A model data-driven method for constructing spatiotemporal graph convolutional neural network datasets

Abstract

The present invention discloses a model data-driven method for constructing a spatiotemporal graph convolutional neural network dataset, including a method for constructing a structure and equipment model dataset, and a method for constructing a pipeline model dataset. The method is based on a three-dimensional design model or a two-dimensional design model, standardizes the model, adds specific model attribute information, specifies reference coordinate points, develops software to read the processed standardized model geometry data and attribute information, and automatically writes the geometry data and attribute information in the model into text files in batches according to the spatiotemporal graph convolutional neural network algorithm dataset format and professional needs. Combined with the time series data of the classic model, a complete time series and spatial dataset is formed for the spatiotemporal graph convolutional neural network model to perform machine learning training and obtain an artificial intelligence analysis model.

Description

Technical Field

[0001] The present invention relates to the field of model processing technology, and in particular to a method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset. Background Art

[0002] Convolutional Neural Networks (CNNs) are a type of feedforward neural network with a deep structure that incorporates convolutional computations. They are one of the leading algorithms in deep learning. CNNs possess representational learning capabilities and can perform shift-invariant classification of input information based on their hierarchical structure. Hence, they are also known as shift-invariant artificial neural networks (SIANNs).

[0003] Convolutional neural networks are currently rarely used in environmental governance, especially in sewage treatment. They cannot be combined with the time series data of the classic sewage treatment plant model to form a time series spatial dataset for machine learning training of the spatiotemporal graph convolutional neural network model. The accuracy is low, and the efficiency of relying on manual modeling dataset input is low. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a model data-driven spatiotemporal graph convolutional neural network dataset construction method. The method is based on a three-dimensional design model or a two-dimensional design model, standardizes the model, adds specific model attribute information, specifies reference coordinate points, and develops software to read the processed standardized model geometric data and attribute information. According to the spatiotemporal graph convolutional neural network algorithm dataset format and professional needs, the geometric data and attribute information in the model are automatically written into text files in batches. Combined with the time series data of the classic model, a complete time series and spatial dataset is formed for the spatiotemporal graph convolutional neural network model to perform machine learning training and obtain an artificial intelligence analysis model.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset, including a method for constructing a structure and equipment model dataset, specifically comprising the following steps:

[0007] Step 1: Standardize the model layer conversion and formulate layer conversion rules. Convert the equipment model to a special "equipment" layer. Perform the same conversion on other structures. The converted structures and equipment include the pool, cover, railings, stairs, pipes, and hole nodes.

[0008] Step 2: Generate datasets in batches intelligently.

[0009] Preferably, step 1 further includes the following sub-steps:

[0010] Step 1.1: Model Cleanup: Clean up the model by deleting redundant lines and arcs. Develop specific tools to filter and select, and perform various command set operations. Convert to composite lines: Develop specific tools, select the model, and convert the lines, arcs, and circles in the drawing into composite lines of a specified width. The composite line width is usually set to 0. Connect composite lines into closed line strings: Develop specific tools, select the model, and link the composite lines in the drawing into line strings. When the model is well organized, the composite lines can be closed to form closed line strings.

[0011] Step 1.2: Check whether the line strings are closed. By developing a specific tool and selecting the model, the closed line strings in the diagram will be selected. The selected closed line strings can be moved and the unclosed line strings can be processed as needed.

[0012] Step 1.3: Add fences and reference points. Around the trimmed model, match the rectangular frame layer to a standard layer. Copy the reference point block to the reference point location to add reference points to the model for generating the dataset.

[0013] Step 1.4: Add dataset name and comments, using Chinese and English respectively, specify the name of the dataset to be generated and the Chinese comments. Add the dataset name and comments.

[0014] A method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset, also including a pipeline model dataset construction method, specifically comprising the following steps:

[0015] Step A: Model standardization;

[0016] Step B: Add pipe diameter and elevation information to the 2D model's attributes, filter and select the pipe centerline, also known as the polyline, and perform various command set operations to decompose the pipe centerline from the polyline into lines and arcs.

[0017] Step C: Intelligent batch generation of datasets

[0018] Step C1: Read model data, read pipeline model layer data, determine the classification type of the model according to the definition rules, and obtain pipeline parameters according to the attribute parameters of the model;

[0019] Step C2: Write the model data into a file. According to the spatial-temporal graph convolutional neural network algorithm dataset format and professional requirements, automatically write the geometric data and attribute information in the model into a text file in batches.

[0020] Preferably, step A further comprises the following sub-steps:

[0021] Step A1: Layer conversion: formulate layer conversion rules and convert the pipeline model into a special "pipeline" layer. The converted pipelines include sewage pipes, grey water pipes, rainwater pipes, dosing pipes, air pipes, etc.

[0022] Step A2: Model cleaning: perform model cleaning, delete redundant models and model auxiliary lines, develop specific tools to implement filtering and selection, perform multiple command set operations, and improve model cleaning efficiency;

[0023] Step A3: Check for closure. Structures connected to the pipeline and pipeline accessories such as valves and gates can serve as closed ends of the pipeline. Check whether the pipeline is closed. By developing a specific tool and selecting a model, closed pipelines in the diagram will be selected. These selected closed pipelines can be moved, and treatment measures can be taken as needed for unclosed pipelines.

[0024] Step A4: Add fences and reference points. Match the rectangular frame layer to the standard layer around the organized model. Copy the reference point block to the reference point location to add the reference point to the model for generating the dataset.

[0025] Step A5: Add dataset name and comments, using Chinese and English respectively, specify the name of the dataset to be generated and the Chinese comments. Add dataset name and comments.

[0026] Preferably, step B further comprises the following sub-steps:

[0027] Step B1: Add pipe diameter information, filter and select the pipe centerline, assign a value to the pipe diameter, select a reference object, and all object features will become reference features to batch select 2D pipelines. Enter the pipeline attribute data type: PipeDn; 2-PipeElevation; 3-PipeEqipment; 4-PipeEqipmentDist; 5-PipeDnLeft; 6-PipeDnRight; 7-PipeElevationLeft; 8-PipeElevationRight; enter the pipeline attribute data type and the pipeline diameter attribute data.

[0028] Step B2: Add pipeline center elevation information, filter and select pipeline center lines, implement filter selection, perform multiple command set operations, and assign pipeline elevation values;

[0029] Step B3: Add additional equipment information to the pipeline, filter and select the pipeline centerline, perform various command set operations, and add additional equipment information to the pipeline;

[0030] Step B4: Add variable pipe diameter information, filter and select the pipe centerline, perform various command set operations, and assign diameters to both sides of the pipe. The default setting is the starting point diameter and the ending point diameter, which may cause errors. The error part needs to be manually modified in the generated data set.

[0031] Step B5: Add variable elevation information, filter and select the pipeline centerline, implement filter selection, perform multiple command set operations, and assign elevations to both sides of the pipeline. The default elevation is the starting point elevation and the ending point elevation, which may cause errors. The error part needs to be manually modified in the generated data set.

[0032] The advantages and technical effects of the present invention are:

[0033] The present invention provides a method for constructing a spatiotemporal graph convolutional neural network dataset driven by model data, formulates specific equipment, structure, and pipeline standard settings, develops corresponding tool software, standardizes the structure and equipment and pipeline models, implements layer conversion, model cleaning, and closure inspection in batches, and adds fences and benchmarks, dataset names, and annotations. Develop corresponding tool software to read the equipment model layer data, determine the classification type of the model according to the definition rules, obtain the equipment parameters according to the attribute parameters of the model, and automatically write the model geometry data and attribute information in batches into text files according to the spatiotemporal graph convolutional neural network algorithm dataset format and professional needs. Combined with the time series data of the classic model of the sewage treatment plant, a time series spatial dataset is formed for the spatiotemporal graph convolutional neural network model to perform machine learning training and obtain an artificial intelligence analysis model.

[0034] By driving intelligent construction of spatiotemporal graph convolutional neural network algorithm datasets through model data, it is possible to build an algorithm dataset for a model with more than 1,000 nodes in a sewage treatment plant within 5 minutes, with a dataset accuracy rate of more than 95%. The algorithm modeling efficiency is more than 10 times higher than that of manually inputting data. DETAILED DESCRIPTION

[0035] In order to further understand the content, features and effects of the present invention, the following embodiments are given as examples and described in detail with reference to the accompanying drawings. It should be noted that the embodiments are illustrative and not restrictive, and the scope of protection of the present invention cannot be limited thereby.

[0036] A method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset, including a method for constructing a structure and equipment model dataset, specifically comprising the following steps:

[0037] Step 1: Standardize the model layer conversion and formulate layer conversion rules. Convert the equipment model to a special "equipment" layer. Perform the same conversion on other structures. The converted structures and equipment include the pool, cover, railings, stairs, pipes, and hole nodes.

[0038] Step 2: Generate datasets in batches intelligently.

[0039] Preferably, step 1 further includes the following sub-steps:

[0040] Step 1.1: Model Cleanup: Clean up the model by deleting redundant lines and arcs. Develop specific tools to filter and select, and perform various command set operations. Convert to composite lines: Develop specific tools, select the model, and convert the lines, arcs, and circles in the drawing into composite lines of a specified width. The composite line width is usually set to 0. Connect composite lines into closed line strings: Develop specific tools, select the model, and link the composite lines in the drawing into line strings. When the model is well organized, the composite lines can be closed to form closed line strings.

[0041] Step 1.2: Check whether the line strings are closed. By developing a specific tool and selecting the model, the closed line strings in the diagram will be selected. The selected closed line strings can be moved and the unclosed line strings can be processed as needed.

[0042] Step 1.3: Add fences and reference points. Around the trimmed model, match the rectangular frame layer to a standard layer. Copy the reference point block to the reference point location to add reference points to the model for generating the dataset.

[0043] Step 1.4: Add dataset name and comments, using Chinese and English respectively, specify the name of the dataset to be generated and the Chinese comments. Add the dataset name and comments.

[0044] A method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset, also including a pipeline model dataset construction method, specifically comprising the following steps:

[0045] Step A: Model standardization;

[0046] Step B: Add pipe diameter and elevation information to the 2D model's attributes, filter and select the pipe centerline, also known as the polyline, and perform various command set operations to decompose the pipe centerline from the polyline into lines and arcs.

[0047] Step C: Intelligent batch generation of datasets

[0048] Step C1: Read model data, read pipeline model layer data, determine the classification type of the model according to the definition rules, and obtain pipeline parameters according to the attribute parameters of the model;

[0049] Step C2: Write the model data into a file. According to the spatial-temporal graph convolutional neural network algorithm dataset format and professional requirements, automatically write the geometric data and attribute information in the model into a text file in batches.

[0050] Preferably, step A further comprises the following sub-steps:

[0051] Step A1: Layer conversion: formulate layer conversion rules and convert the pipeline model into a special "pipeline" layer. The converted pipelines include sewage pipes, grey water pipes, rainwater pipes, dosing pipes, air pipes, etc.

[0052] Step A2: Model cleaning: perform model cleaning, delete redundant models and model auxiliary lines, develop specific tools to implement filtering and selection, perform multiple command set operations, and improve model cleaning efficiency;

[0053] Step A3: Check for closure. Structures connected to the pipeline and pipeline accessories such as valves and gates can serve as closed ends of the pipeline. Check whether the pipeline is closed. By developing a specific tool and selecting a model, closed pipelines in the diagram will be selected. These selected closed pipelines can be moved, and treatment measures can be taken as needed for unclosed pipelines.

[0054] Step A4: Add fences and reference points. Match the rectangular frame layer to the standard layer around the organized model. Copy the reference point block to the reference point location to add the reference point to the model for generating the dataset.

[0055] Step A5: Add dataset name and comments, using Chinese and English respectively, specify the name of the dataset to be generated and the Chinese comments. Add dataset name and comments.

[0056] Preferably, step B further comprises the following sub-steps:

[0057] Step B1: Add pipe diameter information, filter and select the pipe centerline, assign a value to the pipe diameter, select a reference object, and all object features will become reference features to batch select 2D pipelines. Enter the pipeline attribute data type: PipeDn; 2-PipeElevation; 3-PipeEqipment; 4-PipeEqipmentDist; 5-PipeDnLeft; 6-PipeDnRight; 7-PipeElevationLeft; 8-PipeElevationRight; enter the pipeline attribute data type and the pipeline diameter attribute data. The corresponding parameters have the following meanings:

[0058] (regapp"PIPEDNLEFT")——pipe diameter data, Left is the side of the starting point;

[0059] (regapp"PIPEDNRIGHT")——pipe diameter data, Right is the end point side;

[0060] (regapp"PIPEELEVATIONLEFT")——Elevation data of the pipeline, Left is the side of the starting point;

[0061] (regapp"PIPEElevationRIGHT")——Elevation data of the pipeline, Right is the end point side;

[0062] (regapp"PIPEEQTYPE")——pipe type data;

[0063] (regapp"PIPEEQDIST")——Equipment distance data of the pipeline, one side from the starting point.

[0064] Step B2: Add pipeline center elevation information, filter and select pipeline center lines, implement filter selection, perform multiple command set operations, and assign pipeline elevation values;

[0065] Step B3: Add additional equipment information to the pipeline, filter and select the pipeline centerline, perform various command set operations, and add additional equipment information to the pipeline;

[0066] Step B4: Add variable pipe diameter information, filter and select the pipe centerline, perform various command set operations, and assign diameters to both sides of the pipe. The default setting is the starting point diameter and the ending point diameter, which may cause errors. The error part needs to be manually modified in the generated data set.

[0067] Step B5: Add variable elevation information, filter and select the pipeline centerline, implement filter selection, perform multiple command set operations, and assign elevations to both sides of the pipeline. The default elevation is the starting point elevation and the ending point elevation, which may cause errors. The error part needs to be manually modified in the generated data set.

[0068] In order to more clearly illustrate the specific embodiment of the present invention, an embodiment is provided below:

[0069] Model dataset example

[0070] A. Code form model dataset

[0071] Various forms of data sets are generated according to the algorithm requirements. An example of a data set of pipelines and accessories in code format is shown below.

[0072] co=1;

[0073] CNG.trafficSignViewType = 6; / / 3D pipeline model......

[0074] la="PIPE";

[0075] bp=CNG.AcGePoint3d(xx

[29] ,yy

[163] ,0.00)——;ep=CNG.AcGePoint3d(xx

[35] ,yy

[163] ,0.00)——;CNWaterTreatment.GdrLine(ts,bp,ep,la,co,lsty)——;

[0076] ptaList.Clear()——;ptaList.Add(800)——;ptaList.Add(800)——;ptaList.Add(6.0)——;ptaList.Add(6.0)——;ptaList.Add(1)——;

[0077] ptaList.Add(1500)——;

[0078] ts.pipeDnList.Add(ptaList)——;

[0079] bp=CNG.AcGePoint3d(xx

[42] ,yy[4],0.00)——;ep=CNG.AcGePoint3d(xx

[42] ,yy[3],0.00)——;

[0080] CNWaterTreatment.GdrLine(ts,bp,ep,la,co,lsty)——;ptaList.Clear()——;ptaList.Add(800)——;ptaList.Add(800)——;ptaList.Add(6.0)——;ptaList.Add(6.0)——;ptaList.Add(1)——;

[0081] ptaList.Add(1500)——;

[0082] ts.pipeDnList.Add(ptaList)——;

[0083] bp=CNG.AcGePoint3d(xx

[45] ,yy[2],0.00)——;ep=CNG.AcGePoint3d(xx

[46] ,yy[2],0.00)——;

[0084] CNWaterTreatment.GdrLine(ts,bp,ep,la,co,lsty)——;

[0085] ptaList.Clear()——;ptaList.Add(800)——;ptaList.Add(800)——;ptaList.Add(6.0)——;ptaList.Add(6.0)——;ptaList.Add(1)——;

[0086] ptaList.Add(1500)——;

[0087] ts.pipeDnList.Add(ptaList)——;

[0088] B. Spatial datasets

[0089] According to the dataset format of the spatiotemporal graph convolutional neural network algorithm, the generated pipeline and accessory dataset examples are as follows:

[0090] Table 1 Spatial datasets of spatiotemporal graph convolutional neural network algorithms

[0091]

[0092]

[0093] C. Time series dataset

[0094] According to the dataset format of the spatiotemporal graph convolutional neural network algorithm and professional needs, a time series dataset is created by combining the time series data of the classic model.

[0095] Table 2 Time series datasets of spatiotemporal graph convolutional neural network algorithms

[0096]

[0097]

[0098] Finally, it is preferred that the present invention adopts mature products and mature technical means in the prior art for all the parts not described in the present invention.

[0099] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A model data-driven spatiotemporal graph convolutional neural network dataset construction method, characterized in that: The method for constructing a structure and equipment model dataset includes the following steps: Step 1: Standardize the model layer conversion. Establish layer conversion rules and convert the equipment model to a special "equipment" layer. Perform the same conversion on other structures. The converted structures and equipment include the pool, cover, railings, stairs, pipes, and hole nodes. Step 2: Intelligent batch generation of data sets; The step 1 also includes the following sub-steps: Step 1.1: Model Cleanup: Clean up the model by deleting redundant lines and arcs. Develop specific tools to filter and select, and perform various command set operations. Convert to composite lines: Develop specific tools, select the model, and convert lines, arcs, and circles in the layer into composite lines of specified width. Set the composite line width to 0. Connect composite lines into closed line strings: Develop specific tools, select the model, and link composite lines in the layer into line strings. When the model is well organized, composite lines can be closed to form closed line strings. Step 1.2: Check whether the line string is closed. By developing a specific tool and selecting the model, the closed line strings in the layer will be selected. The selected closed line strings can be moved and the unclosed line strings can be processed as needed. Step 1.3: Add fences and reference points. Around the trimmed model, match the rectangular frame layer to a standard layer. Copy the reference point block to the reference point location to add reference points to the model for generating the dataset. Step 1.4: Add the dataset name and annotation, using Chinese and English respectively, specify the name of the dataset to be generated and the Chinese annotation. Add the dataset name and annotation; The method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset also includes a pipeline model dataset construction method, which specifically includes the following steps: Step A: Model standardization; Step B: Add pipe diameter and elevation information to the 2D model's attributes, filter and select the pipe centerline, and perform various command set operations to decompose the pipe centerline from a polyline into lines and arcs. Step C: Intelligent batch generation of datasets Step C1: Read model data, read pipeline model layer data, determine the classification type of the model according to the definition rules, and obtain pipeline parameters according to the attribute parameters of the model; Step C2: Write the model data into a file. According to the spatial-temporal graph convolutional neural network algorithm dataset format and professional requirements, automatically write the geometric data and attribute information in the model into a text file in batches.

2. The method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset according to claim 1, characterized in that: The step A further comprises the following sub-steps: Step A1: Layer conversion: formulate layer conversion rules and convert the pipeline model into a special "pipeline" layer. The converted pipelines include sewage pipes, grey water pipes, rainwater pipes, dosing pipes, and air pipes. Step A2: Model cleaning: perform model cleaning, delete redundant models and model auxiliary lines, develop specific tools to implement filtering and selection, perform multiple command set operations, and improve model cleaning efficiency; Step A3: Check for closure. The structures connected to the pipeline and the valves and gates attached to the pipeline are considered as the closed ends of the pipeline. Check whether the pipeline is closed. By developing a specific tool and selecting the model, the closed pipelines in the layer will be selected. The selected closed pipelines can be moved, and treatment measures can be taken as needed for the unclosed pipelines. Step A4: Add fences and reference points. Around the trimmed model, match the rectangular frame layer to the standard layer. Copy the benchmark block to the benchmark position to add the benchmark point of the generated data set to the model; Step A5: Add dataset name and comments, using Chinese and English respectively, specify the name of the dataset to be generated and the Chinese comments. Add dataset name and comments.

3. The method for constructing a model data-driven spatiotemporal graph convolutional neural network dataset according to claim 1, characterized in that: The step B further comprises the following sub-steps: Step B1: Add pipe diameter information, filter and select the pipe centerline, assign a value to the pipe diameter, select a reference object, where all object features will become reference features, implement batch selection of 2D pipelines, and enter the pipeline attribute data type: PipeDn; 2-PipeElevation; 3-PipeEqipment; 4-PipeEqipmentDist; 5-PipeDnLeft; 6-PipeDnRight; 7-PipeElevationLeft; 8-PipeElevationRight; input pipeline attribute data type, input pipeline diameter attribute data; Step B2: Add pipeline center elevation information, filter and select pipeline center lines, implement filter selection, perform multiple command set operations, and assign pipeline elevation values; Step B3: Add additional equipment information to the pipeline, filter and select the pipeline centerline, perform various command set operations, and add additional equipment information to the pipeline; Step B4: Add variable pipe diameter information, filter and select the pipe centerline, perform various command set operations, and assign diameters to both sides of the pipe. The default setting is the starting point diameter and the ending point diameter, which may cause errors. The error part needs to be manually modified in the generated data set. Step B5: Add variable elevation information, filter and select the pipeline centerline, implement filter selection, perform multiple command set operations, and assign elevations to both sides of the pipeline; the default elevation is the starting point elevation and the ending point elevation, which will cause errors. The error part needs to be manually modified in the generated data set.

Images