Pressure prediction model training and prediction method, system, device and storage medium

Abstract

This invention provides a training and prediction method, system, device, and storage medium for a pressure prediction model, belonging to the field of urban intelligent water supply network technology. The training method for a multi-monitoring-point network pressure prediction model includes: acquiring historical pressure data of the monitored urban water supply network; constructing a pressure prediction model, which includes a temporal convolutional network model and a long short-term memory network model; jointly training the convolutional network model and the long short-term memory network model based on historical pressure data to obtain a trained pressure prediction model; and using this trained pressure model to predict the pressure data of each monitoring point on the monitored urban water supply network based on the historical pressure data. This invention provides a training method for a pressure prediction model that, based on the correlation and periodicity of multiple monitoring points on an urban water supply network, efficiently and accurately predicts the pressure data of each monitoring point on the network.

Description

Technical Field

[0001] This invention relates to the field of urban intelligent water supply network technology, specifically to training and prediction methods, systems, equipment, and storage media for pressure prediction models. Background Technology

[0002] In urban water supply networks, the widespread use of IoT and sensor technologies has led to a substantial accumulation of network pressure data. IoT technology provides the necessary foundation and support for data management in urban water supply systems. Today, data and its understanding are becoming increasingly important; analyzing water supply network pressure data can ensure timely and effective water supply safety.

[0003] A city's DMA (District Metering Area) water supply network is a multi-source, multi-node flow system used to deliver water with sufficient pressure and quantity to users. Its characteristics include multiple pressure monitoring points within a DMA area, network connectivity, and non-linear and spatially distributed pressure data. In city water supply networks, most monitoring involves measuring network pressure data. Water supply personnel typically observe the network's operational status based on this measured pressure data, controlling the overall network pressure within a reasonable range and assessing any anomalies based on local pressure loss. Therefore, accurate and rapid prediction of pressure at multiple monitoring points is crucial for real-time scheduling and scientific management.

[0004] Considering the connectivity of water supply networks, pressure prediction for these networks differs from prediction for a single point. Pressure values ​​at monitoring points are influenced by pressure values ​​from other points within the network. Water supply operators need to focus on the changing trends of pressure data from multiple monitoring points within the region. Existing technologies such as differential autoregressive moving average models and recurrent neural networks struggle to perform parallel analysis of pressure data from various monitoring points in water supply networks, hindering rapid and efficient prediction. Temporal Convolutional Networks (TCNs), however, demonstrate superior performance compared to traditional methods and recurrent neural networks in handling long-input time series. The inherent limitations of convolutional neural networks restrict their application in time series prediction, while TCNs incorporate fundamental techniques such as local connections and weight sharing found in traditional convolutional neural networks. Furthermore, TCNs possess a simple network structure, offering significant advantages in capturing long-range temporal patterns.

[0005] Therefore, there is a need to provide a training and prediction method, system, device, and storage medium for a stress prediction model to solve the above problems. Summary of the Invention

[0006] In view of the shortcomings of the prior art, the purpose of this invention is to provide a training and prediction method, system, device and storage medium for a pressure prediction model, so as to improve the technical problem in the prior art that differential autoregressive moving average model and recurrent neural network are difficult to analyze the pressure data of each monitoring point in parallel and cannot quickly and efficiently predict the pressure data of multiple monitoring points in the prediction of pressure data of multiple monitoring points in water supply network.

[0007] To achieve the above and other related objectives, this invention provides a training method for a multi-monitoring-point pipeline network pressure prediction model, comprising the following steps:

[0008] Obtain historical pressure data of the monitored city's water supply network;

[0009] A stress prediction model is constructed, which includes a temporal convolutional network model and a long short-term memory network model.

[0010] The convolutional network model and the long short-term memory network model are jointly trained based on the historical pressure data to obtain the trained pressure prediction model. The pressure prediction model predicts the pressure data of each monitoring point on the monitored urban water supply network based on the historical pressure data.

[0011] In one embodiment of the present invention, the present invention also provides a method for predicting the pressure of a multi-monitoring-point pipeline network. The method employs a multi-monitoring-point pipeline network pressure prediction model trained using the training method described in any of the above embodiments. The method includes:

[0012] Obtain historical pressure data of the monitored city's water supply network;

[0013] The historical pressure data is input into the multi-monitoring point pipeline pressure prediction model to obtain the pressure prediction results of each monitoring point on the monitored urban water supply pipeline.

[0014] In one embodiment of the present invention, the present invention also provides a training system for a multi-monitoring-point pipeline network pressure prediction model, the system comprising:

[0015] The data acquisition unit is used to acquire historical pressure data of the monitored city's water supply network;

[0016] The model building unit is used to build a stress prediction model, which includes a temporal convolutional network model and a long short-term memory network model.

[0017] The joint training unit performs joint training on the convolutional network model and the long short-term memory network model based on the historical pressure data to obtain the trained pressure prediction model. The pressure prediction model predicts the pressure data of each monitoring point on the monitored urban water supply network based on the historical pressure data.

[0018] In one embodiment of the present invention, a computer device is also provided, including a processor coupled to a memory storing program instructions, wherein the processor executes the program instructions stored in the memory to implement the method described in any of the preceding claims.

[0019] In one embodiment of the present invention, a computer-readable storage medium is also provided, including a program that, when run on a computer, causes the computer to perform the method described in any one of the above descriptions.

[0020] This invention provides a training and prediction method, system, device, and storage medium for a pressure prediction model. The model is applied to urban DMA water supply network pressure prediction scenarios with multiple monitoring points. It fully considers the attribute information of the monitoring points in the water supply network, capturing the temporal and periodic features of the monitoring points through a temporal convolutional network model and a self-attention mechanism. Then, it uses pressure data with temporal and periodic features as input to a long short-term memory network to learn the long-term dependencies of the water supply network pressure data, further extracting the temporal features of the monitoring point pressure data, and finally obtaining the final prediction result through a fully connected layer.

[0021] In summary, the training and prediction methods, systems, equipment, and storage media of the pressure prediction model can process historical pressure data from multiple monitoring points on the urban water supply network in parallel using a temporal convolutional network model. Simultaneously, a self-attention mechanism is employed to acquire the periodic characteristics of the monitoring points themselves. Finally, the long-term memory capability of the LSTM (Long Short-Term Memory) network is utilized to obtain the temporal dependence of the monitoring points. Based on the correlation and periodicity of multiple monitoring points on the urban water supply network, this model can efficiently and accurately predict the pressure data of each monitoring point on the network. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is an overall framework diagram of the pressure prediction model in one embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of the residual module in a temporal convolutional network model according to an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram illustrating the process of dilated causal convolution in a temporal convolutional network model according to an embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of the structure of a long short-term memory network model in one embodiment of the present invention;

[0027] Figure 5 This is a flowchart illustrating the training method for a multi-monitoring point pipeline pressure prediction model in one embodiment of the present invention.

[0028] Figure 6 This is a flowchart illustrating step S1 in one embodiment of the present invention;

[0029] Figure 7 This is a flowchart illustrating step S3 in one embodiment of the present invention;

[0030] Figure 8 This is a data comparison chart of the prediction results and actual results of the multi-monitoring point pipeline pressure prediction model in one embodiment of the present invention;

[0031] Figure 9 This is a schematic diagram illustrating the optimal parameter selection of the pressure prediction model under the MAE index in one embodiment of the present invention;

[0032] Figure 10 This is a structural block diagram of a training system for a multi-monitoring-point pipeline pressure prediction model in one embodiment of the present invention.

[0033] Component designation explanation:

[0034] 10. Training system for multi-monitoring point pipeline pressure prediction model; 11. Data acquisition unit; 12. Model calling unit; 13. Joint training unit. Detailed Implementation

[0035] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features can be combined with each other. It should also be understood that the terminology used in the embodiments of the present invention is for describing specific implementation schemes and not for limiting the scope of protection of the present invention. Test methods in the following embodiments that do not specify specific conditions are generally performed under conventional conditions or according to the conditions recommended by the respective manufacturers.

[0036] Please see Figures 1 to 10 It should be understood that the structures, proportions, sizes, etc., depicted in the accompanying drawings are merely for illustrative purposes to aid those skilled in the art and to facilitate understanding. They are not intended to limit the scope of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, without affecting the effectiveness and purpose of the invention, should still fall within the scope of the technical content disclosed herein. Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0037] When numerical ranges are given in the embodiments, it should be understood that, unless otherwise stated in the present invention, both endpoints of each numerical range and any value between the two endpoints may be selected. Unless otherwise defined, all technical and scientific terms used in this invention, as well as the prior art known to those skilled in the art and the description of this invention, may be implemented using any prior art methods, devices, and materials similar to or equivalent to those described, used, or made of materials in the embodiments of this invention.

[0038] Please see Figures 1 to 6 The purpose of this invention is to provide a training and prediction method, system, device and storage medium for a pressure prediction model, so as to improve the technical problem in the prior art that differential autoregressive moving average model and recurrent neural network are difficult to analyze the pressure data of each monitoring point in parallel and cannot quickly and efficiently predict the pressure data of multiple monitoring points in water supply network with multiple monitoring points.

[0039] Please see Figures 1 to 5The pressure prediction model trained based on the training method of this invention simultaneously processes multi-dimensional pressure time-series data by constructing a temporal convolutional network model and employing a self-attention module to enhance the correlation between the components in the output time-series feature vector. At the same time, a long short-term memory network model is used to further extract the time-series features from the time-series feature vector to transform the time-series feature vector into a prediction feature vector. Finally, a fully connected layer is used to integrate the prediction feature vector to obtain the pressure prediction results of multiple monitoring points on the monitored urban water supply network.

[0040] Please see Figures 1 to 5 , Figure 5 This is a flowchart illustrating a training method for a multi-monitoring-point pipeline network pressure prediction model according to an embodiment of the present invention. In one embodiment of the present invention, a training method for a multi-monitoring-point pipeline network pressure prediction model is provided, comprising the following steps:

[0041] Step S1: Obtain historical pressure data of the monitored city's water supply network;

[0042] Specifically, such as Figure 6 As shown, step S1 includes the following process:

[0043] S11. Obtain historical pressure data of the monitored urban water supply network, and divide the historical pressure data into a training set and a test set according to time sequence; wherein, the historical pressure data includes historical pressure data of multiple monitoring points on the urban water supply network;

[0044] For example, the historical pressure data includes pressure data collected from 12 pressure monitoring points in the water supply network within the DMA area from July 1, 2020 to July 15, 2020. The data sampling frequency is once every five minutes, and each monitoring point has a total of 4320 pressure data points. In step S11, the historical pressure data collected in the first 12 days is divided into a training set for training the pressure prediction model; the historical pressure data collected in the remaining 3 days is divided into a test set for verifying the accuracy of the pressure prediction model's prediction results.

[0045] S12. Based on multiple monitoring points on the monitored urban water supply network, the historical pressure data corresponding to each monitoring point in the training set is divided into multiple time-series components, which are used to represent pressure data in different time periods. Then, the multiple time-series components are summarized in chronological order for different monitoring points to obtain pressure time-series data, and the pressure time-series data is preprocessed to facilitate input into the pressure prediction model.

[0046] The pressure time-series data is a multi-dimensional matrix constructed based on information from multiple monitoring points in the spatial dimension and time-series components arranged chronologically in the temporal dimension. This pressure time-series data is represented as follows: In the formula mRepresents the serial number of the monitoring point. n This represents the time series length (i.e., the number of time series components) corresponding to the monitoring point.

[0047] In this invention, the preprocessing of pressure time series data includes: firstly, normalizing the pressure time series data to normalize each time series component within the (0,1) interval, in order to accelerate the training speed of the model and obtain better prediction results; then, according to the length of the input time series of the temporal convolutional network model... L Convert pressure time series data into a three-dimensional matrix vector, for example, convert the first... m The pressure time-series data of each monitoring point are represented as follows: Pressure time series data (Converted to a 3D matrix vector) The input stress prediction model is a temporal convolutional network model. After feature extraction by the temporal convolutional network model and the long short-term memory network model, a temporal length of [value missing] is obtained. K Output data The prediction result is represented as .

[0048] Step S2: Construct a stress prediction model, which includes a Temporal Convolutional Network (TCN) model and a Long Short-Term Memory (LSTM) model.

[0049] The inventors discovered that urban water supply networks are multi-node monitoring systems, with interconnected monitoring points. Traditional methods for predicting water supply network pressure are mostly based on historical data from a single node, which limits the analysis of the network's operational status. Considering the connectivity of the water supply network, water supply operators focus on the changing trends of pressure data from multiple monitoring points within the region. Therefore, to address the shortcomings of traditional methods in predicting pressure data from multiple monitoring points in water supply networks, temporal convolutional networks can effectively solve this problem. Temporal convolutional networks are suitable for sequential data with temporal characteristics and large receptive fields, and are suitable for parallel and distributed computing, enabling them to effectively extract data features.

[0050] Please see Figures 1 to 3The temporal convolutional network model is used to extract temporal features representing the pressure correlation among multiple monitoring points in the pressure time series data and summarize them into a temporal feature vector. The temporal convolutional network model includes at least two residual modules, and based on a self-attention mechanism, a self-attention module is set at the output of each residual module in the model. The residual modules are used to extract the temporal feature vector from the pressure time series data, and the self-attention modules are used to capture the periodic features of the temporal feature vector itself, and to weight the temporal feature vector based on the periodic features to obtain a temporal feature vector with periodic features, thereby enhancing the correlation between the temporal components in the temporal feature vector.

[0051] like Figure 2 and Figure 3 As shown, the temporal convolutional network model uses residual blocks to avoid the gradient explosion and vanishing problems of deep networks. The residual block has two causal convolutional layers, and the output of each causal convolution is sequentially configured with weight normalization, a Rectified Linear Unit (ReLU) function, and a Dropout output layer. Furthermore, the residual block sets an additional residual connection between the input and output of the two causal convolutional layers, allowing the input data to be directly added to the output data, thereby avoiding performance saturation of the deep network and ensuring its stability and performance.

[0052] Specifically, the residual module performs convolution processing on the input stress time-series data using causal convolution, utilizing convolution kernels. Time series of a monitoring point in the input data t Time sequence t Previous timing length L The temporal components are convolved to obtain the temporal data corresponding to the monitoring point. t - L To time sequence t Temporal characteristics of stage pressure time series data By summarizing the time-series features, the time-series feature vector of the pressure time-series data can be obtained. Among them, As a convolution operator, when the temporal length of the input data is less than the designed convolution kernel size, zero-padding is usually applied to the left side of the input temporal sequence to achieve causal convolution.

[0053] And, as Figure 3 As shown, in this invention, causal convolution expands the receptive field of the temporal convolutional network model by introducing dilated convolution and performing interval sampling on the input of the convolution. The dilated causal convolution can review longer historical samples. The sampling rate of the causal convolution is determined by the coefficients of the dilated convolution. d Determines the expansion factor of hidden layers in causal convolution.d As network depth increases exponentially, deep networks with causal convolutions can converge quickly, but their performance saturates. The causal convolution operation, after introducing dilated convolution, is defined as follows: In the formula, d For the coefficients of the dilated convolution, as a special case, when d When = 1, the dilated causal convolution degenerates into a normal causal convolution.

[0054] like Figure 1 and Figure 2 As shown, in one embodiment of the present invention, the temporal convolutional network model employs two residual modules to extract features from the stress time series data. In the first residual module, the causal convolution kernel size is 5, the dilation coefficient is 1, and the number of filters is 32. In the second residual module, the causal convolution dilation coefficient is 2, and other parameters remain the same as the first residual module. A 1×1 convolution is set between the input and output of each residual module to ensure that the input and output data of each residual module are of the same size.

[0055] As seen above, temporal convolutional neural networks (TCNNs), as an alternative model for sequence modeling, have a greater advantage than recurrent neural networks (RNNs) in capturing long-range temporal patterns. The backpropagation of TCNNs from the output layer to the input layer is the same as other feedforward networks, thus enabling them to handle massive amounts of highly nonlinear sequence data. While deep, multi-layered convolutional neural networks experience performance degradation in feature extraction, TCNNs, using dilated convolutions, can capture more historical input data within the same network depth, resulting in a more flexible receptive field when processing data in parallel. TCNNs are suitable for parallel and distributed computing, and are well-suited for sequence data with temporal characteristics and large receptive fields. They incorporate not only dilated causal convolutions and residual connections, but also the advantages of local connections and weight sharing found in convolutional neural networks, giving them better feature extraction and historical memory capabilities when processing sequence data with multi-dimensional features, leading to better prediction results.

[0056] In addition, such as Figure 1 As shown, in one embodiment of the present invention, in order to provide greater weight to the hidden states that are closely related to the prediction, the temporal convolutional network model sets a self-attention module at the output of each residual module. The self-attention module can extract periodic features based on the temporal feature vector output by the residual module, and process the temporal feature vector with weights based on the periodic features to obtain a temporal feature vector with periodic features.

[0057] Specifically, the self-attention module obtains the temporal feature vector output by the residual module, and initializes it to obtain three weight matrices. , and The query vector is obtained by multiplying each dimension of the time-series feature vector by the three weight matrices. key vector Sum value vector By query vector With key vector Perform a dot product, then use softmax to calculate the attention score, and finally combine the attention score with the value vector. The weighted value is obtained by multiplication to get the output. The formula for the self-attention mechanism is as follows:

[0058]

[0059] in, , and These are the query vector, key vector, and value vector matrix. The temporal feature vectors are weighted appropriately using a self-attention mechanism to highlight more critical influencing factors and achieve high-level feature learning. Through the weighted temporal feature vectors, each temporal component can fully utilize the periodic features carried by the input sequence. Therefore, a self-attention mechanism is used to extract the feature information of pressure from multiple monitoring points, with each different monitoring point treated as a separate time series data point.

[0060] like Figure 1 and Figure 4 As shown, the Long Short-Term Memory (LSTM) network model is used to sort out the temporal feature vectors into predicted feature vectors, and then integrate the predicted feature vectors through a fully connected layer to obtain the stress prediction result and output the stress prediction result.

[0061] The Long Short-Term Memory (LSTM) network, by introducing storage units and gating mechanisms, can perform better with long sequences. For example... Figure 4 As shown, LSTM selectively retains key information from the temporal feature vector of the input by introducing an input gate, a forget gate, and an output gate. The forget gate determines how much state information from the previous time step is retained; the input gate determines how much state information from the current time step is retained; and the output gate determines the output quantity of the current unit state. LSTM uses three gating mechanisms to solve the problems of recurrent neural networks, enabling it to perform well in learning long-term features from sequence data and to capture long-term dependencies in sequences. The output of different stress points in the time dimension is obtained through LSTM. The calculation formula of LSTM is as follows:

[0062]

[0063]

[0064]

[0065]

[0066]

[0067] Where σ and tanh represent the sigmoid and tanh operations, respectively; , , These represent the input gate, forget gate, and output gating mechanisms, respectively. Indicates input data, and These represent cell state information. and This represents the output of the LSTM unit at the previous and current time steps. W and b These represent the weights and bias vectors of the gating mechanism, respectively. Represents the Hadamard product.

[0068] like Figure 1 As shown, the fully connected layer (MLP), serving as the output of the pressure prediction model, excels at learning the nonlinear mapping relationship between input and output data. It is used to integrate prediction feature vectors to obtain pressure prediction results from multiple monitoring points on the monitored urban water supply network. The fully connected layer comprises an input layer, a hidden layer, and an output layer. The hidden layer processes the data from the input layer; each neuron performs a weighted sum of the input data based on its weights and bias vectors, and then passes the sum through a nonlinear activation function to obtain the output value. The output layer processes the data from the hidden layer to generate the output.

[0069] Step S3: Based on the historical pressure data, the convolutional network model and the long short-term memory network model are jointly trained to obtain the trained pressure prediction model. The pressure prediction model predicts the pressure data of each monitoring point on the monitored urban water supply network based on the historical pressure data.

[0070] Specifically, such as Figure 7 As shown, step S3 includes the following process:

[0071] S31. Input the training set into the temporal convolutional network model.

[0072] S32. Obtain the temporal feature vector of the pressure time series data through the temporal convolutional network model; specifically, obtain the temporal feature vector of the pressure time series data through the residual module in the temporal convolutional network model, and then perform weighted processing on the temporal feature vector based on the periodicity of the temporal feature vector itself through the self-attention module to obtain a temporal feature vector with periodic characteristics.

[0073] S33. The time-series feature vectors with periodic characteristics are sorted into predictive feature vectors through the Long Short-Term Memory Network model. In particular, the Long Short-Term Memory Network selectively retains the key information of the input time-series feature vectors by introducing storage units and gating mechanisms, thereby sorting the time-series feature vectors into predictive feature vectors.

[0074] S34. By integrating the predicted feature vectors through a fully connected layer, the pressure prediction results of multiple monitoring points on the monitored urban water supply network are obtained.

[0075] S35. Compare the stress prediction results with the test set to obtain the loss function of the stress prediction model. Minimize the loss function through iterative training to obtain the trained stress prediction model. In one embodiment of the invention, the Adam optimizer is used to optimize the loss function of the stress prediction model. The temporal convolutional network model and the long short-term memory network model are trained using supervised learning with backpropagation through fully connected layers. The loss function of the stress prediction model is minimized during iterative training to obtain the trained stress prediction model.

[0076] The loss function includes at least one of mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE); the mean absolute error is... The root mean square error is, The mean absolute percentage error is, ,in, This represents the true value of the pressure data. Indicates the first i The pressure data at the monitoring point is the first j Predicted values ​​at each time point m and n These represent the pressure quantity and sample quantity at the monitoring points, respectively.

[0077] Furthermore, the temporal length of the input data, the number of LSTM neurons, the size of the convolutional kernel, and the number of filters in the stress prediction model of this invention have a significant impact on the model's prediction performance and learning efficiency during training. For example... Figure 9 As shown, in one embodiment of the present invention, under the MAE evaluation index, when the input time sequence length of the stress prediction model is 30, the number of LSTM neurons is 100, the convolution kernel size is 5 and the number of filters is 32, the prediction error of the stress prediction model reaches the minimum value.

[0078] To verify the improvement in pressure prediction accuracy of the pressure prediction model (TSA-LSTM) in this invention compared to existing models, experiments were conducted to compare the prediction results of the TSA-LSTM model with those of several existing models. As shown in Table 1, the detailed description is as follows: the TSA-LSTM model significantly improves the prediction results, indicating that it achieves excellent performance in predicting pressure data. For example, for 5-minute predictions, compared to the SVR model, the MAE and RMSE of the TSA-LSTM model are reduced by approximately 32.95% and 12.09%, respectively. Compared to the XGBoost model, the MAE and RMSE are reduced by approximately 13.37% and 4.48%, respectively. This is mainly because methods such as SVR and XGBoost face certain challenges in feature extraction from highly nonlinear pipeline pressure data. Compared with the CNN-GRU model, MAE and RMSE were reduced by approximately 5.41% and 3.90%, respectively; compared with the TCN model, MAE and RMSE were reduced by approximately 4.37% and 4.19%, respectively. Although the CNN-GRU and TCN models have certain advantages in long-term memory, they have certain limitations in the simultaneous prediction of stress data from multiple monitoring points in parallel.

[0079] Table 1. Prediction results of different models

[0080]

[0081] Furthermore, to verify the effectiveness of the stress prediction model (TSA-LSTM) in this invention, which combines a Temporal Convolutional Network (TCN), a self-attention mechanism, and a Long Short-Term Memory (LSTM) network, the TSA-LSTM model was compared with both the TCN-SA and TCN-LSTM models. As shown in Table 1, the TSA-LSTM model clearly demonstrates better prediction results across both MAE and RMSE evaluation metrics. Taking the 5-minute prediction results as an example, compared to the TCN-SA model, the TSA-LSTM model reduced MAE and RMSE by approximately 2.78% and 3.03%, respectively; compared to the TCN-LSTM model, the MAE and RMSE were reduced by approximately 1.69% and 2.44%, respectively. Figure 8 This is a visualization of the 5-minute pressure prediction results of the TSA-LSTM model for a certain monitoring point. It can be seen that the predicted value is close to the actual value. The TSA-LSTM model can utilize the good feature extraction capability of TCN. TCN is suitable for parallel and distributed computing because it uses the basic weight sharing characteristic. The self-attention mechanism can focus on the periodic features of the data. Finally, LSTM can achieve long-term prediction results for sequence data.

[0082] Please see Figures 1 to 5The present invention also provides a method for predicting pipeline pressure at multiple monitoring points. The method uses the training method described above to train the multi-monitoring-point pipeline pressure prediction model, and includes:

[0083] Obtain historical pressure data of the monitored city's water supply network;

[0084] The historical pressure data is input into the multi-monitoring point pipeline pressure prediction model to obtain the pressure prediction results of each monitoring point on the monitored urban water supply pipeline.

[0085] Please see Figure 10 , Figure 10 The diagram shows a structural block diagram of a training system for a multi-monitoring-point pipeline network pressure prediction model according to an embodiment of the present invention. The training system 10 includes a data acquisition unit 11, a model building unit 12, and a joint training unit 13. The data acquisition unit 11 acquires historical pressure data of the monitored urban water supply network; the model building unit 12 constructs a pressure prediction model, which includes a temporal convolutional network model and a long short-term memory network model; the joint training unit 13 jointly trains the convolutional network model and the long short-term memory network model based on the historical pressure data to obtain the trained pressure prediction model, which predicts the pressure data of each monitoring point on the monitored urban water supply network based on the historical pressure data.

[0086] It should be noted that, in order to highlight the innovative aspects of this invention, this embodiment does not include modules that are not closely related to solving the technical problems proposed by this invention, but this does not mean that there are no other modules in this embodiment.

[0087] Furthermore, those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the embodiments provided by this invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for example, the division of modules is merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0088] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0089] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional units.

[0090] This embodiment also proposes a computer device, which includes a processor and a memory, coupled together. The memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the aforementioned task management method is implemented. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The memory may include Random Access Memory (RAM) and may also include Non-Volatile Memory, such as at least one disk storage device. The memory can be an internal memory of the Random Access Memory (RAM) type. The processor and memory can be integrated into one or more independent circuits or hardware, such as an Application Specific Integrated Circuit (ASIC). It should be noted that when the computer program in the aforementioned memory is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention.

[0091] This embodiment also proposes a computer-readable storage medium storing computer instructions for instructing a computer to execute the task management method described above. The storage medium can be an electronic medium, magnetic medium, optical medium, electromagnetic medium, infrared medium, or a semiconductor system or propagation medium. The storage medium may also include semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disc. Optical discs may include optical disc-read-only memory (CD-ROM), optical disc-read / write (CD-RW), and DVD.

[0092] This invention provides a training and prediction method, system, device, and storage medium for a pressure prediction model. The model is applied to urban DMA water supply network pressure prediction scenarios with multiple monitoring points. It fully considers the attribute information of the monitoring points in the water supply network, capturing the temporal and periodic features of the monitoring points through a temporal convolutional network model and a self-attention mechanism. Then, it uses pressure data with temporal and periodic features as input to a long short-term memory network to learn the long-term dependencies of the water supply network pressure data, further extracting the temporal features of the monitoring point pressure data, and finally obtaining the final prediction result through a fully connected layer.

[0093] In summary, the training and prediction methods, systems, equipment, and storage media of the pressure prediction model can process historical pressure data from multiple monitoring points on the urban water supply network in parallel using a temporal convolutional network model. Simultaneously, a self-attention mechanism is employed to acquire the periodic characteristics of the monitoring points themselves. Finally, the long-term memory capability of the LSTM network is used to obtain the temporal dependence of the monitoring points. Based on the correlation and periodicity of multiple monitoring points on the urban water supply network, this model can efficiently and accurately predict the pressure data of each monitoring point on the network.

[0094] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for training a multi-monitoring-point pipe network pressure prediction model, characterized in that, include: Obtain historical pressure data of the monitored city's water supply network; A stress prediction model is constructed, which includes a temporal convolutional network model and a long short-term memory network model. The convolutional network model and the long short-term memory network model are jointly trained based on the historical pressure data to obtain the trained pressure prediction model. The pressure prediction model predicts the pressure data of each monitoring point on the monitored urban water supply network based on the historical pressure data. The step of acquiring historical pressure data of the monitored urban water supply network includes: acquiring historical pressure data of the monitored urban water supply network, dividing the historical pressure data into a training set and a test set according to time sequence; the historical pressure data includes historical pressure data of multiple monitoring points on the urban water supply network; dividing the historical pressure data of each monitoring point in the training set into multiple time-series components according to time sequence, summarizing the multiple time-series components for different monitoring points to obtain pressure time-series data, and preprocessing the pressure time-series data; The temporal convolutional network model includes at least two residual modules. The model incorporates a self-attention module at the output of each residual module based on a self-attention mechanism. The residual modules extract temporal feature vectors from the stress time-series data, while the self-attention module captures the periodic features of the temporal feature vectors themselves and performs weighted processing based on these periodic features to obtain a temporal feature vector with periodic characteristics.

2. The method of claim 1, wherein the method further comprises: The step of jointly training the convolutional network model and the long short-term memory network model based on the historical stress data to obtain the trained stress prediction model includes: The training set is input into the temporal convolutional network model; The temporal feature vector of the pressure time series data is obtained through the temporal convolutional network model. The temporal feature vectors are sorted into predicted feature vectors using a long short-term memory network model; By integrating the predicted feature vectors through a fully connected layer, the pressure prediction results of multiple monitoring points on the monitored urban water supply network are obtained. By comparing the stress prediction results with the test set, the loss function of the stress prediction model is obtained. The loss function is minimized through iterative training to obtain the trained stress prediction model.

3. The method of claim 2, wherein the method further comprises: The loss function includes at least one of mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE); the mean absolute error is... The root mean square error is, The mean absolute percentage error is, ,in, This represents the true value of the pressure data. Indicates the first i The pressure data at the monitoring point is the first j Predicted values ​​at each time point m and n These represent the pressure quantity and sample quantity at the monitoring points, respectively.

4. The method of claim 2, wherein the method further comprises: The fully connected layer trains the temporal convolutional network model and the long short-term memory network model through supervised learning via backpropagation based on the loss function.

5. A multi-monitoring-point pipe network pressure prediction method, characterized by, The multi-monitoring point pipeline pressure prediction model is trained using the training method of the multi-monitoring point pipeline pressure prediction model according to any one of claims 1 to 4, wherein the multi-monitoring point pipeline pressure prediction method includes: Obtain historical pressure data of the monitored city's water supply network; The historical pressure data is input into the multi-monitoring point pipeline pressure prediction model to obtain the pressure prediction results of each monitoring point on the monitored urban water supply pipeline.

6. A training system of a multi-monitoring-point pipe network pressure prediction model, characterized in that, include: The data acquisition unit is used to acquire historical pressure data of the monitored city's water supply network; The model building unit is used to build a stress prediction model, which includes a temporal convolutional network model and a long short-term memory network model. The joint training unit performs joint training on the convolutional network model and the long short-term memory network model based on the historical pressure data to obtain the trained pressure prediction model. The pressure prediction model predicts the pressure data of each monitoring point on the monitored urban water supply pipeline based on the historical pressure data. The data acquisition unit is used to obtain historical pressure data of the monitored urban water supply network, and divide the historical pressure data into a training set and a test set according to time sequence; the historical pressure data includes historical pressure data of multiple monitoring points on the urban water supply network; the historical pressure data of each monitoring point in the training set is divided into multiple time series components according to time sequence, and the multiple time series components are summarized for different monitoring points to obtain pressure time series data, and the pressure time series data is preprocessed. The temporal convolutional network model includes at least two residual modules. The model incorporates a self-attention module at the output of each residual module based on a self-attention mechanism. The residual modules extract temporal feature vectors from the stress time-series data, while the self-attention module captures the periodic features of the temporal feature vectors themselves and performs weighted processing based on these periodic features to obtain a temporal feature vector with periodic characteristics.

7. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images