Pollution source VOCs emission accounting method and system based on electric power data analysis

Abstract

The invention provides a pollution source VOCs emission accounting method and system based on electric power data analysis, and relates to the technical field of city management, and the pollution source VOCs emission accounting method comprises the following steps: S1, obtaining a basic data set of an enterprise in a grid based on a grid matching strategy; the basic data set comprises release time, and VOCs monitoring data, meteorological data and electric power data collected within a moment T; s2, mining data features of the release time, the VOCs monitoring data, the meteorological data and the electric power data, and integrating the data features into the basic data set to obtain a feature data set of an enterprise in the grid; and S3, based on a model fusion strategy of a linear regression model, a Light GBM model and a neural network model, estimating the VOCs at the moment T according to the feature data set to obtain VOCs estimation data at the moment T.

Description

Technical Field

[0001] The present invention relates to the technical field of urban management, and in particular to a method and system for calculating VOCs emissions from pollution sources based on power data analysis. Background Art

[0002] VOCs from pollution sources pose a serious threat to human health and the ecological environment. In terms of the human body, long-term inhalation can cause coughing, vomiting, and even destroy the immune system. In terms of the environment, the main impacts are acid rain, soil and water environment ecology, etc. Therefore, for urban management, the monitoring and control of VOCs from pollution sources is imminent.

[0003] At present, the existing monitoring technologies include installing fixed monitoring sensors and VOCs UAVs. On the one hand, due to the large number of industrial enterprises and the small size of fixed monitoring sensors, insufficient coverage is inevitable; on the other hand, for areas where fixed monitoring sensors are not installed, VOCs UAVs are usually used for regular monitoring. The premise of regular monitoring requires planning vehicle travel routes, which is more difficult, and the efficiency of vehicle travel is low, making it difficult to quickly and comprehensively monitor large areas. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for calculating VOCs emissions from pollution sources based on power data analysis. The VOCs monitoring data are mainly collected through fixed monitoring sensors and VOCs mobile vehicles. The collected VOCs monitoring data are deeply integrated based on the model fusion strategy of linear regression model, LightGBM model and neural network model. The VOCs emissions from pollution sources of target enterprises can be quickly and accurately estimated, providing a solid data foundation for urban governance of VOCs from pollution sources.

[0005] In order to solve the above technical problems, the present invention adopts the following solutions: A method for calculating VOCs emissions from pollution sources based on power data analysis, the method comprising the following steps: S1. Obtaining basic data sets of enterprises in the grid based on grid matching strategy; The basic data set includes the release time and VOCs monitoring data, meteorological data, and power data collected within time T; S2, mining the data features of release time, VOCs monitoring data, meteorological data, and power data, and integrating them into the basic data set to obtain the characteristic data set of enterprises in the grid; S3, based on the model fusion strategy of linear regression model, LightGBM model and neural network model, estimates the VOCs at time T according to the feature data set to obtain the VOCs estimation data at time T.

[0006] Furthermore, the process of obtaining the basic data set of enterprises in the grid based on the grid matching strategy is as follows: The current area is gridded, the longitude and latitude of the enterprise in the current area are obtained, and the enterprise is matched with the grid. When the longitude and latitude of the enterprise are greater than or equal to the longitude and latitude of the lower left point of the grid and less than or equal to the longitude and latitude of the upper right point of the grid, the enterprise is determined to be in the grid, and the basic data set of the enterprise in the grid is obtained based on the grid.

[0007] Furthermore, the data sources of the VOCs monitoring data include fixed sensors and UAVs; when both the fixed sensors and the UAVs collect VOCs monitoring data, the VOCs monitoring data collected by the fixed sensors is selected and integrated into the basic data set. Furthermore, the VOCs monitoring data, meteorological data, and power data collected within time T all correspond to a collection time. Then, in S2, the process of mining the data features of release time, VOCs monitoring data, meteorological data, and power data is as follows: Extract features of the time unit to which the release time belongs, and use the obtained time unit as the time feature corresponding to the release time; Based on the collection time, the power data in the historical time period is aggregated according to different indexes, and the obtained aggregation results are used as the corresponding power index features in the historical time period; Based on the collection time, the VOCs monitoring data at the current collection time and the VOCs monitoring data at the previous collection time are trend-converted, and the obtained trend state is used as the VOCs trend feature corresponding to the current collection time; Based on the collection time, the meteorological data in the historical time period are averaged according to different meteorological elements, and the obtained average value is used as the meteorological index feature corresponding to the historical time period.

[0008] Furthermore, the process of trend conversion is: Based on the collection time, the VOCs monitoring data at the current collection time and the VOCs monitoring data at the previous collection time are obtained in the basic data set, and the difference between the two is calculated; Based on the enterprise, the corresponding enterprise dynamic threshold is obtained, including an upward dynamic threshold and a downward dynamic threshold; When the difference is greater than or equal to the rising dynamic threshold, it means that the current trend state is an rising trend, and the rising trend is used as the VOCs trend feature corresponding to the current collection time; when the difference is less than or equal to the falling dynamic threshold, it means that the current trend state is a falling trend, and the falling trend is used as the VOCs trend feature corresponding to the current collection time; when the difference is between the rising dynamic threshold and the falling dynamic threshold, it means that the current trend state is a stable trend, and the stable trend is used as the VOCs trend feature corresponding to the current collection time.

[0009] Furthermore, the process of obtaining the corresponding enterprise dynamic threshold is as follows: According to the basic data set of the enterprise, all VOCs monitoring data in the basic data set are first-order differentiated to obtain positive and negative differential numbers; The positive difference numbers and negative difference numbers are respectively formed into an ascending list and a descending list, and both the ascending list and the descending list are sorted in ascending order. The data of the preset quantile is found in the ascending list as the ascending dynamic threshold, and the data of the preset quantile is found in the descending list as the descending dynamic threshold.

[0010] Furthermore, S3 includes the following steps: S31, constructing and training a linear regression model based on the feature data set, and predicting the first VOCs monitoring data at time T and the VOCs trend characteristics at time T through the linear regression model; S32, adding the VOCs trend feature at time T to the feature data set, that is, updating the feature data set; S33, constructing and training a LightGBM model based on the updated feature data set, and obtaining the second VOCs monitoring data at time T through the LightGBM model prediction; S34, based on time T, obtain the VOCs monitoring data at time T-1 in the feature data set, perform trend conversion on the first VOCs monitoring data and the second VOCs monitoring data respectively with the VOCs monitoring data at time T-1, and determine whether the two trend states are consistent. If so, use the second VOCs monitoring data as the VOCs estimation data at time T; if not, go to step S35; S35, constructing and training a neural network model based on the updated feature data set, and obtaining the third VOCs monitoring data at time T through prediction by the neural network model; S36, based on time T, obtain the VOCs monitoring data at time T-1 in the feature data set, perform trend conversion on the first VOCs monitoring data and the third VOCs monitoring data respectively with the VOCs monitoring data at time T-1, and determine whether the two trend states are consistent, if so, go to step S37; S37, calculating the error value between the second VOCs monitoring data and the third VOCs monitoring data, performing error correction on the second VOCs monitoring data according to a comparison between the error value and a preset threshold, and using the corrected second VOCs monitoring data as the VOCs estimation data at time T.

[0011] Further, in S37, the error correction process is: When the error value is greater than a preset threshold, the time feature corresponding to the release time is obtained, the working status is obtained according to the time feature, the VOCs monitoring data at the corresponding time T is obtained based on the working status screening, and its average value is calculated to determine whether the second VOCs monitoring data is greater than the average value. If so, the second VOCs monitoring data is revised downward; if not, the second VOCs monitoring data is revised upward.

[0012] Furthermore, the linear regression model uses the power data and power index features in the feature data set as input features and the VOCs monitoring data as output labels; The LightGBM model uses the data other than the VOCs monitoring data and the VOCs trend features in the updated feature data set as input features, and uses the VOCs monitoring data as output labels; The neural network model uses data other than VOCs monitoring data and VOCs trend features in the updated feature data set as input features, and uses VOCs monitoring data as output labels.

[0013] A pollution source VOCs emission accounting system based on power data analysis, characterized in that the pollution source VOCs emission accounting method based on power data analysis is applied, including: Basic data collection module: based on the grid matching strategy, the basic data set of the enterprises in the grid is obtained; the basic data set includes the release time and the VOCs monitoring data, meteorological data, and power data collected within time T; Feature data mining module: mines the data features of release time, VOCs monitoring data, meteorological data, and power data, and integrates them into the basic data set to obtain the feature data set of enterprises in the grid; VOCs estimation module: Based on the model fusion strategy of linear regression model, LightGBM model and neural network model, the VOCs at time T are estimated according to the feature data set to obtain the VOCs estimation data at time T.

[0014] Beneficial effects of the present invention: The present invention provides a method and system for calculating VOCs emissions from pollution sources based on power data analysis. Fixed monitoring sensors and VOCs mobile vehicles commonly used in the prior art are used to collect VOCs monitoring data, and the collected VOCs monitoring data are merged with conventional power data, meteorological data, and time data. Data features are fully utilized, different data features are extracted through feature engineering, and historical and assumed future feature mining strategies are cleverly performed to enhance data diversity.

[0015] In addition, based on the diversity of features, the present invention adopts a model fusion strategy of linear regression model, LightGBM model, and neural network model, obtains diversified prediction results according to different models, and then fuses the prediction results to improve the accuracy of the prediction data. It can quickly and accurately estimate the VOCs emissions from pollution sources of target enterprises, provide a solid data foundation for urban governance of VOCs pollution sources, and help improve urban air quality and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the process of calculating VOCs emissions from pollution sources in Example 1 of the present invention; Figure 2 This is a schematic diagram of the process of VOCs estimation in Example 1 of the present invention. DETAILED DESCRIPTION

[0017] 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, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. 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.

[0018] The relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

[0019] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0020] Additionally, descriptions of well-known structures, functions, and configurations may be omitted for clarity and conciseness.One of ordinary skill in the art will recognize that various changes and modifications may be made to the examples described herein without departing from the spirit and scope of the present disclosure.

[0021] Technologies, methods, and apparatus known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, such technologies, methods, and apparatus should be considered part of the authorization specification.

[0022] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0023] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments: Example 1 In this embodiment, if Figure 1 As shown, a method for calculating VOCs emissions from pollution sources based on power data analysis is provided. Fixed monitoring sensors and VOCs mobile vehicles commonly used in the prior art are used to collect VOCs monitoring data. The collected VOCs monitoring data are merged with conventional power data, meteorological data, enterprise data, and grid data to obtain diverse features. Therefore, based on the diversity of features, a model fusion strategy of linear regression model, LightGBM model, and neural network model is used to estimate VOCs. This can quickly and accurately estimate the VOCs emissions from pollution sources of the target enterprise, provide a solid data foundation for urban governance of VOCs from pollution sources, and help improve urban air quality and sustainable development.

[0024] Specifically, the method for calculating VOCs emissions from pollution sources includes the following steps: S1. Obtaining basic data sets of enterprises in the grid based on grid matching strategy; The basic data set includes the release time and VOCs monitoring data, meteorological data, and power data collected within time T; S2, mining the data features of release time, VOCs monitoring data, meteorological data, and power data, and integrating them into the basic data set to obtain the characteristic data set of enterprises in the grid; S3, based on the model fusion strategy of linear regression model, LightGBM model and neural network model, estimates the VOCs at time T according to the feature data set to obtain the VOCs estimation data at time T.

[0025] First of all, the pollution source VOCs emission accounting method mainly adopts the fixed monitoring sensors and VOCs mobile vehicles commonly used in the prior art to collect VOCs monitoring data. At the same time, meteorological data, enterprise data, grid data, and power data are obtained, and then the collected VOCs monitoring data are merged with conventional meteorological data, enterprise data, grid data, and power data to facilitate the subsequent estimation of the pollution source VOCs emitted by the target enterprise. The key point is that the present invention considers estimating based on power data, deeply integrating power data with VOCs monitoring data, and constructing a set of scientific and reasonable estimation and accounting methods. With this method, the VOCs emission estimation of the target enterprise can be realized quickly and accurately, providing a solid data foundation for urban governance of VOCs pollution, and helping to improve urban air quality and sustainable development.

[0026] In S1, the basic data set of the enterprise in the grid is obtained based on the grid matching strategy. The basic data set refers to the VOCs monitoring data, meteorological data, enterprise data, grid data, and power data combined together by the target enterprise. Specifically, the VOCs monitoring data, meteorological data, enterprise data, grid data, and power data are respectively as follows: The VOCs monitoring data is VOCs sensor monitoring data or VOCs UAV monitoring data; The VOCs sensor monitoring data includes the following information: company name, industry type, company address, fixed VOCs monitoring value, and collection time; The VOCs UAV monitoring data includes the following information: company name, industry type, company address, mobile VOCs monitoring value, and collection time.

[0027] The meteorological data includes the following information: temperature, humidity, wind speed, wind direction, solar radiation, grid ID, and collection time; The enterprise data includes the following information: enterprise name, industry type, enterprise address latitude and longitude; The grid data includes the following information: grid ID, grid center point longitude and latitude, grid lower left point longitude and latitude, grid upper right point longitude and latitude; The power data includes the following information: enterprise name, industry type, power consumption index, and collection time.

[0028] When collecting VOCs monitoring data, meteorological data, and power data, the collection time point can be recorded. According to the collection time, the VOCs monitoring data, meteorological data, enterprise data, grid data, and power data within time T can be merged to form a basic data set. The specific process is as follows: (1) Data type conversion: The collected enterprise name and industry type can be converted into string type; the collected fixed VOCs monitoring value, mobile VOCs monitoring value, temperature, humidity, wind speed, wind direction, solar radiation, grid ID, grid center point longitude and latitude, grid lower left point longitude and latitude, grid upper right point longitude and latitude, enterprise address longitude and latitude, and electricity consumption index can be converted into floating point type; the collected release time and collection time can be converted into time type; (2) VOCs data merging: Since there are two types of VOCs data sources in this embodiment, one is fixed sensor data and the other is UAV data. There may be data duplication. In one embodiment, the strategy adopted is that fixed sensors have a higher priority than UAVs. The reason is that there is a certain distance factor between the UAV and the enterprise's pollution equipment, and there is a certain data error, so the priority of fixed sensors is relatively large. For example, if two data appear at the same time in Enterprise A, one is VOCs sensor monitoring data and the other is VOCs UAV monitoring data, then the VOCs sensor monitoring data is selected as the VOCs monitoring data of Enterprise A for that hour. (3) Enterprise grid matching: Based on the grid matching strategy, the enterprise is matched to the grid, and the rectangular characteristics of the grid are used for matching. If the longitude and latitude of the enterprise are greater than or equal to the longitude and latitude of the lower left point of the grid, and the longitude and latitude of the enterprise are less than or equal to the longitude and latitude of the upper right point of the grid, the enterprise is judged to be in the grid.

[0029] ; Among them, True means that the condition is met. represents the longitude of company A, represents the latitude of enterprise A, Indicates the longitude of the lower left point of the grid, Indicates the latitude of the lower left point of the grid, Indicates the longitude of the upper right point of the grid, Indicates the latitude of the upper right point of the grid; (4) Data merging: Merge fields according to release time and grid ID, merge meteorological data, power data, and VOCs data. The final field sequence is "company name, VOCs monitoring value, temperature, humidity, wind speed, wind direction, solar radiation, electricity consumption index, release time", recorded as DATA.

[0030] The VOCs monitoring value in the field sequence is VOCs monitoring data, the temperature, humidity, wind speed, wind direction, and solar radiation are meteorological data, the electricity consumption index is electricity data, and the release time is used to mark the time characteristics of the field sequence.

[0031] Based on the above principle, in one embodiment, in S2, the process of mining the data features of release time, VOCs monitoring data, meteorological data, and power data is as follows: Extract features of the time unit to which the release time belongs, and use the obtained time unit as the time feature corresponding to the release time; Based on the collection time, the power data in the historical time period is aggregated according to different indexes, and the obtained aggregation results are used as the corresponding power index features in the historical time period; Based on the collection time, the VOCs monitoring data at the current collection time and the VOCs monitoring data at the previous collection time are trend-converted, and the obtained trend state is used as the VOCs trend feature corresponding to the current collection time; Based on the collection time, the meteorological data in the historical time period are averaged according to different meteorological elements, and the obtained average value is used as the meteorological index feature corresponding to the historical time period.

[0032] In one embodiment, the process of trend conversion is: Based on the collection time, the VOCs monitoring data at the current collection time and the VOCs monitoring data at the previous collection time are obtained in the basic data set, and the difference between the two is calculated; Based on the enterprise, the corresponding enterprise dynamic threshold is obtained, including an upward dynamic threshold and a downward dynamic threshold; When the difference is greater than or equal to the rising dynamic threshold, it means that the current trend state is an rising trend, and the rising trend is used as the VOCs trend feature corresponding to the current collection time; when the difference is less than or equal to the falling dynamic threshold, it means that the current trend state is a falling trend, and the falling trend is used as the VOCs trend feature corresponding to the current collection time; when the difference is between the rising dynamic threshold and the falling dynamic threshold, it means that the current trend state is a stable trend, and the stable trend is used as the VOCs trend feature corresponding to the current collection time.

[0033] In one embodiment, the process of obtaining the corresponding enterprise dynamic threshold is specifically as follows: According to the basic data set of the enterprise, all VOCs monitoring data in the basic data set are first-order differentiated to obtain positive and negative differential numbers; The positive difference numbers and negative difference numbers are respectively formed into an ascending list and a descending list, and both the ascending list and the descending list are sorted in ascending order. The data of the preset quantile is found in the ascending list as the ascending dynamic threshold, and the data of the preset quantile is found in the descending list as the descending dynamic threshold.

[0034] Specifically, the release time and the VOCs monitoring data, meteorological data, and power data collected within time T can be obtained according to the field sequence; wherein the VOCs monitoring data, meteorological data, and power data all have time characteristics, and the corresponding VOCs monitoring data, meteorological data, and power data obtained at the collection time are mined and fused to provide effective statistical data for subsequent model estimation.

[0035] Time feature mining: Extract the release time in the DATA dataset to extract features such as quarter, month, day, week, hour, and whether it is a holiday.

[0036] Power index feature mining: Based on the power consumption index in the DATA data set, the maximum, minimum, average, variance and other power index features of the power data in the past 5 hours are mined, and the future 5 hours of the current hour of the historical day are mined. For example, if the current time is 13:00 on January 5, 2024, it is necessary to locate the time to the future 5 hours of 13:00 on January 4, 2024. The final mining time is from 14:00 on January 4, 2024 to 18:00 on January 4, 2024.

[0037] VOCs trend feature mining: If the current 1-hour VOCs monitoring data is to be trended, it is first necessary to obtain the historical 1-hour VOCs monitoring data and calculate the difference between the two; Among them, since the emission ranges of different enterprises are inconsistent, different enterprise dynamic thresholds are set, which may include an increasing dynamic threshold, a decreasing dynamic threshold, and a stable dynamic threshold; the stable dynamic threshold refers to a value between the increasing dynamic threshold and the decreasing dynamic threshold.

[0038] The process of setting dynamic thresholds for different enterprises is as follows: according to the basic data set of the enterprise, all VOCs monitoring data in the basic data set are first-order differentiated, the lists of positive and negative differentials are processed in ascending order, and the 10% quantile is obtained as the dynamic threshold, which is specifically: ; Among them, 0 represents an upward trend, 1 represents a downward trend, and 2 represents a stable trend. Represents the T moment value, Represents T-1 moment value, Represents the 10% quantile in the ascending list, Represents the 10% quantile in the descending list; It can be seen that when the difference is less than or equal to the downward dynamic threshold, it means that the current trend state is a downward trend, and the downward trend is used as the VOCs trend feature corresponding to the current acquisition time; when the difference is between the upward dynamic threshold and the downward dynamic threshold, it means that the current trend state is a stable trend, and the stable trend is used as the VOCs trend feature corresponding to the current acquisition time.

[0039] Meteorological data feature mining: Extract meteorological features based on the DATA data set, including mining the historical three-hour average values ​​of temperature, humidity, wind speed, and solar radiation.

[0040] Based on the above-mentioned time characteristics, power index characteristics, VOCs trend characteristics, and meteorological index characteristics, data integration is performed, and the final output feature sequence is obtained as "company name, VOCs monitoring value, temperature, humidity, wind speed, historical three-hour temperature average, historical three-hour humidity average, historical three-hour wind speed average, wind direction, power index, historical five-hour maximum power, historical five-hour minimum power, historical five-hour average power, historical five-hour power sequence variance, historical three-hour maximum power, historical three-hour minimum power, historical three-hour average power, historical five-hour power sequence variance, historical three-hour maximum power, historical three-hour minimum power, historical three-hour average power, historical three-hour power variance, quarter, month, day, week, hour, whether it is a holiday, VOCs trend characteristics, release time", recorded as DF, that is, the feature data set.

[0041] In one embodiment, S3 includes the following steps: S31, constructing and training a linear regression model based on the feature data set, and predicting the first VOCs monitoring data at time T and the VOCs trend characteristics at time T through the linear regression model; S32, adding the VOCs trend feature at time T to the feature data set, that is, updating the feature data set; S33, constructing and training a LightGBM model based on the updated feature data set, and obtaining the second VOCs monitoring data at time T through the LightGBM model prediction; S34, based on time T, obtain the VOCs monitoring data at time T-1 in the feature data set, perform trend conversion on the first VOCs monitoring data and the second VOCs monitoring data respectively with the VOCs monitoring data at time T-1, and determine whether the two trend states are consistent. If so, use the second VOCs monitoring data as the VOCs estimation data at time T; if not, go to step S35; S35, constructing and training a neural network model based on the updated feature data set, and obtaining the third VOCs monitoring data at time T through prediction by the neural network model; S36, based on time T, obtain the VOCs monitoring data at time T-1 in the feature data set, perform trend conversion on the first VOCs monitoring data and the third VOCs monitoring data respectively with the VOCs monitoring data at time T-1, and determine whether the two trend states are consistent, if so, go to step S37; S37, calculating the error value between the second VOCs monitoring data and the third VOCs monitoring data, performing error correction on the second VOCs monitoring data according to a comparison between the error value and a preset threshold, and using the corrected second VOCs monitoring data as the VOCs estimation data at time T.

[0042] In one embodiment, in S37, the error correction process is: When the error value is greater than a preset threshold, the time feature corresponding to the release time is obtained, the working status is obtained according to the time feature, the VOCs monitoring data at the corresponding time T is obtained based on the working status screening, and its average value is calculated to determine whether the second VOCs monitoring data is greater than the average value. If so, the second VOCs monitoring data is revised downward; if not, the second VOCs monitoring data is revised upward.

[0043] Specifically, in this embodiment, the present invention mainly adopts the model fusion strategy of linear regression model, LightGBM model and neural network model based on the diversity of features, obtains diversified prediction results according to different models, and then fuses the prediction results to improve the accuracy of the prediction data.

[0044] The linear regression model is a trend model. Since the power index is linearly positively correlated with the time factor and the emission concentration, the linear regression model can fully utilize this feature to capture the data trend.

[0045] The LightGBM model is a decision tree selection model. Since the tree model has relatively good generalization ability, takes less training time under the same conditions, and does not require complex parameter adjustment to achieve good results, the model can be set as the main prediction model in this embodiment.

[0046] The neural network model is a parameter correction model. When running under specific circumstances, the neural network model can fully explore the nonlinear semantics and representational relationships between features, and the model accuracy is high. However, due to the large calculation parameters and slow calculation, the model is set as a parameter correction model in this embodiment.

[0047] like Figure 2 As shown in the figure, the process of VOCs estimation using the model fusion strategy of linear regression model, LightGBM model and neural network model is as follows: (1) Training the linear regression algorithm: Filter out the relevant power data and VOCs monitoring data in the feature data set DF and construct a unit linear regression. Use power data and power index features as input features and VOCs as output labels for training. The parameter settings are 5000 training batches, 0.001 learning rate, and MSE loss function. Finally, a linear regression model is obtained. (2) Linear trend acquisition: The VOCs monitoring data at time T is predicted by the linear regression model, and then the VOCs monitoring data at time T-1 is obtained. The linear results are trend-converted to serve as the VOCs trend features at time T. At the same time, the VOCs trend features at time T are added to the feature data set, that is, the feature data set is updated; (3) Training the LightGBM model: The features of the updated feature data set DF except the VOCs monitoring data are used as input features, and VOCs is used as the output label for training. The parameters are set to 1500 training rounds, a learning rate of 0.05, a maximum tree depth of 32, a minimum leaf node of 10, and L2 regularization to prevent overfitting, and finally a decision tree model is obtained; (4) Obtaining the prediction results of the LightGBM model: Since the VOCs trend features at time T are missing in the feature data set at time T when predicting the future, the VOCs trend features at time T are added to the feature data set to obtain a complete decision tree prediction feature set; the invention uses the decision tree model to predict the VOCs monitoring data at time T, and calculates the result trend of the decision tree model to check whether the linear result trend is consistent with the decision tree prediction trend. If not, recursively re-predict until the predicted trend results are consistent; (5) Training the neural network algorithm: The features based on the updated feature data set DF except the VOCs monitoring data are used as input features, and the first-order difference of the VOCs monitoring data is used as the output label for training. The parameters are set to 10,000 training rounds, a learning rate of 0.01, and a loss function of MSE. Finally, the neural network model is obtained; (6) Obtaining the prediction results of the neural network model: Since the VOCs trend features at time T are missing in the feature data set at time T, the VOCs trend features at time T are added to the feature data set to obtain a complete neural network prediction feature set; the neural network model is then used to predict the first-order difference value of the VOCs monitoring data at time T, and the differential trend of the neural network model is calculated to check whether the linear result trend is consistent with the neural network prediction trend. If not, recursively re-predict until the predicted trend results are consistent.

[0048] (7) Corrected output: Get the prediction results of the LightGBM model, and determine whether the error between the prediction results of the neural network model and the prediction results of the LightGBM model is too large. If so, first determine the time characteristics corresponding to the release time. You can check the working status of the day, where the working status includes working days, non-working days, and holidays. Then, according to different working statuses, select the concentration average values ​​of the same 5 historical days at the same time T, recorded as T5_Mean, and determine whether the prediction result of the LightGBM model is greater than 7% of T5_Mean and less than 7% of 15_Mean. If it is greater than 7%, the prediction result of the LightGBM model is subtracted from 0.07% of the prediction result of the neural network model. If it is less than 7%, the prediction result of the LightGBM model is added with 0.07% of the prediction result of the neural network model. If it is less than this interval, it is determined that the prediction is correct and does not need to be corrected and is directly written to the database.

[0049] In summary, the present invention innovatively integrates pollution monitoring data of different scales in depth, giving full play to the advantages of machine learning technology, and can quickly and accurately estimate the VOCs parameters of monitoring blind spots. This technical means not only greatly improves the monitoring efficiency, but also effectively achieves the goal of reducing costs and increasing efficiency.

[0050] In the application, this method adopts the cloud server deployment mode, which is extremely convenient and efficient. For the estimation of industrial enterprises at the city level, it only takes 10 minutes to complete the VOCs parameter estimation using this method. In addition, this method does not have strict requirements on the hardware configuration of the cloud server. Ordinary conventional servers can easily handle the deployment tasks, which greatly reduces the application threshold and cost investment, and provides strong support for wide promotion and application.

[0051] Example 2 A pollution source VOCs emission accounting system based on power data analysis, using a pollution source VOCs emission accounting method based on power data analysis as described above, comprising: Basic data collection module: based on the grid matching strategy, the basic data set of the enterprises in the grid is obtained; the basic data set includes VOCs monitoring data, meteorological data, power data and the collection time within time T; Feature data mining module: mines the features of VOCs monitoring data, meteorological data, power data and the collection time within time T, and integrates them into the basic data set to obtain the feature data set of the enterprises in the grid; VOCs estimation module: Based on the model fusion strategy of linear regression model, LightGBM model and neural network model, the VOCs at time T are estimated according to the feature data set to obtain the VOCs estimation data at time T.

[0052] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. According to the technical essence of the present invention, within the spirit and principles of the present invention, any simple modification, equivalent replacement and improvement made to the above embodiment still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for calculating VOCs emissions from pollution sources based on power data analysis, characterized in that: The method for calculating VOCs emissions from pollution sources comprises the following steps: S1. Obtaining basic data sets of enterprises in the grid based on grid matching strategy; The basic data set includes the release time and VOCs monitoring data, meteorological data, and power data collected within time T; S2, mining the data features of release time, VOCs monitoring data, meteorological data, and power data, and integrating them into the basic data set to obtain the characteristic data set of enterprises in the grid; S3, based on the model fusion strategy of linear regression model, LightGBM model and neural network model, estimates the VOCs at time T according to the feature data set to obtain the VOCs estimation data at time T.

2. According to claim 1, a method for calculating VOCs emissions from pollution sources based on power data analysis is characterized in that: The process of obtaining the basic data set of enterprises in the grid based on the grid matching strategy is as follows: The current area is gridded, the longitude and latitude of the enterprise in the current area are obtained, and the enterprise is matched with the grid. When the longitude and latitude of the enterprise are greater than or equal to the longitude and latitude of the lower left point of the grid and less than or equal to the longitude and latitude of the upper right point of the grid, the enterprise is determined to be in the grid, and the basic data set of the enterprise in the grid is obtained based on the grid.

3. According to claim 1, a method for calculating VOCs emissions from pollution sources based on power data analysis is characterized in that: The data sources of the VOCs monitoring data include fixed sensors and UAVs; when both the fixed sensors and the UAVs collect VOCs monitoring data, the VOCs monitoring data collected by the fixed sensors is selected and integrated into the basic data set.

4. According to claim 1, a method for calculating VOCs emissions from pollution sources based on power data analysis is characterized in that: The VOCs monitoring data, meteorological data, and power data collected within time T all correspond to a collection time. Then, in S2, the process of mining the data features of release time, VOCs monitoring data, meteorological data, and power data is as follows: Extract features of the time unit to which the release time belongs, and use the obtained time unit as the time feature corresponding to the release time; Based on the collection time, the power data in the historical time period is aggregated according to different indexes, and the obtained aggregation results are used as the corresponding power index features in the historical time period; Based on the collection time, the VOCs monitoring data at the current collection time and the VOCs monitoring data at the previous collection time are trend-converted, and the obtained trend state is used as the VOCs trend feature corresponding to the current collection time; Based on the collection time, the meteorological data in the historical time period are averaged according to different meteorological elements, and the obtained average value is used as the meteorological index feature corresponding to the historical time period.

5. A method for calculating VOCs emissions from pollution sources based on power data analysis according to claim 4, characterized in that: The process of trend conversion is: Based on the collection time, the VOCs monitoring data at the current collection time and the VOCs monitoring data at the previous collection time are obtained in the basic data set, and the difference between the two is calculated; Based on the enterprise, the corresponding enterprise dynamic threshold is obtained, including an upward dynamic threshold and a downward dynamic threshold; When the difference is greater than or equal to the rising dynamic threshold, it means that the current trend state is an upward trend, and the upward trend is used as the VOCs trend feature corresponding to the current collection time; When the difference is less than or equal to the downward dynamic threshold, it means that the current trend state is a downward trend, and the downward trend is used as the VOCs trend feature corresponding to the current collection time; When the difference is between the rising dynamic threshold and the falling dynamic threshold, it means that the current trend state is a stable trend, and the stable trend is used as the VOCs trend feature corresponding to the current collection time.

6. A method for calculating VOCs emissions from pollution sources based on power data analysis according to claim 5, characterized in that: The specific process of obtaining the corresponding enterprise dynamic threshold is as follows: According to the basic data set of the enterprise, all VOCs monitoring data in the basic data set are first-order differentiated to obtain positive and negative differential numbers; The positive difference numbers and negative difference numbers are respectively formed into an ascending list and a descending list, and both the ascending list and the descending list are sorted in ascending order. The data of the preset quantile is found in the ascending list as the ascending dynamic threshold, and the data of the preset quantile is found in the descending list as the descending dynamic threshold.

7. A method for calculating VOCs emissions from pollution sources based on power data analysis according to claim 4, characterized in that: The S3 includes the following steps: S31, constructing and training a linear regression model based on the feature data set, and predicting the first VOCs monitoring data at time T and the VOCs trend characteristics at time T through the linear regression model; S32, adding the VOCs trend feature at time T to the feature data set, that is, updating the feature data set; S33, constructing and training a LightGBM model based on the updated feature data set, and obtaining the second VOCs monitoring data at time T through the LightGBM model prediction; S34, based on time T, obtain the VOCs monitoring data at time T-1 in the feature data set, perform trend conversion on the first VOCs monitoring data and the second VOCs monitoring data respectively with the VOCs monitoring data at time T-1, and determine whether the two trend states are consistent. If so, use the second VOCs monitoring data as the VOCs estimation data at time T; if not, go to step S35; S35, constructing and training a neural network model based on the updated feature data set, and obtaining the third VOCs monitoring data at time T through prediction by the neural network model; S36, based on time T, obtain the VOCs monitoring data at time T-1 in the feature data set, perform trend conversion on the first VOCs monitoring data and the third VOCs monitoring data respectively with the VOCs monitoring data at time T-1, and determine whether the two trend states are consistent, if so, go to step S37; S37, calculating the error value between the second VOCs monitoring data and the third VOCs monitoring data, performing error correction on the second VOCs monitoring data according to a comparison between the error value and a preset threshold, and using the corrected second VOCs monitoring data as the VOCs estimation data at time T.

8. The method for calculating VOCs emissions from pollution sources based on power data analysis according to claim 7 is characterized in that: In S37, the error correction process is: When the error value is greater than a preset threshold, the time feature corresponding to the release time is obtained, the working status is obtained according to the time feature, the VOCs monitoring data at the corresponding time T is obtained based on the working status screening, and its average value is calculated to determine whether the second VOCs monitoring data is greater than the average value. If so, the second VOCs monitoring data is revised downward; if not, the second VOCs monitoring data is revised upward.

9. The method for calculating VOCs emissions from pollution sources based on power data analysis according to claim 7 is characterized in that: The linear regression model uses the power data and power index features in the feature data set as input features and the VOCs monitoring data as output labels; The LightGBM model uses the data other than the VOCs monitoring data and the VOCs trend features in the updated feature data set as input features, and uses the VOCs monitoring data as output labels; The neural network model uses data other than VOCs monitoring data and VOCs trend features in the updated feature data set as input features, and uses VOCs monitoring data as output labels.

10. A pollution source VOCs emission accounting system based on power data analysis, characterized in that: A method for calculating VOCs emissions from pollution sources based on power data analysis as described in any one of claims 1 to 9 is applied, comprising: Basic data collection module: based on the grid matching strategy, the basic data set of the enterprises in the grid is obtained; the basic data set includes the release time and the VOCs monitoring data, meteorological data, and power data collected within time T; Feature data mining module: mines the data features of release time, VOCs monitoring data, meteorological data, and power data, and integrates them into the basic data set to obtain the feature data set of enterprises in the grid; VOCs estimation module: Based on the model fusion strategy of linear regression model, LightGBM model and neural network model, the VOCs at time T are estimated according to the feature data set to obtain the VOCs estimation data at time T.

Images