025] Next, we process the rows that contain invalid strings in the "Pump Flow" field. We first convert the values ​​in the "Pump Flow" field to string type, and then use the str.contains function to check whether these strings contain invalid strings.

[0026] For missing values ​​and invalid values ​​in other fields, they are handled similarly. First, invalid values ​​are converted to NaN using the pd.to_numeric function, and then missing values ​​are filled with the mean or the value with the highest number of occurrences as needed.

[0027] Data denoising refers to removing noise from data, that is, data fluctuations caused by random errors or measurement errors. In the pump operation data, noise may come from the accuracy limitations of the sensor or environmental interference.

[0028] In this embodiment, taking the processing of noise in "water pump flow" as an example, a numpy array named flow_data is first defined, which contains the measured value of the water pump flow. Next, a function named moving_average is implemented, which is used to perform moving average filtering on the input data. The function accepts two parameters: data and window_size. Among them, data is the data array to be processed, and window_size is the window size of the moving average filter, which is 3.

[0029] Inside the moving_average function, the np.cumsum function is first used to calculate the cumulative sum of the input data. Then the average value of the data in each window is calculated through slicing. Specifically, the cumulative sum of the previous window is subtracted from the cumulative sum array to get the sum of the data in the current window, and then it is divided by the window size to get the average value of the data in the current window. Finally, the array of average values ​​from the second window to the last window is returned.

[0030] Finally, the moving_average function is used to perform moving average filtering on the flow_data array. By calling this function and passing in the flow_data array and window size parameters, a denoised water pump flow data array cleaned_flow_data is obtained.

[0031] Data normalization refers to scaling the data so that it falls into a small specific interval (such as between 0 and 1). Normalization helps speed up model training and improve model performance. In the pump operation data, the value ranges of different features may vary greatly, and normalization can eliminate this dimensional difference.

[0032] In this embodiment, taking the processing of "water pump flow" as an example, a dictionary named data is first defined, which contains the flow characteristic values ​​of the water pump. Next, the data dictionary is converted into a pandas DataFrame object df. Then the MinMaxScaler class in the sklearn library is introduced, and a MinMaxScaler object scaler is created. MinMaxScaler is a tool for data preprocessing provided by the sklearn library, which can scale the characteristic value to between a given minimum and maximum value.

[0033] After preparing the data and processing tools, call the fit_transform method of the scaler object to normalize the data in df. The fit_transform method first calculates the maximum and minimum values ​​in the data, and then scales the data to between 0 and 1 based on these values.

[0034] Step S300: Using a deep learning algorithm based on PatchTST, combined with pre-processed historical water pump operation data and historical water pump performance data, to build a water pump performance prediction model.

[0035] Specifically, the historical water pump operation data includes water pump position coordinates, historical water pump flow, historical water pump pressure, historical water pump speed, historical water pump motor current, historical water pump motor voltage, historical water pump ambient temperature and timestamp.

[0036] The structure of the pump performance prediction model includes: The first input layer is used to input the preprocessed historical water pump operation data and historical water pump performance data. The historical water pump performance data includes historical water pump flow, historical water pump head, historical water pump power and historical water pump efficiency. The PatchTST feature extraction layer is used to divide the input data into multiple time series subsequences and extract features for each subsequence to capture the local and global features of the time series data. The self-attention mechanism layer assigns dynamic weights to each subsequence through the self-attention mechanism to obtain the long-term dependency of the time series data. The first output layer receives the outputs of the PatchTST feature extraction layer and the self-attention mechanism layer, and outputs the water pump flow prediction, water pump head prediction, water pump power prediction and water pump efficiency prediction within the preset time in the future.

[0037] Pump flow rate refers to the volume of liquid delivered by a pump per unit time, measured in cubic meters per hour (m³ / h). The flow rate depends on factors such as the pump design, speed, and inlet and outlet resistance. In the pump performance prediction model, historical pump flow rate data is used as input to learn the flow rate change pattern over time and thus predict future flow rates.

[0038] Pump head refers to the gravitational potential energy that a pump needs to overcome to transport liquid from a low place to a high place, measured in meters (m). The size of the head is related to factors such as the structure and speed of the pump, as well as the density and viscosity of the liquid. In the pump performance prediction model, historical pump head data is used as input to capture the change trend of the head.

[0039] Pump power refers to the electrical or mechanical energy consumed by a pump during operation, measured in kilowatts (kW). The power depends on factors such as the flow rate, head, and efficiency of the pump. In the pump performance prediction model, historical pump power data is used to learn the relationship between power and parameters such as flow rate and head, so as to predict future power changes.

[0040] Pump efficiency refers to the proportion of input power converted into useful work during the actual working process of the pump, expressed as a percentage (%). The efficiency reflects the performance of the pump and the efficiency of energy utilization. In the pump performance prediction model, historical pump efficiency data is used as input to learn the relationship between efficiency and parameters such as flow rate and head.

[0041] In this embodiment, the first input layer includes a linear layer, the input dimension of the linear layer is input_dim, and the output dimension is 128. Wherein, input_dim represents the number of features of the input data, and the input data dimension is 12, including the pump number, the pump position coordinates, the historical pump flow, the historical pump pressure, the historical pump speed, the historical pump motor current, the historical pump motor voltage, the historical pump ambient temperature, the timestamp, the historical pump head, the historical pump power, and the historical pump efficiency. When the input data enters the first input layer, it is first dimensionally converted through the linear layer, and then nonlinearly transformed through the ReLU activation function to obtain the converted feature data.

[0042] The PatchTST feature extraction layer is used to split the input data into multiple time series subsequences and perform feature extraction on each subsequence. The PatchTST feature extraction layer includes multiple linear layers, each of which corresponds to the feature extraction of a time series subsequence. Specifically, the input data is first reshaped into the shape of (batch_size, sequence_length, feature_dim), where batch_size represents the batch size, sequence_length represents the length of the time series, and feature_dim represents the feature dimension. Then, the time series is evenly divided into num_patches subsequences, each of which is of size patch_size. For each subsequence, the corresponding linear layer is used for feature extraction to obtain a feature vector with an embedding dimension of embedding_dim. Finally, the feature vectors of all subsequences are concatenated along the feature dimension to obtain the final feature representation.

[0043] The self-attention mechanism layer is used to perform self-attention calculations on the feature representations output by the PatchTST feature extraction layer to capture the dependencies between different features. The self-attention mechanism layer includes multiple self-attention heads, each of which includes three linear layers: query, key, and value, and an output linear layer. Specifically, the feature representations are first input into the query, key, and value linear layers to obtain the query matrix Q, key matrix K, and value matrix V. Then, the dot product of the query matrix Q and the key matrix K is calculated and multiplied by the scaling factor scale to obtain the attention score. Next, the attention score is normalized using the softmax function to obtain the attention weight. Finally, the attention weight is multiplied by the value matrix V to obtain the self-attention output. The outputs of all self-attention heads are concatenated along the feature dimension and linearly transformed through the output linear layer to obtain the final self-attention feature representation.

[0044] The first output layer is used to convert the feature representation output by the self-attention mechanism layer into a prediction result. In this embodiment, the first output layer includes a linear layer whose input dimension is embedding_dim × num_patches and whose output dimension is output_dim. After the self-attention feature representation enters the first output layer, it is linearly transformed through the linear layer to obtain the final prediction result.

[0045] Specifically, in this embodiment, the input data dimension is 12, the time series length is 20, and it is divided into 4 patches, each patch size is 5, the embedding dimension is 32, and the output dimension is 4 (predicting the water pump flow, head, power and efficiency). First, a water pump performance prediction model is constructed based on these parameters. Then, input data of the shape of (batch_size, input_dim ×sequence_length) is generated, where batch_size is 2. Finally, the input data is input into the model to obtain the prediction result. The shape of the prediction result is (batch_size, output_dim), that is, each batch contains 2 prediction results, and each prediction result includes the prediction values ​​of 4 dimensions, predicting the water pump flow, water pump head, water pump power and water pump efficiency in the next 2 hours.

[0046] Step S400: Using a deep learning algorithm based on a graph neural network, combined with pre-processed historical water pump operation data, historical water pump energy consumption data, and historical regional water flow demand data, a water pump energy consumption control model is constructed.

[0047] Specifically, the structure of the water pump energy consumption control model includes: The second input layer is used to input the historical pump operation data, historical pump energy consumption data, and historical regional water flow demand data of each pump after preprocessing. The graph neural network layer constructs each node and edge of the graph neural network. The node is the pump and the edge is the association between the pumps. The graph neural network is used to extract the features of the nodes and edges in the pump network to capture the mutual influence between the pumps and their operating characteristics. The graph attention mechanism layer introduces the attention mechanism to assign different weights to each node in the graph. The second output layer receives the output of the graph attention mechanism layer and outputs the results of the pump energy consumption control model, including the speed and flow of each pump in the graph.

[0048] In this embodiment, the first is the second input layer, which is responsible for inputting the preprocessed historical water pump operation data, historical water pump energy consumption data, and historical regional water flow demand data into the model. This is achieved by defining a class called "SecondInputLayer", which inherits from PyTorch's "nn.Module". In the class initialization method, a linear layer is defined to convert the input dimension to 128. In the forward propagation method, the input data is processed by a linear layer and a ReLU activation function.

[0049] The graph neural network layer is used to construct each node and edge of the graph neural network, where the nodes are water pumps and the edges are the associations between water pumps. This is achieved by defining a class called "GraphNeuralNetworkLayer", which also inherits from "nn.Module". In the class initialization method, a GCN convolution layer is defined. In the forward propagation method, the input data and edge index are processed by the GCN convolution layer and the ReLU activation function.

[0050] The graph attention mechanism layer is used to assign different weights to each node in the graph. This layer is implemented by defining a class called "GraphAttentionNetworkLayer". In the class initialization method, a GAT convolution layer is defined and multiple parameters are set, including the number of input channels, the number of output channels, the number of attention heads, whether to connect the output, the dropout rate, the negative slope, and the alpha value. In the forward propagation method, the input data and edge index are processed by the GAT convolution layer and the ReLU activation function.

[0051] The second output layer is responsible for outputting the results of the water pump energy consumption control model, including the speed and flow rate of each water pump in the figure. This is achieved by defining a class called "SecondOutputLayer". In the class initialization method, a linear layer is defined to convert the input dimension into the output dimension (i.e. the number of speeds and flows). In the forward propagation method, the input data is processed by the linear layer.

[0052] The above layers are integrated into a complete model to build a class called "PumpEnergyConsumptionModel". In the class initialization method, the second input layer, graph neural network layer, graph attention mechanism layer and second output layer are defined in sequence. In the forward propagation method, the input data and edge index are processed in sequence through each layer to finally get the output result.

[0053] Step S500: dynamically adjusting the water pump operation strategy of the area according to the output result of the water pump energy consumption control model.

[0054] Specifically, the real-time pump operation data is input into the pump performance prediction model to predict the pump flow, pump head, pump power and pump efficiency within 2 hours. The real-time water flow demand data is input into the pump energy consumption control model to output the speed and flow of each pump in the area that meets the regional water flow demand. Combined with the existing speed and flow of each pump in the area, a pump operation strategy is formulated. According to the pump operation strategy, the speed and flow of each pump are adjusted in real time to achieve energy-saving operation of the pump.

[0055] In this embodiment, the real-time pump operation data is first input into the trained pump performance prediction model. These data include key parameters such as the current flow rate, head, power, efficiency, etc. of each pump, as well as the corresponding timestamp. Based on these data, the model predicts the change trend of the flow rate, head, power and efficiency of each pump in the next 2 hours.

[0056] The real-time water flow demand data is input into the water pump energy consumption control model. These data reflect the water flow demand in each period of the region. The model analyzes these data and combines the correlation between water pumps to output the optimal speed and flow of each water pump in the region that meets the water flow demand.

[0057] When formulating the pump operation strategy, multiple factors are considered comprehensively. First, it is necessary to ensure that the water flow demand in the next 2 hours is met, that is, the water flow in each period must reach the preset target value. Secondly, it is necessary to reduce the overall energy consumption as much as possible and improve the operation efficiency of the pump. To this end, the prediction results of the pump performance prediction model and the output results of the pump energy consumption control model are combined, and the current speed and flow of each pump are compared to formulate the optimal pump operation strategy.

[0058] Specifically, the following steps are followed: Data integration: Integrate the real-time pump operation data and water flow demand data to analyze the current pump operation status and future demand change trends. Forecast and evaluation: Use the pump performance prediction model to predict the flow, head, power and efficiency of the pump in the next 2 hours, and use the pump energy consumption control model to output the optimal speed and flow of each pump that meets the water flow demand. Strategy formulation: Formulate a pump operation strategy based on the prediction results and actual needs.

[0059] The strategy includes the following: Pump speed adjustment: formulate a speed adjustment plan based on the difference between the optimal speed and current speed of each pump. Pump flow distribution: distribute the flow of each pump according to the regional water flow demand to ensure that the demand is met. Energy consumption optimization: minimize overall energy consumption by adjusting the pump speed and flow while ensuring the water flow demand. Strategy implementation and monitoring: implement the formulated pump operation strategy into the actual pump control system, and monitor the operating status of the pump and changes in water flow demand in real time. Strategy adjustment and optimization: during the operation process, adjust and optimize the pump operation strategy based on real-time monitoring data and feedback to ensure the effectiveness and adaptability of the strategy.

[0060] If it is predicted that the water flow demand in a certain area will increase significantly, and the current speed and flow of the water pump cannot meet this demand, the speed and flow of the relevant water pump will be appropriately increased according to the output of the water pump energy consumption control model to ensure that the water flow demand is met. At the same time, if it is predicted that the efficiency of a water pump will decrease in the future, its speed and flow will be adjusted in advance to avoid energy waste.

[0061] In one embodiment, the method also includes continuously monitoring the operating status and energy consumption of the water pump, and feeding back the real-time monitoring data to the water pump performance prediction model and the water pump energy consumption control model to optimize the model prediction accuracy and control effect.

[0062] Specifically, during the monitoring process, the focus is on key performance indicators such as the flow rate, head, power, efficiency, and energy consumption data of the water pump. These data are collected in real time through various sensors installed on the water pump system and transmitted to the central monitoring system for unified processing and analysis.

[0063] On the one hand, real-time monitoring data verifies the accuracy of the pump performance prediction model. By comparing the prediction results with the real-time monitoring data, the performance of the model in predicting the pump flow, head, power and efficiency can be evaluated. If it is found that there is a large deviation between the prediction results and the actual data, the model can be adjusted in parameters or optimized in structure to improve its prediction accuracy.

[0064] On the other hand, real-time monitoring data is also used to optimize the water pump energy consumption control model. By comparing the energy consumption of different water pumps under different operating conditions, water pumps with high energy consumption and their potential problems can be more accurately identified. At the same time, based on the real-time monitoring data feedback, the water pump energy consumption control model can adjust the water pump speed and flow distribution strategy in real time to achieve more refined energy consumption control.

[0065] In addition, continuous monitoring can also help to promptly detect faults or abnormal conditions in the pump system. For example, when the flow rate or head of the pump suddenly drops, it may be a sign of problems such as pump blockage or wear. Through the analysis of real-time monitoring data, these potential problems can be discovered in time and corresponding maintenance measures can be taken to avoid further deterioration of the fault and cause greater losses.

[0066] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A water pump energy-saving operation method based on artificial intelligence, characterized in that: The method comprises: Step S100: acquiring the water pump operation data in the area in real time; Step S200: preprocessing the water pump operation data; Step S300: using a deep learning algorithm based on PatchTST, combined with pre-processed historical water pump operation data and historical water pump performance data, to build a water pump performance prediction model; Step S400: using a deep learning algorithm based on a graph neural network, combining pre-processed historical water pump operation data, historical water pump energy consumption data, and historical regional water flow demand data, to build a water pump energy consumption control model; Step S500: dynamically adjust the water pump operation strategy of the area in combination with the output results of the water pump performance prediction model and the water pump energy consumption control model.

2. The water pump energy-saving operation method according to claim 1, characterized in that: The water pump operation data includes water pump position coordinates, water pump flow, water pump pressure, water pump speed, water pump motor current, water pump motor voltage and water pump ambient temperature.

3. The water pump energy-saving operation method according to claim 1, characterized in that: The data preprocessing includes data cleaning, data denoising and data normalization.

4. The water pump energy-saving operation method according to claim 1, characterized in that: The historical water pump operation data includes water pump position coordinates, historical water pump flow, historical water pump pressure, historical water pump speed, historical water pump motor current, historical water pump motor voltage, historical water pump ambient temperature and timestamp.

5. The water pump energy-saving operation method according to claim 1, characterized in that: The structure of the water pump performance prediction model includes: The first input layer is used to input pre-processed historical water pump operation data and historical water pump performance data, wherein the historical water pump performance data includes historical water pump flow, historical water pump head, historical water pump power and historical water pump efficiency; The PatchTST feature extraction layer is used to split the input data into multiple time series subsequences and perform feature extraction on each subsequence to capture the local and global features of the time series data; The self-attention mechanism layer assigns dynamic weights to each subsequence through the self-attention mechanism to obtain the long-term dependencies of time series data; The first output layer receives the output of the PatchTST feature extraction layer and the self-attention mechanism layer, and outputs the pump flow prediction, pump head prediction, pump power prediction, and pump efficiency prediction within a preset time in the future.

6. The water pump energy-saving operation method according to claim 1, characterized in that: The structure of the water pump energy consumption control model includes: The second input layer is used to input the historical water pump operation data, historical water pump energy consumption data and historical regional water flow demand data of each water pump after preprocessing; The graph neural network layer constructs each node and edge of the graph neural network. The nodes are water pumps, and the edges are the associations between water pumps. The graph neural network is used to extract the features of the nodes and edges in the water pump network, capturing the mutual influence between water pumps and their operating characteristics. Graph attention mechanism layer, which introduces the attention mechanism and assigns different weights to each node in the graph; The second output layer receives the output of the graph attention mechanism layer and outputs the results of the water pump energy consumption control model, including the speed and flow rate of each water pump in the graph.

7. The water pump energy-saving operation method according to claim 1, characterized in that: The step S500 specifically includes: Input the real-time water pump operation data into the water pump performance prediction model to predict the water pump flow, water pump head, water pump power and water pump efficiency within a preset time in the future; Input the real-time water flow demand data into the water pump energy consumption control model, and output the speed and flow of each water pump in the area that meets the regional water flow demand; Formulate a pump operation strategy based on the existing speed and flow of each pump in the area; According to the water pump operation strategy, the speed and flow rate of each water pump are adjusted in real time to achieve energy-saving operation of the water pump.

8. The water pump energy-saving operation method according to claim 1, characterized in that: The method also includes continuously monitoring the operating status and energy consumption of the water pump, and feeding back the real-time monitoring data to the water pump performance prediction model and the water pump energy consumption control model to optimize the model prediction accuracy and control effect.

Images