Environmental protection equipment digital twin modeling method based on multi-source data fusion

Abstract

The invention discloses an environmental protection equipment digital twin modeling method based on multi-source data fusion, relates to the technical field of digital model establishment, and solves the technical problems of how to effectively process data and carry out targeted improvement and optimization on a model. In combination with the model purpose, deep analysis is carried out, specific implementation modes and application scenes of feature layer fusion and decision-making layer fusion are described in detail, the data fusion effect is improved for feature extraction algorithms of different data sources and decision evaluation and fusion strategies in decision-making layer fusion, and the model function judgment accuracy is improved. According to the method, clear deviation judgment standards and processes are formulated, and comprehensive and highly-targeted optimization strategies are respectively given for different deviation conditions, so that the digital twinborn model can better adapt to the complex and changeable operation environment of the environmental protection equipment, and the prediction accuracy and reliability of the operation state of the environmental protection equipment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital model establishment, and in particular to a digital twin modeling method for environmental protection equipment based on multi-source data fusion. Background Art

[0002] With the continuous improvement of environmental protection requirements, environmental protection equipment is becoming more and more critical in environmental pollution control and resource recycling. Digital twin technology builds a virtual model of physical entities to achieve real-time mapping and status monitoring of environmental protection equipment, providing new ideas for its intelligent management. The operation of environmental protection equipment will generate multi-source data with a wide range of sources and types, such as sensor data, equipment operation data, environmental data, and operation and maintenance data. Integrating this data using multi-source data fusion technology can provide richer data support for digital twin modeling, and has become a hot topic in current research and application.

[0003] According to a patent application with application number CN202110998629.X, a method for digital twin modeling of a circulating water pump is disclosed, the method comprising: performing digital twin analysis of functional feature analysis, structural feature analysis and operating characteristic analysis on the circulating water pump to determine the subsystem set, structural features and operating parameters of the circulating water pump; constructing a digital twin virtual model of the circulating water pump by analyzing the physical object entity of the circulating water pump; the digital twin virtual model includes mechanism model modeling, data model modeling and knowledge model modeling; the obtained subsystem set, structural features and operating parameters of the circulating water pump are interacted and integrated with the constructed digital twin virtual model of the circulating water pump, the communication requirements of the digital twin virtual modeling of the circulating water pump are first analyzed, and then the data transmission scheme is determined to perform data mapping and data fusion to achieve virtual-reality mapping.

[0004] However, some existing digital twin models, when in use, have difficulty in effectively acquiring, preprocessing and fusing multi-source data to build accurate and reliable digital twin models of environmental protection equipment. Data from different sources have problems such as inconsistent formats, noise and error values, and complex data dimensions and features, which bring challenges to data processing and fusion.

[0005] In practical applications, how to accurately determine whether the digital twin model functions normally, and how to quickly locate the problem and effectively optimize it when an abnormality occurs. For example, when predicting the operating status of environmental protection equipment, there is a deviation between the model prediction results and the actual results. How to analyze these deviations and improve the model in a targeted manner to improve its prediction accuracy and reliability? Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a digital twin modeling method for environmental protection equipment based on multi-source data fusion, which solves the problem of how to effectively process data and improve and optimize the model in a targeted manner.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a digital twin modeling method for environmental protection equipment based on multi-source data fusion, the method specifically comprising the following steps:

[0008] Acquire multi-source data and preprocess them to obtain preprocessed data, fuse the preprocessed data to obtain fused data, and establish a digital twin model based on the fused data;

[0009] Collect real-time data of environmental protection equipment and input the real-time data into the digital twin model to obtain the prediction results. At the same time, judge the deviation of the prediction results to generate large stable situation, large unstable situation and small situation;

[0010] The analysis method for large instability is to analyze and optimize the model structure and generate corresponding optimization information;

[0011] For the analysis of the large stability situation, the model parameters are optimized and the data is enhanced and expanded, and the corresponding optimization information is generated;

[0012] Analyze the undersized situation and generate corresponding optimization information by analyzing the algorithm and features of the model;

[0013] Obtain the optimization model and multiple groups of measured parameters at the same time, and input the optimization model to generate prediction results. According to the relationship between the prediction results and the measured results, generate normal signals and optimization signals, and perform secondary optimization processing based on the prediction results.

[0014] As a further solution of the present invention, the specific method of establishing the digital twin model is as follows:

[0015] First, multi-source data is acquired and preprocessed to obtain preprocessed data. Then, the model purpose of the digital twin model is acquired, and a fusion method is selected based on the acquired model purpose to obtain a standard fusion method. The preprocessed data is fused according to the standard fusion method to obtain a fusion result. Then, a suitable model structure is selected based on the fusion result, and a digital twin model is established.

[0016] As a further solution of the present invention, the preprocessing operation includes data cleaning, and the specific methods of data cleaning include removing duplicate data, correcting erroneous values, and filling missing values.

[0017] As a further solution of the present invention, the fusion method includes feature layer fusion and decision layer fusion;

[0018] Feature layer fusion means selecting the most representative, highly relevant and low-redundancy features from a large number of extracted features, and concatenating or fusing the selected features from different data sources;

[0019] Decision-level fusion means fusing multiple decision results using appropriate fusion strategies based on the results of decision evaluation.

[0020] As a further solution of the present invention, the specific method for judging the deviation of the prediction result is:

[0021] Input the real-time data of environmental protection equipment into the digital twin model to obtain the predicted results, and compare the predicted results with the actual results. If the predicted results are the same as the actual results, a normal monitoring signal is generated. Otherwise, if the predicted results are different from the actual results, an abnormal analysis result is generated.

[0022] Analyze the abnormal analysis results, obtain multiple groups of real-time data, and obtain the corresponding prediction results. At the same time, calculate the numerical difference between the actual result and the predicted result, and then identify the numerical distribution of the numerical difference to obtain the large stable situation, the large unstable situation and the small situation.

[0023] As a further solution of the present invention, the specific method of troubleshooting and optimizing the model structure is as follows:

[0024] Check the model structure. For neural network models, if the number of model layers is too small, gradually increase the number of hidden layers and observe the changes in deviations on the validation set; for physical modeling of environmental protection equipment.

[0025] As a further solution of the present invention, the specific method of optimizing the parameters of the model and enhancing and expanding the data is as follows:

[0026] Analysis of hyperparameter adjustment: Try to use a preset step size and observe the loss value on the validation set. If the loss value rises rapidly, it indicates that the learning rate is too large, and the step size should be reduced; if it decreases slowly, the step size should be increased.

[0027] For data enhancement analysis, increase diversity, add Gaussian noise with a reasonable intensity set according to the data fluctuation range, or perform data translation.

[0028] As a further solution of the present invention, the specific method of analyzing the algorithm and characteristics of the model is as follows:

[0029] Analyze algorithm improvements, introduce more advanced algorithms, use more efficient algorithms in data fusion, and apply new optimization algorithms in model training to improve model performance.

[0030] Analyze feature optimization, re-sort features, and explore more representative features.

[0031] The present invention provides a digital twin modeling method for environmental protection equipment based on multi-source data fusion. Compared with the prior art, it has the following beneficial effects:

[0032] The present invention selects the fusion method and conducts in-depth analysis in combination with the purpose of the model, and explains in detail the specific implementation methods and applicable scenarios of feature layer fusion and decision layer fusion, such as feature extraction algorithms for different data sources and decision evaluation and fusion strategies in decision layer fusion, to help users more accurately select the appropriate fusion method and improve the data fusion effect. In terms of model function judgment, clear deviation judgment standards and processes are formulated, and comprehensive and targeted optimization strategies are given for different deviation situations to improve model performance. This enables the digital twin model to better adapt to the complex and changeable operating environment of environmental protection equipment and improve the accuracy and reliability of the prediction of the operating status of environmental protection equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a diagram of the steps of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] See also Figure 1 The present application provides a digital twin modeling method for environmental protection equipment based on multi-source data fusion, which specifically includes the following steps:

[0036] Step 1: Acquire multi-source data, pre-process the acquired multi-source data to obtain pre-processed data, fuse the pre-processed data to obtain fused data, and establish a digital twin model based on the fused data.

[0037] First, multi-source data is acquired, and multi-source data is represented as data from different sources, such as sensor data (including real-time data collected by various sensors such as temperature, pressure, flow, etc., which is used to reflect the operating status of environmental protection equipment), equipment operation data (such as equipment on / off time, operating time, energy consumption, etc., which records the operating history of the equipment), environmental data (environmental factors such as ambient temperature, humidity, pollutant concentration, etc. have an important impact on the operation of environmental protection equipment, and operation and maintenance data include equipment maintenance records, fault reports, etc., which provide a basis for equipment health management), and at the same time, the obtained multi-source data is pre-processed to obtain pre-processed data, and the pre-processing operation specifically refers to data cleaning and data transformation of multi-source data, where data cleaning means identifying and processing these problem data, and improving the accuracy and completeness of the data by removing duplicate data, correcting erroneous values, and filling missing values. For example, for abnormal jump values ​​in sensor data, filtering algorithms can be used for smoothing; for missing operation and maintenance data, reasonable estimates can be made based on the historical maintenance rules of the equipment and the operation and maintenance records of similar equipment to fill in the data, and data transformation means standardizing or normalizing the data;

[0038] Then, the model purpose of the digital twin model is obtained, and the model purpose here is expressed as operation monitoring, fault prediction and other purposes, and a fusion method is selected based on the obtained model purpose to obtain a standard fusion method, and the fusion methods include feature layer fusion and decision layer fusion. The feature layer fusion means selecting the most representative, highly relevant and low-redundancy features from a large number of extracted features, and splicing or otherwise fusing the features of different selected data sources. Specifically, for different types of data sources, corresponding feature extraction algorithms are used. For the time series data of the sensor, statistical features such as mean, variance, and peak index can be extracted; for image data, algorithms such as scale-invariant feature transform (SIFT) and accelerated robust features (SURF) can be used to extract key feature points of the image; for operation and maintenance data in text form, methods such as bag of words and TF-IDF (term frequency-inverse document frequency) can be used to extract text features. Decision layer fusion means that according to the results of decision evaluation, appropriate fusion strategies are used to fuse multiple decision results. For example, for fault diagnosis of environmental protection equipment, sensor data is input into a fault diagnosis model based on neural networks, and equipment operation data is input into a fault diagnosis system based on rule reasoning. Each of them outputs preliminary decision results on whether the equipment has a fault and the type of fault. Each independent decision result is evaluated to determine its reliability and accuracy. Its performance can be measured by calculating indicators such as confidence and recall rate of each decision result. Different weights are assigned according to indicators such as confidence of each decision result, and then weighted average is performed to obtain the final decision. The preprocessed data is fused according to the obtained standard fusion method to obtain the fusion result. Then, based on the fusion result, a suitable model structure is selected and a digital twin model is established. If the fused data presents an obvious linear relationship, and the application goal is a simple prediction of operating parameters, such as predicting the flow of sewage treatment equipment, a linear regression model can be selected. If the fused data features are complex, there is a nonlinear relationship, and it is used for complex fault diagnosis, a neural network model can be selected. Taking the fault diagnosis of garbage incineration equipment as an example, its fused data contains complex information such as multi-dimensional parameters of equipment operation and combustion image features. Selecting a neural network with multiple hidden layers can better learn complex patterns in the data. This is a prior art and will not be elaborated on here.

[0039] Step 2: Collect real-time data of environmental protection equipment and input the real-time data into the digital twin model to obtain prediction results. At the same time, judge the deviation of the prediction results and generate deviation identification results.

[0040] Acquire real-time data of environmental protection equipment, input the real-time data into the digital twin model for analysis, obtain prediction results, and compare the obtained prediction results with the actual results. If the prediction results are the same as the actual results, and the same means that the difference between the prediction results and the actual results is within the preset range, it means that the overall function of the digital twin model is normal and a normal monitoring signal is generated. On the contrary, if the prediction results are different from the actual results, it means that the overall function of the digital twin model is abnormal, and an abnormal analysis result is generated;

[0041] For example, the digital twin model predicts the sewage treatment efficiency and effluent water quality at the next moment based on the current sewage flow, water quality and equipment operating parameters, and presets a reasonable error range. This range is determined based on the accuracy requirements of environmental protection equipment, actual operating experience and industry standards. For the prediction of COD concentration in the effluent water quality of sewage treatment equipment, the preset error range is ±5mg / L. If the difference between the predicted result and the actual result is within the preset range, the predicted result is considered to be the same as the actual result, indicating that the overall function of the digital twin model is normal, the system generates normal monitoring signals, and continues to perform routine operation monitoring work. Conversely, if the difference exceeds the preset range, the predicted result is different from the actual result.

[0042] Process the obtained abnormal analysis results, obtain multiple sets of real-time data, and obtain the corresponding prediction results. At the same time, calculate the numerical difference between the actual result and the predicted result, and then identify the numerical distribution of the numerical difference to obtain the deviation identification result. The specific deviation identification results include the large stable situation, the large unstable situation and the small situation. And here, the numerical difference is compared with the preset value to determine whether it is too large or too small. The specific value of the preset value is set by the operator. Assume that the operator sets the preset value of the ammonia nitrogen concentration in the effluent water quality of the sewage treatment equipment to ±0.5mg / L, and obtain 5 sets of real-time data;

[0043] The corresponding actual ammonia nitrogen concentrations are 1.8mg / L, 2.2mg / L, 1.9mg / L, 2.3mg / L, and 2.0mg / L respectively;

[0044] The prediction results of the digital twin model were 2.5mg / L, 2.8mg / L, 2.4mg / L, 1.5mg / L, and 2.7mg / L;

[0045] Through calculation, the numerical differences of the five groups of data are 0.7mg / L, 0.6mg / L, 0.5mg / L, -0.8mg / L, and 0.7mg / L respectively;

[0046] Among them, 0.7mg / L, 0.6mg / L, and 0.7mg / L are greater than the preset upper limit of 0.5mg / L, and the difference is relatively stable in multiple measurements, which is a relatively large and stable situation;

[0047] For the data set of 1.5mg / L, the difference of -0.8mg / L is less than the preset lower limit of -0.5mg / L, which is a relatively small situation. If the difference of the data obtained within a period of time fluctuates greatly in the larger direction, such as the difference is 0.5mg / L, 1.2mg / L, and 0.8mg / L respectively, it is a relatively large and unstable situation.

[0048] Step 3: Analyze the large stable situation, large unstable situation and small situation separately. The analysis method for the large unstable situation is to check and optimize the model structure and generate corresponding optimization information. The specific processing method is as follows:

[0049] Neural network model: When the deviation of the model fluctuates greatly in different time periods or different data subsets, it may be that the model structure is unreasonable. For example, if the number of model layers is too small, for complex environmental protection equipment fault diagnosis tasks, it is impossible to fully extract the deep-level features in the data, resulting in large deviations. At this time, the number of hidden layers can be gradually increased, and the model can be retrained after each layer is added, and the deviation changes on the validation set can be observed. If the deviation gradually decreases and tends to be stable, it means that increasing the number of layers is effective; conversely, if the deviation does not improve or even increases, other structural parameters may need to be adjusted, such as the number of nodes;

[0050] Physical model: In the physical modeling of environmental protection equipment, if the model structure does not match the actual physical process, it will also lead to unstable deviations. For example, when establishing a fluid mechanics model for sewage treatment equipment, if too many simplified assumptions are made and key physical factors are ignored, the model prediction results will deviate greatly from the actual situation.

[0051] For the analysis of the large stability situation, the model parameters are optimized and the data is enhanced and expanded, and the corresponding optimization information is generated. The specific processing method is as follows:

[0052] Hyperparameter adjustment: The model may fall into a poor local optimal solution. Taking the common gradient descent algorithms (such as stochastic gradient descent SGD, Adagrad, Adadelta, Adam, etc.) as an example, the learning rate should be adjusted first. If the learning rate is too large, the model may not converge or even diverge; if the learning rate is too small, the model converges very slowly. Generally, you can first try with a larger step size (such as 0.1, 0.01, 0.001) to observe the changes in the loss value of the model on the validation set. If the loss value rises rapidly, it means that the learning rate is too large; if the loss value decreases slowly, the learning rate can be appropriately increased. After determining the approximate range, fine-tune with a smaller step size;

[0053] Data enhancement: If insufficient data leads to poor model generalization, for image data, use Python's OpenCV, Scikit-Image and other libraries to perform rotation, flipping, and scaling operations. For example, the appearance image of environmental protection equipment can be randomly rotated by a certain angle (such as ±15°), flipped horizontally or vertically, and scaled proportionally (such as 0.8-1.2 times) to increase the diversity of the image. For time series data, Gaussian noise can be added, and the noise intensity can be reasonably set according to the fluctuation range of the data; or data can be shifted to simulate the operation data at different starting times.

[0054] Data expansion: In addition to data enhancement, more actual operation data can be collected. Cooperate with environmental protection companies to obtain environmental protection equipment operation data under different working conditions and time periods. You can also generate some data through simulation, add it to the data set, retrain the model, and check whether the deviation has improved.

[0055] For the undersized situation, we analyze the algorithm and features of the model to generate corresponding optimization information. The specific processing method is as follows:

[0056] Algorithm improvement: You can try to introduce more advanced algorithms. For example, use a more efficient fusion algorithm in the data fusion stage, or use a new optimization algorithm in model training to improve model performance and further reduce deviation.

[0057] Feature optimization: Re-examine feature engineering and explore more representative features. For the operation data of environmental protection equipment, we can combine domain knowledge to extract new features or combine and transform existing features to improve the model's understanding and prediction capabilities of data.

[0058] The digital twin model is optimized based on the obtained optimization information to generate an optimized model.

[0059] Step 4. Obtain the optimization model and multiple sets of measured parameters at the same time, and input the optimization model to generate prediction results. According to the relationship between the prediction results and the measured results, generate normal signals and optimized signals, and perform secondary optimization processing based on the prediction results.

[0060] Some of the data in the above formulas are calculated by taking their numerical values, and parameter units are not input for calculation. At the same time, the contents not described in detail in this specification belong to the existing technologies known to those skilled in the art.

[0061] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A digital twin modeling method for environmental protection equipment based on multi-source data fusion, characterized in that: The method specifically comprises the following steps: Acquire multi-source data and preprocess them to obtain preprocessed data, fuse the preprocessed data to obtain fused data, and establish a digital twin model based on the fused data; Collect real-time data of environmental protection equipment and input the real-time data into the digital twin model to obtain the prediction results. At the same time, judge the deviation of the prediction results to generate large stable situation, large unstable situation and small situation; The analysis method for large instability is to analyze and optimize the model structure and generate corresponding optimization information; For the analysis of the large stability situation, the model parameters are optimized and the data is enhanced and expanded, and the corresponding optimization information is generated; Analyze the undersized situation and generate corresponding optimization information by analyzing the algorithm and features of the model; Obtain the optimization model and multiple groups of measured parameters at the same time, and input the optimization model to generate prediction results. According to the relationship between the prediction results and the measured results, generate normal signals and optimization signals, and perform secondary optimization processing based on the prediction results.

2. According to claim 1, a digital twin modeling method for environmental protection equipment based on multi-source data fusion is characterized in that: The specific methods of building a digital twin model are as follows: First, multi-source data is acquired and preprocessed to obtain preprocessed data. Then, the model purpose of the digital twin model is acquired, and a fusion method is selected based on the acquired model purpose to obtain a standard fusion method. The preprocessed data is fused according to the standard fusion method to obtain a fusion result. Then, a suitable model structure is selected based on the fusion result, and a digital twin model is established.

3. The method for digital twin modeling of environmental protection equipment based on multi-source data fusion according to claim 2 is characterized in that: Preprocessing operations include data cleaning, and specific methods of data cleaning include removing duplicate data, correcting erroneous values, and filling missing values.

4. The method for digital twin modeling of environmental protection equipment based on multi-source data fusion according to claim 2 is characterized in that: Fusion methods include feature layer fusion and decision layer fusion; Feature layer fusion means selecting the most representative, highly relevant and low-redundancy features from a large number of extracted features, and concatenating or fusing the selected features from different data sources; Decision-level fusion means fusing multiple decision results using appropriate fusion strategies based on the results of decision evaluation.

5. The method for digital twin modeling of environmental protection equipment based on multi-source data fusion according to claim 1 is characterized in that: The specific method for judging the deviation of the prediction results is as follows: Input the real-time data of environmental protection equipment into the digital twin model to obtain the predicted results, and compare the predicted results with the actual results. If the predicted results are the same as the actual results, a normal monitoring signal is generated. Otherwise, if the predicted results are different from the actual results, an abnormal analysis result is generated. Analyze the abnormal analysis results, obtain multiple groups of real-time data, and obtain the corresponding prediction results. At the same time, calculate the numerical difference between the actual result and the predicted result, and then identify the numerical distribution of the numerical difference to obtain the large stable situation, the large unstable situation and the small situation.

6. The method for digital twin modeling of environmental protection equipment based on multi-source data fusion according to claim 1 is characterized in that: The specific methods for troubleshooting and optimizing the model structure are as follows: Check the model structure. For neural network models, if the number of model layers is too small, gradually increase the number of hidden layers and observe the changes in deviations on the validation set; for physical modeling of environmental protection equipment.

7. The method for digital twin modeling of environmental protection equipment based on multi-source data fusion according to claim 1 is characterized in that: The specific methods for optimizing model parameters and enhancing and expanding data are as follows: Analysis of hyperparameter adjustment: Try to use a preset step size and observe the loss value on the validation set. If the loss value rises rapidly, it indicates that the learning rate is too large, and the step size should be reduced; if it decreases slowly, the step size should be increased. For data enhancement analysis, increase diversity, add Gaussian noise with a reasonable intensity set according to the data fluctuation range, or perform data translation.

8. The method for digital twin modeling of environmental protection equipment based on multi-source data fusion according to claim 1 is characterized in that: The specific methods for analyzing the algorithm and features of the model are as follows: Analysis of algorithm improvement, introduction of more advanced algorithms, use of more efficient algorithms in data fusion, and application of new optimization algorithms in model training to improve model performance; Analyze feature optimization, re-sort features, and explore more representative features.

Images