Urban sewerage network system missing data reconstruction method based on simulation model

Abstract

The application discloses a city drainage pipe network system missing data reconstruction method based on a simulation model, and comprises the following steps: 1, building a simulation model of the city drainage system; 2, calculating the constant inflow flow data to obtain the liquid level data of the downstream end through the simulation model; 3, making cumulative calculation changing with time to obtain the inflow flow volume data changing with time; 4, constructing a ConvGRUNet neural network model based on the simulation model; 5, taking the real monitoring liquid level data level as the input of the ConvGRUNet neural network model to obtain the cumulative water volume volume contained by the city drainage system; 6, obtaining the relationship function f (level, volume) and performing derivation to obtain the x level of the entire drainage system corresponding to h square ; 7, performing cyclic calculation on the flow at each moment to reconstruct the missing flow data. The application improves the accuracy of the reconstructed drainage pipe network data.

Description

Technical Field

[0001] This invention relates to the field of technology, and in particular to a method for reconstructing missing data in urban drainage network systems based on simulation models. Background Technology

[0002] Against the backdrop of smart city development, urban water supply and drainage networks, as an important infrastructure for social development, play a key role in realizing the integration of intelligent city plants, networks, and stations.

[0003] However, drainage pipe networks differ from water supply pipe networks. Water supply pipe networks are pressurized and the water is clean and free of foreign matter, while drainage pipe networks are gravity-flowed and contain more impurities. The influencing factors of the pipe network system are more complex, and the deployment of intelligent monitoring data in drainage systems is significantly lagging behind that of water supply pipe networks. This results in fewer types and quantities of data that can be collected from drainage systems (generally liquid level data), more missing and abnormal data, and a smaller amount of historical accumulated data available, which in turn affects the accuracy of information feature extraction.

[0004] The data of urban drainage pipe network have both temporal and spatial correlations, are greatly affected by hydrological phenomena, have weak periodicity, and exhibit obvious hydraulic process characteristics.

[0005] Traditional methods for missing data repair mainly rely on existing statistical analysis to find similar data for filling in the gaps or on deep learning to build mathematical models that depend on rich prior knowledge for prediction. However, drainage network data is affected by many factors and is characterized by small data volume, high mechanism complexity, difficulty in representation, and weak periodicity. It is difficult to reconstruct missing drainage network data that conforms to the actual operation of urban drainage network systems using existing methods.

[0006] Existing methods for reconstructing missing data focus on the data itself, using purely mathematical methods to reconstruct the data, while ignoring the impact of hydrological and hydraulic characteristics on urban drainage network data. This results in reconstructed drainage network data that does not conform to actual operational experience of urban drainage systems, and the model has poor generalization ability in urban drainage network data.

[0007] Therefore, how to address the shortcomings of existing technologies in reconstructing drainage network data, which do not conform to the actual operational experience of urban drainage systems and have poor generalization ability in urban drainage network data, has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] In view of the above-mentioned deficiencies of the prior art, the present invention provides a method for reconstructing missing data of urban drainage pipe network system based on simulation model. The purpose is to reconstruct missing data of urban drainage pipe network that conforms to the cognitive experience of actual operation of urban drainage system, improve the accuracy of drainage pipe network data after reconstruction, and make up for the lack of variety, quantity and quality of detectable data of drainage pipe network caused by the lag in the deployment of drainage pipe network detection instruments and the complexity of the internal environment of real drainage pipe network.

[0009] To achieve the above objectives, this invention discloses a method for reconstructing missing data in urban drainage pipe network systems based on simulation models; comprising the following steps:

[0010] Step 1: Build a simulation model of the urban drainage system;

[0011] Step 2: Randomly select the constant inflow flow data V1 of the urbanized drainage pipe network of a node as input, and calculate the downstream liquid level data V2 through the simulation model;

[0012] Step 3: Data preprocessing; Perform cumulative calculations on V1 over time to obtain inflow volume data V3 that varies over time.

[0013] Step 4: Based on the simulation model, construct a ConvGRUNet neural network model and train it using dynamic data.

[0014] Step 5: Use the acquired real monitoring liquid level data (level) as input to the ConvGRUNet neural network model to obtain the calculation result of the ConvGRUNet neural network model, which is used as the cumulative water volume that the urban drainage system can accommodate.

[0015] Step 6: Obtain the relational function f(level, volume) based on the multiple data obtained in Step 5; take the derivative of the relational function f(level, volume) to obtain the corresponding h. level The entire drainage system x square ;

[0016] Step 7: According to the law of conservation of mass, the water level change at any monitoring point in the urban drainage system network at the current moment is related to the flow rate flowing into the monitoring point at the current moment and the flow rate discharged by the pumping station at the current moment. Therefore, according to the formula... By iteratively calculating the traffic at each moment, the missing traffic data can be reconstructed.

[0017] Where y represents missing traffic data;

[0018] x square This represents the current floor area of ​​the drainage pipe network system.

[0019] dealt h This represents the change in liquid level at the current moment.

[0020] v p This represents the amount of water discharged by the pumping station at the current time relative to the previous time.

[0021] t interval The time step for inputting monitoring data.

[0022] Preferably, step 4 includes the following steps:

[0023] Step 4.1: ConvGRUNet neural network model construction, including input layer, feature extraction layer, fully connected layer and output layer;

[0024] The feature extraction layer includes a one-dimensional convolutional layer and a gated recurrent unit layer;

[0025] The one-dimensional convolutional layer uses one-dimensional convolutions with kernel sizes of 2, 3, and 4, and a kernel count of 32, respectively, to extract local features from the input data.

[0026] The local features, after being concatenated, serve as the feature input for the doorway circulation unit layer.

[0027] The gated recurrent unit layer includes a gated neural unit with 32 neurons and a gated neural unit with 16 neurons.

[0028] The fully connected layer consists of fully connected structures with 16, 8, and 4 neurons respectively, and the activation function between each layer is ReLU;

[0029] Step 4.2: Setting model training parameters;

[0030] The parameters that need to be set during the training of the ConvGRUNet neural network model include the number of training epochs, the batch size of samples in each training session, the optimizer, the learning rate lr, the learning rate reduction factor θ in each training epoch, and the training loss function loss.

[0031] Step 4.3: Formation of the training sample dataset;

[0032] Using downstream data S2 as input and data to be reconstructed S1 as output, the training dataset is constructed using a sliding window method. During the construction process, the size s of the sliding window and the sliding step l of the window need to be set.

[0033] Step 4.4: Training the ConvGRUNet neural network model;

[0034] Training is performed using the backpropagation algorithm, as detailed below:

[0035] Step 4.4.1: Initialize the neural network weight parameters w of the ConvGRUNet neural network model.

[0036] Step 4.4.2: The input data x is used to obtain the model prediction result through forward propagation. And calculate the model loss based on the training loss function loss.

[0037] Step 4.4.3: Update the parameters of the ConvGRUNet neural network model using the backpropagation algorithm. new ←w old -lr×adam(w old (,loss), where adam is the selected optimizer;

[0038] Step 4.4.4: Update the learning rate (lr) new =θ×lr old ;

[0039] Step 4.4.5: Repeat steps 4.4.2 to 4.4.4 until the training is complete.

[0040] The beneficial effects of this invention are:

[0041] This invention can reconstruct missing data of urban drainage pipe networks that conforms to the actual operation and cognitive experience of urban drainage systems, improve the accuracy of the reconstructed drainage pipe network data, and make up for the deficiencies in the types, quantity and quality of detectable data of drainage pipe networks caused by the lag in the deployment of drainage pipe network detection instruments and the complexity of the internal environment of the actual drainage pipe network.

[0042] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0043] Figure 1 A flowchart of an embodiment of the present invention is shown.

[0044] Figure 2 The diagram shows the structure of the ConvGRUNet neural network model in one embodiment of the present invention. Detailed Implementation

[0045] Example

[0046] like Figure 1 and Figure 2 As shown, a method for reconstructing missing data in an urban drainage network system based on a simulation model is presented, including the following steps:

[0047] Step 1: Use urban drainage network watershed simulation software to build a simulation model of the urban drainage system;

[0048] Step 2: Randomly select a node of the urban drainage network and use the constant inflow flow rate V1 as input, in cubic meters per second, and calculate the downstream liquid level data V2 through the simulation model, in meters.

[0049] Step 3: Data preprocessing; Perform cumulative calculations on V1 over time to obtain inflow volume data V3, which varies over time and is in cubic meters.

[0050] Step 4: Based on the simulation model, construct a ConvGRUNet neural network model for training and learning on dynamic data, automatically extract the changing characteristics of the urban drainage network system, and realize data mining through a combination of quantitative and qualitative analysis.

[0051] Step 5: Use the acquired real monitoring liquid level data (level, in meters) as input to the ConvGRUNet neural network model to obtain the calculation result of the ConvGRUNet neural network model, which is used as the cumulative water volume that the urban drainage system can hold (volume, in cubic meters).

[0052] Step 6: Obtain the relational function f(level, volume) based on the multiple data obtained in Step 5; take the derivative of the relational function f(level, volume) to obtain the corresponding h. level The entire drainage system x square ;

[0053] Step 7: According to the law of conservation of mass, the water level change at any monitoring point in the urban drainage system is related to the flow rate flowing into the monitoring point at the current moment and the flow rate discharged by the pumping station at the current moment. Therefore, according to the formula... By iteratively calculating the traffic at each moment, the missing traffic data can be reconstructed.

[0054] Where y represents the missing traffic data, in meters (m). 3 / s;

[0055] x square The current floor area of ​​the drainage pipe network system is expressed in square meters.

[0056] dealt h This represents the change in liquid level at the current moment, in meters.

[0057] v p This refers to the amount of water discharged by the pumping station at the current time relative to the previous time, expressed in meters (m³). 3 ;

[0058] t interval The time step for inputting monitoring data.

[0059] This invention uses a calibrated urban drainage simulation model as a foundation, and then uses a ConvGRUNet neural network model, which combines convolutional neural networks and GRU networks, to train and learn the dynamic data between simulated urban drainage pipe networks. This model can automatically extract the changing characteristics of the urban drainage pipe network system and construct a simulated drainage pipe network data mining model that combines quantitative and qualitative analysis.

[0060] Then, by using relevant data that are detectable in reality and have hydraulic relationships with the urban drainage network data to be reconstructed as input to the simulation drainage network data mining model, the inverse reconstruction of the missing urban drainage network data is completed.

[0061] This invention combines a simulation model with a ConvGRUNet neural network model to maximize the utilization of the hydraulic and hydrological process characteristics of urban drainage systems. It reconstructs missing data of urban drainage networks that conforms to the cognitive experience of actual operation of urban drainage systems, improves the accuracy of the reconstructed drainage network data, and makes up for the deficiencies in the types, quantity, and quality of detectable data of drainage networks caused by the lag in the deployment of drainage network detection instruments and the complexity of the internal environment of real drainage networks.

[0062] The missing data of the urban drainage network system reconstructed by this invention can be used to build a subsequent urban flooding early warning and prediction model, enriching the data types and quantity of the urban flooding early warning and prediction model and improving the accuracy of prediction.

[0063] In some embodiments, step 4 includes the following steps:

[0064] Step 4.1: ConvGRUNet neural network model construction, including input layer, feature extraction layer, fully connected layer and output layer;

[0065] The feature extraction layer includes a one-dimensional convolutional layer and a gated recurrent unit layer;

[0066] One-dimensional convolutional layers use one-dimensional convolutions with kernel sizes of 2, 3, and 4, as well as one-dimensional convolutions with 32 kernels, to extract local features from the input data.

[0067] Local features, after being concatenated, serve as the feature input for the doorway loop unit layer.

[0068] The gated recurrent unit layer includes gated neural units with 32 neurons and gated neural units with 16 neurons;

[0069] The fully connected layer consists of fully connected structures with 16, 8, and 4 neurons respectively, and the activation function between each layer is ReLU;

[0070] Step 4.2: Setting model training parameters;

[0071] The parameters that need to be set during training of the ConvGRUNet neural network model include the number of training epochs, the batch size of samples in each training session, the optimizer, the learning rate lr, the learning rate reduction factor θ in each training epoch, and the training loss function loss.

[0072] Step 4.3: Formation of the training sample dataset;

[0073] Using downstream data S2 as input and data to be reconstructed S1 as output, the training dataset is constructed using a sliding window method. During the construction process, the size s of the sliding window and the sliding step l of the window need to be set.

[0074] Step 4.4: Training the ConvGRUNet neural network model;

[0075] Training is performed using the backpropagation algorithm, as detailed below:

[0076] Step 4.4.1: Initialize the neural network weight parameters w of the ConvGRUNet neural network model.

[0077] Step 4.4.2: The input data x is used to obtain the model prediction result through forward propagation. And calculate the model loss based on the training loss function loss.

[0078] Step 4.4.3: Update the parameters of the ConvGRUNet neural network model using the backpropagation algorithm. new ←w old -lr×adam(w old (,loss), where adam is the selected optimizer;

[0079] Step 4.4.4: Update the learning rate (lr) new =θ×lr old ;

[0080] Step 4.4.5: Repeat steps 4.4.2 to 4.4.4 until the training is complete.

[0081] In practical applications, the purpose of one-dimensional convolutional layers in extracting local features from input data is to obtain the local dependencies between input features using different convolutional kernels.

[0082] Gated recurrent unit layers are used to elucidate the connections between different dimensions of local features after feature concatenation and to prevent gradient explosion and gradient vanishing during model training.

[0083] In practical applications, urban drainage pipe network systems consist of three parts: drainage pipes, booster pump stations, and sewage treatment plants. These drainage systems are interconnected and influence each other through these three parts of the drainage pipe network.

[0084] This invention first couples the operation and scheduling of various drainage systems by establishing a unified urban drainage simulation model, and preserves the dynamic hydraulic operation relationship between drainage networks under real hydrological phenomena.

[0085] Then, a neural network combining convolutional neural networks and GRU networks is used to train and learn the dynamic data of simulated urban drainage pipe network systems with hydrological and hydraulic characteristics. The changing characteristics of the urban drainage pipe network system are automatically extracted, and a simulated drainage pipe network data mining model combining quantitative and qualitative analysis is constructed.

[0086] The application of this invention can overcome the dependence of traditional data reconstruction on the amount of actual monitoring data, improve the hydrological and hydraulic characteristics of the reconstructed data, make up for the gap between the reconstructed data of drainage pipe network and the actual operation experience, and has higher generalization ability and robustness.

[0087] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for reconstructing missing data of a city sewerage network system based on a simulation model, characterized in that, Comprising the following steps: Step 1, building a simulation model of the urban drainage system; Step 2, randomly selecting a node of the urban drainage network with constant inflow flow data V1 as input, and calculating the liquid level data V2 at the downstream end through the simulation model; Step 3, data preprocessing; Cumulative calculation of V1 over time to obtain inflow flow volume data V3 over time; Step 4, based on the simulation model, a ConvGRUNet neural network model is constructed for dynamic data training and learning; Step 5, the real monitoring liquid level data level obtained is taken as the input of the ConvGRUNet neural network model, and the calculation result of the ConvGRUNet neural network model is taken as the cumulative water volume volume of the urban drainage system; Step 6, obtaining a relationship function f(level, volume) from the plurality of data obtained according to step 5; deriving the relationship function f(level, volume) to obtain h level x square ; Step 7, according to the law of conservation of mass, the water level change at the current time of any monitoring point in the urban drainage system pipe network is related to the flow into the monitoring point at the current time and the flow pumped away by the pumping station at the current time, so according to the formula The missing flow data can be reconstructed by cyclically calculating the flow at each time. Where y is the missing flow data; x square S is the current moment of the sewer network system bottom area; dealt h is the current moment change in liquid level height; v p Qpump is the water quantity pumped by the pump station at the current time relative to the previous time; t interval is the time step for input monitoring data.

2. The simulation model-based urban sewer system missing data reconstruction method of claim 1, wherein, The step 4 comprises the following steps: Step 4.1, building a ConvGRUNet neural network model, including an input layer, a feature extraction layer, a fully connected layer and an output layer; The feature extraction layer includes a one-dimensional convolution layer and a gated recurrent unit layer; The one-dimensional convolution layer uses one-dimensional convolution kernels with sizes of 2, 3 and 4 respectively and a number of 32 to extract local features from the input data; The local features are used as the feature input of the gated recurrent unit layer after feature splicing; The gated recurrent unit layer includes a gated neural unit with 32 neurons and a gated neural unit with 16 neurons; The fully connected layer includes a fully connected structure with 16, 8 and 4 neurons respectively, and the activation function between each layer is Relu; Step 4.2, setting the model training parameters; The parameters of the ConvGRUNet neural network model that need to be set during training include the number of training rounds epochs, the number of samples batchsize for each training, the model training optimizer optimizer, the training learning rate lr, the training learning rate decay coefficient θ and the training loss function loss; Step 4.3, forming a training sample data set; The downstream data S2 is taken as the input, and the reconstructed data S1 is taken as the output, and the training data set is constructed by the sliding window method, and the size s of the sliding window and the sliding step l of the window need to be set during the construction process; Step 4.4, training of the ConvGRUNet neural network model; The back propagation algorithm is used for training, specifically as follows: Step 4.4.1, initializing the neural network weight parameters w of the ConvGRUNet neural network model Step 4.4.2, the input data x is calculated by forward propagation to obtain the model prediction result And the model loss is calculated according to the training loss function loss Step 4.4.

3. Update the parameters of the ConvGRU Net neural network model, w new ← w - lr x adam(w old , loss), where adam is the optimizer chosen. old , loss), where adam is the optimizer chosen. Step 4.4.4, update learning rate lr new = θ x lr old ; Step 4.4.5, repeating steps 4.4.2 to 4.4.4 until the training is completed.

Images