Enterprise pollution discharge behavior analysis method based on electric power data
By mining the correlation between enterprise electricity consumption information and production and pollution control equipment from power data, and building an artificial intelligence equipment operation status identification model, the problems of low efficiency and high cost of enterprise pollution discharge supervision are solved, and intelligent supervision and green transformation are achieved.
Patent Information
- Application Number
- CN202410901119.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-05
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-07-05
AI Technical Summary
Enterprise pollutant discharge supervision faces problems such as a wide variety of equipment, complex production processes, imperfect supervision mechanisms and insufficient supervisory staff, resulting in inefficient supervision and high cost.
By mining the connotation laws from power data, a correlation analysis method between enterprise electricity consumption information and production and pollution control equipment is constructed, and a device operation status recognition model based on the CNN-LSTM-Attention algorithm is built to realize intelligent supervision of enterprise pollution discharge behavior.
It has improved the efficiency of pollutant discharge supervision, reduced the supervision cost, realized data-driven analysis and evaluation of corporate pollutant discharge behaviors, and promoted the green transformation of enterprises.
Smart Images

Figure CN118981599A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of environmental supervision, and in particular to a method for analyzing pollution discharge behavior of an enterprise based on power data. Background Art
[0002] As global climate change and environmental problems become increasingly severe, green, low-carbon and circular have become the new trend of economic development, and the green transformation of enterprises is an inevitable choice that cannot be delayed. However, the supervision of pollution production and discharge of enterprises still faces severe practical difficulties. On the one hand, the number of enterprises is huge, the types are diverse, and the production processes of enterprises are complex; some enterprises, while pursuing economic benefits, ignore the responsibilities and obligations of environmental protection, which makes the supervision of pollution discharge difficult. On the other hand, the supervision methods of relevant environmental supervision units on enterprises generally adopt the extensive supervision of surprise inspections and regular inspections. The supervision mechanism is imperfect, resulting in low supervision efficiency and poor supervision effect. In addition, the number of supervisors is far from meeting the actual supervision needs, resulting in high pressure and heavy supervision tasks, and the scope of supervision is difficult to cover all enterprises. Therefore, how to strengthen the supervision of enterprise pollution discharge and promote the green transformation of enterprises is an important issue that needs to be solved urgently.
[0003] The production of an enterprise cannot be separated from equipment, and the operation of equipment cannot be separated from electricity. Electricity data has the characteristics of large volume, multiple dimensions and high value. Therefore, the present invention starts with electricity data, and by mining the connotation rules hidden in electricity data, constructs an analysis method for the correlation between enterprise electricity consumption information and production and pollution control equipment, and relies on artificial intelligence technology to build a pollution-producing and pollution control equipment state recognition model to realize the analysis and evaluation of enterprise pollution discharge behavior. The present invention provides a new idea for enterprise pollution discharge supervision, which is conducive to improving the efficiency of pollution discharge supervision, alleviating supervision pressure, and reducing supervision costs, and has certain practical value. Summary of the invention
[0004] The purpose of the present invention is to provide a method for analyzing enterprise pollution discharge behavior based on power data. By performing data cleaning and information mining on the power data of the enterprise's production equipment and pollution control equipment, the correlation between the enterprise's electricity consumption information and the enterprise's production and pollution control equipment operating status can be accurately perceived. The key features in the power data can be extracted through feature reconstruction and Pearson correlation analysis technology, and a production equipment and pollution control equipment operating status recognition model based on the CNN-LSTM-Attention algorithm can be built based on artificial intelligence technology, thereby assisting in intelligent decision-making for pollution discharge supervision and achieving a technological breakthrough in the transformation of pollution discharge supervision from manual supervision to intelligent and automated supervision.
[0005] To achieve the above object, the present invention provides a method for analyzing enterprise pollution discharge behavior based on power data, comprising the following steps:
[0006] S1. Obtain the original power data of production equipment and pollution control equipment, build a power history database, classify and label the power data;
[0007] S2. Clean the power data;
[0008] S3, based on the original power data features, construct new power features through feature combination;
[0009] S4, mining the correlation between power data features and the correlation between power data features and equipment operating status, and for highly correlated feature pairs, eliminating redundant features to obtain the original feature set;
[0010] S5, fusing the new power feature constructed in step S3 with the original feature set after removing redundant features in step S4 to obtain a key feature library of power data;
[0011] S6. Build a production equipment and pollution control equipment operation status recognition model based on the CNN-LSTM-Attention algorithm;
[0012] S7. Training and parameter optimization of the production equipment and pollution control equipment operation status recognition model constructed in step S6.
[0013] Preferably, in step S1, the original power data of the enterprise's production equipment and pollution control equipment collected by the contactless smart meter is first obtained, and the original power data includes voltage, current, power and power parameter data, and a power history database is constructed, and the original power data is classified and labeled according to the operating status of the production equipment and the pollution control equipment.
[0014] Preferably, the power data is divided into four categories according to the operating status of the production equipment and the pollution control equipment: both the production equipment and the pollution control equipment are operating, neither the production equipment nor the pollution control equipment is operating, the production equipment is operating and the pollution control equipment is not operating, and the production equipment is not operating and the pollution control equipment is operating.
[0015] Preferably, in step S2, the data cleaning of the power data specifically includes:
[0016] S21. Use the Raida criterion to remove outliers in the power data;
[0017] S22. Use mean filling method to fill missing values in power data;
[0018] S23. Use Savitzky-Golay filter to reduce noise of power data.
[0019] Preferably, in step S3, three new power characteristics are constructed, namely the power difference between the production equipment and the pollution control equipment, the voltage ratio between the production equipment and the pollution control equipment, and the current ratio between the production equipment and the pollution control equipment.
[0020] Preferably, in step S4, for the original power data feature library, by calculating the Pearson correlation coefficients between the power data features and between the power data features and the equipment operation status categories, a Pearson correlation coefficient matrix is constructed, wherein the Pearson correlation coefficient is used to measure the linear correlation between two variables; assuming that the X feature data is X = {X1, X2, ...., X n}, Y feature data is Y={Y1,Y2,....,Y n}, then the Pearson correlation coefficient between features X and Y is calculated as follows:
[0021]
[0022] in, is the mean of the X feature data, is the mean of the Y feature data;
[0023] A correlation coefficient threshold is set. For the constructed Pearson correlation coefficient matrix, features with correlation coefficient values greater than the threshold are identified as highly correlated feature pairs, and redundant features in the highly correlated feature pairs are eliminated.
[0024] Preferably, in step S6, a production equipment and pollution control equipment operation status recognition model based on the CNN-LSTM-Attention algorithm is built, the key features obtained in step S5 are used as the input of the model, and the output result is used as the operation status category label of the production equipment and pollution control equipment in step S1. The specific process is as follows;
[0025] The CNN layer extracts local features from the input data through convolution operations. For a convolution kernel W and input X, the calculation formula for the convolution output Y is:
[0026]
[0027] Among them, b is the bias term, M is the width of the convolution kernel, and N is the height of the convolution kernel;
[0028] The LSTM layer achieves precise control of data information flow through the forget gate, input gate, cell state, and output gate. The update equation of the LSTM unit in one time step is:
[0029] f t =σ(W f ·[h t-1 ,x t ]+b f )i t
[0030] Among them, f t is the activation value of the forget gate, σ represents the sigmoid function, W f is the weight matrix, h t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, b f is the bias term, i t is the activation value of the input gate;
[0031] Apply the Attention mechanism to the output of the LSTM layer: Assume that the output sequence of the LSTM layer is H = [h1,h2,...,h T ],in represents the hidden state of LSTM at time step t, T is the sequence length, d is the dimension of the hidden layer, then the attention score of the i-th position in the sequence at time step t is expressed as:
[0032]
[0033] Among them, w a , W h , W s 、b a is the model parameter, tanh is the hyperbolic tangent activation function, s t-1 represents the hidden state of the previous time step;
[0034] The attention score is converted into attention weight through the softmax function, and the expression formula is:
[0035]
[0036] Among them, α ti represents the attention weight of time step t on the i-th position in the sequence, e ti represents the attention score of time step t on the i-th position in the sequence;
[0037] Use the attention weights to weight the hidden states of the LSTM to get the context vector c t The context vector c is used as input for subsequent layers. t The expression formula is:
[0038]
[0039] Preferably, in step S7, the model training and parameter tuning specifically include the following steps:
[0040] S71, normalizing the power data of the production equipment end and the pollution control equipment end that have undergone data cleaning processing;
[0041] S72, dividing the data set obtained in step S71 into a training set, a validation set, and a test set in proportion;
[0042] S73, input the training set into the production equipment and pollution control equipment operation status recognition model, select accuracy, precision, recall and F1-Score four indicators to evaluate the model performance, and use ten-fold cross validation to obtain the final indicator score of the model;
[0043] S74. Set the model evaluation index standard, verify the model performance on the validation set, and continuously optimize and adjust the model parameters until the model output results meet the index requirements, and obtain a trained production equipment and pollution control equipment operation status recognition model;
[0044] S75. Input the test set into the trained model for testing, output the operating status type information of the production equipment and pollution control equipment, and assist in pollution discharge supervision decision-making.
[0045] Therefore, the present invention adopts the above-mentioned enterprise pollution discharge behavior analysis method based on power data, and the beneficial effects are as follows:
[0046] (1) The present invention analyzes and processes the power data of the enterprise's production equipment and pollution control equipment, thereby realizing the automatic identification of the operating status of the production equipment and pollution control equipment, thereby realizing data-driven analysis and evaluation of the enterprise's pollution discharge behavior, and proposing a new idea for pollution discharge supervision, which can greatly improve the efficiency of pollution discharge supervision, reduce supervision costs, and realize intelligent supervision of enterprise pollution discharge behavior.
[0047] (2) The present invention adopts data cleaning and mining technologies such as the Raida criterion, Savitzky-Golay filtering technology, and Pearson correlation analysis method to effectively improve data quality, enhance data interpretability, and mine implicit information in power data from different dimensions. It constructs a coupling relationship between enterprise electricity consumption information and the operating status of production equipment and pollution control equipment, solves the problem of feature redundancy, and lays a solid foundation for subsequent model building.
[0048] (3) The present invention builds a production equipment and pollution control equipment operating status recognition model based on the CNN-LSTM-Attention algorithm. By combining CNN, LSTM and Attention mechanisms, the model can efficiently extract local features in the data, accurately capture long-term dependencies in time series data, and enhance the ability to focus on key information during model training, which is conducive to improving the accuracy and robustness of the classification model and has good practical application value.
[0049] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 It is an overall flow chart of an embodiment of a method for analyzing enterprise pollution discharge behavior based on power data of the present invention. DETAILED DESCRIPTION
[0051] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.
[0052] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.
[0053] like Figure 1 As shown, a method for analyzing enterprise pollution discharge behavior based on power data includes the following steps:
[0054] S1. Obtain the original power data of production equipment and pollution control equipment, build a power history database, classify and label the power data.
[0055] First, the original power data of the company's production equipment and pollution control equipment collected by non-contact smart meters are obtained. The original power data includes power parameter data such as voltage, current, and power. A power history database is built, and the original power data is classified and labeled according to the operating status of the production equipment and pollution control equipment.
[0056] Specifically, according to the operating status of production equipment and pollution control equipment, the electricity data can be divided into four categories: both production equipment and pollution control equipment are operating, neither production equipment nor pollution control equipment is operating, production equipment is operating and pollution control equipment is not operating, and production equipment is not operating and pollution control equipment is operating. In actual production, the last situation basically does not occur.
[0057] S2. Clean the power data to improve data quality, including:
[0058] S21. Use the Raida criterion to remove outliers in power data.
[0059] Specifically, for a certain power parameter data D = {x1, x2, ..., x n}, calculate the mean and standard deviation of this group of data. The specific formula is as follows:
[0060]
[0061] in, is the mean of the data set, and σ is the standard deviation of the data set. i ,like Then it means x i These are abnormal data and should be removed.
[0062] S22. Use mean filling method to fill missing values in power data;
[0063] Specifically, the python pandas.isnan() function is used to detect whether there are missing values in the data. If there are missing values, the mean filling method is used to fill the missing data, that is, the mean of multiple data before and after the missing data is used to fill the gap. For a certain power parameter data D = {x1, x2, ..., x n}, the calculation formula of the filling value is as follows:
[0064]
[0065] Among them, x' j To fill the value, x j-5 、x j-4 ,....x j+5 The data before and after the missing point.
[0066] S23. Use Savitzky-Golay filter to reduce noise of power data.
[0067] Specifically, the Savitzky-Golay filter is used to process the data noise to achieve data noise reduction and smoothing effects. The Savitzky-Golay filter is a special low-pass filter that fits the value of each data point by constructing a local polynomial and uses the least squares method to optimize the polynomial coefficients to achieve the best fitting effect, thereby achieving the purpose of data denoising.
[0068] For each data point x in the power data i , x after Savitzky-Golay filter processing i ' can be calculated by the following formula:
[0069]
[0070] Where m is the order of the polynomial, c k are the polynomial coefficients calculated by the least squares method, x i+k is the original data sequence x i The data values at the surrounding k distances.
[0071] S3. Based on the original power data features, new power features are constructed through feature combination.
[0072] Specifically, since the original power data contains dimensional information such as voltage, current, and power at the production equipment end and the pollution control equipment end, but these parameter information is not intuitive enough to reflect the correlation between the enterprise's electricity consumption information and the operating status of the production and pollution discharge equipment, in order to explore deeper data connotations, three new and more explanatory power features are constructed by combining the original parameters, namely the power difference between production equipment and pollution control equipment, the voltage ratio between production equipment and pollution control equipment, and the current ratio between production equipment and pollution control equipment.
[0073] S4. By constructing a Pearson correlation coefficient matrix, the correlation between power data features and the correlation between power data features and equipment operating status are mined. Here, power data features refer to parameter features in power data, such as current, voltage, power, etc. in the above steps. Specifically, the correlation between power data features, such as the correlation between voltage and power; the correlation between power data features and equipment operating status, such as the relationship between current and equipment switch status.
[0074] For highly correlated feature pairs, redundant features are removed to obtain the original feature set and achieve feature dimensionality reduction.
[0075] Specifically, for the original power data feature library, the Pearson correlation coefficient matrix is constructed by calculating the Pearson correlation coefficients between power data features and between power data features and equipment operation status categories, where the Pearson correlation coefficient is used to measure the linear correlation between two variables; let X feature data be X = {X1, X2, ...., X n}, Y feature data is Y={Y1,Y2,....,Y n}, then the Pearson correlation coefficient between features X and Y is calculated as follows:
[0076]
[0077] in, is the mean of the X feature data, is the mean of the Y feature data.
[0078] A certain correlation coefficient threshold is set. For the constructed Pearson correlation coefficient matrix, the features with correlation coefficient values greater than the threshold are identified as highly correlated feature pairs. Highly correlated feature pairs often have strong interdependencies and may carry a large amount of overlapping information. In order to reduce the model complexity and computational complexity and improve the model generalization ability, it is necessary to eliminate redundant features in highly correlated feature pairs.
[0079] Specifically, for highly correlated feature pairs, only the feature with the largest absolute value of the correlation coefficient with the equipment operation status category is retained, and the remaining redundant features are eliminated.
[0080] S5. Recombine and fuse the new power features constructed in step S3 with the original feature set after removing redundant features in step S4 to obtain a power data key feature library with minimized dimension and covering almost all power data information, and use these key features as input parameters of subsequent models.
[0081] S6. Build a production equipment and pollution control equipment operation status recognition model based on the CNN-LSTM-Attention algorithm; use the key features obtained in step S5 as the input of the model, and the output result is the operation status category label of the production equipment and pollution control equipment in step S1.
[0082] Specifically, the CNN layer extracts local features from the input data through convolution operations, the core of which is the product and summation process of the convolution kernel and the input features. For a convolution kernel W and input X, the calculation formula for the convolution output Y is:
[0083]
[0084] Among them, b is the bias term, M is the width of the convolution kernel, and N is the height of the convolution kernel.
[0085] The LSTM layer is a special recurrent neural network structure that achieves precise control of data information flow through the forget gate, input gate, cell state, and output gate. The update equation of the LSTM unit in one time step is:
[0086] f t =σ(W f ·[h t-1 ,x t ]+b f )i t ,
[0087] Among them, f t is the activation value of the forget gate, σ represents the sigmoid function, W f is the weight matrix, h t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, b f is the bias term, i t is the activation value of the input gate.
[0088] In this model, the Attention mechanism is applied to the output of the LSTM layer. Suppose the output sequence of the LSTM layer is H = [h1,h2,...,h T ],in represents the hidden state of LSTM at time step t, T is the sequence length, d is the dimension of the hidden layer, then the attention score of the i-th position in the sequence at time step t can be expressed as:
[0089]
[0090] Among them, w a , W h , W s 、b a is the model parameter, tanh is the hyperbolic tangent activation function, s t-1 represents the hidden state at the previous time step.
[0091] Next, the attention score is converted into attention weight through the softmax function, and the expression formula is:
[0092]
[0093] Among them, α ti represents the attention weight of time step t on the i-th position in the sequence, e ti represents the attention score of time step t for the i-th position in the sequence.
[0094] Next, the attention weights are used to perform weighted summation on the hidden states of the LSTM to obtain the context vector c t The context vector c is used as input for subsequent layers. t The expression formula is:
[0095]
[0096] S7. Train and optimize the parameters of the production equipment and pollution control equipment operation status recognition model built in step S6, so that the optimized model can achieve the best recognition effect. When the model recognizes that the pollution control equipment is in the off state when the production equipment is running, it means that the enterprise has not taken pollution control measures and is in violation of production regulations. The enterprise needs to be warned and punished, so as to achieve effective supervision of the enterprise's production and pollution discharge. Model training and parameter tuning specifically include the following steps:
[0097] S71. Normalize the power data of the production equipment end and the pollution control equipment end that have undergone data cleaning.
[0098] S72. Divide the data set obtained in step S71 into a training set, a validation set, and a test set in a ratio of 8:1:1.
[0099] S73. Input the training set into the production equipment and pollution control equipment operation status recognition model, select four indicators including accuracy, precision, recall and F1-Score to evaluate the model performance, and use ten-fold cross validation to obtain the final indicator score of the model.
[0100] The calculation formulas for the four indicators are:
[0101]
[0102] Among them, TP represents true positive examples, TN represents true negative examples, FP represents false positive examples, FN represents false negative examples, and n represents the total number of categories. Precision, recall, and F1-Score are all calculated for specific categories.
[0103] S74. Set the model evaluation index standard, verify the model performance on the validation set, and continuously optimize and adjust the model parameters until the model output results meet the index requirements, thus obtaining a trained production equipment and pollution control equipment operation status identification model.
[0104] S75. Input the test set into the trained model for testing, output the operating status type information of the production equipment and pollution control equipment, and assist in pollution discharge supervision decision-making.
[0105] Therefore, the present invention adopts the above-mentioned enterprise pollution behavior analysis method based on power data, starting from power data, by mining the connotation rules hidden in power data, constructing the correlation analysis method between enterprise electricity information and production and pollution control equipment, and relying on artificial intelligence technology to build a pollution production and pollution control equipment state recognition model to realize the analysis and evaluation of enterprise pollution behavior. It provides a new idea for enterprise pollution supervision, which is conducive to improving the efficiency of pollution supervision, alleviating supervision pressure, and reducing supervision costs, and has certain practical value.
[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution 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 they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A method for analyzing enterprise pollution discharge behavior based on power data, It is characterized in that The following steps are involved: S1. Obtain the original power data of production equipment and pollution control equipment, build a power history database, classify and label the power data; S2. Clean the power data; S3, based on the original power data features, construct new power features through feature combination; S4, mining the correlation between power data features and the correlation between power data features and equipment operating status, and for highly correlated feature pairs, eliminating redundant features to obtain the original feature set; S5, fusing the new power feature constructed in step S3 with the original feature set after removing redundant features in step S4 to obtain a key feature library of power data; S6. Build a production equipment and pollution control equipment operation status recognition model based on the CNN-LSTM-Attention algorithm; S7. Training and parameter optimization of the production equipment and pollution control equipment operation status recognition model constructed in step S6.
2. According to the method for analyzing enterprise pollution discharge behavior based on power data in claim 1, It is characterized in that In step S1, the original power data of the enterprise's production equipment and pollution control equipment collected by the contactless smart meter is first obtained, and the original power data includes voltage, current, power and power parameter data, and a power history database is constructed. The original power data is classified and labeled according to the operating status of the production equipment and pollution control equipment.
3. According to claim 2, a method for analyzing enterprise pollution discharge behavior based on power data, It is characterized in that According to the operating status of the production equipment and the pollution control equipment, the power data is divided into four categories: both the production equipment and the pollution control equipment are operating, neither the production equipment nor the pollution control equipment is operating, the production equipment is operating and the pollution control equipment is not operating, and the production equipment is not operating and the pollution control equipment is operating.
4. According to the method for analyzing enterprise pollution discharge behavior based on power data in claim 3, It is characterized in that In step S2, the power data is cleaned, specifically including: S21. Use the Raida criterion to remove outliers in the power data; S22. Use mean filling method to fill missing values in power data; S23. Use Savitzky-Golay filter to reduce noise of power data.
5. According to claim 4, a method for analyzing enterprise pollution discharge behavior based on power data, It is characterized in that In step S3, three new power characteristics are constructed, namely, the power difference between the production equipment and the pollution control equipment, the voltage ratio between the production equipment and the pollution control equipment, and the current ratio between the production equipment and the pollution control equipment.
6. According to claim 5, a method for analyzing enterprise pollution discharge behavior based on power data, It is characterized in that In step S4, for the original power data feature library, the Pearson correlation coefficient matrix is constructed by calculating the Pearson correlation coefficients between the power data features and between the power data features and the equipment operation status categories, where the Pearson correlation coefficient is used to measure the linear correlation between two variables; let X feature data be X={X 1 ,X 2 ,....,X n }, Y feature data is Y = {Y 1 ,Y 2 ,....,Y n }, then the Pearson correlation coefficient between features X and Y is calculated as follows: in, is the mean of the X feature data, is the mean of the Y feature data; A correlation coefficient threshold is set. For the constructed Pearson correlation coefficient matrix, features with correlation coefficient values greater than the threshold are identified as highly correlated feature pairs, and redundant features in the highly correlated feature pairs are eliminated.
7. According to claim 6, a method for analyzing enterprise pollution discharge behavior based on power data, It is characterized in that In step S6, a production equipment and pollution control equipment operation status recognition model based on the CNN-LSTM-Attention algorithm is built, the key features obtained in step S5 are used as the input of the model, and the output result is used as the operation status category label of the production equipment and pollution control equipment in step S1. The specific process is as follows; The CNN layer extracts local features from the input data through convolution operations. For a convolution kernel W and input X, the calculation formula for the convolution output Y is: Among them, b is the bias term, M is the width of the convolution kernel, and N is the height of the convolution kernel; The LSTM layer achieves precise control of data information flow through the forget gate, input gate, cell state, and output gate. The update equation of the LSTM unit in one time step is: f t =σ(W f ·[h t-1 ,x t ]+b f )i t Among them, f t is the activation value of the forget gate, σ represents the sigmoid function, W f is the weight matrix, h t-1 represents the hidden state of the previous time step, x t represents the input of the current time step, b f is the bias term, i t is the activation value of the input gate; Apply the Attention mechanism to the output of the LSTM layer: Let the output sequence of the LSTM layer be H = [h 1 ,h 2 ,...,h T ],in represents the hidden state of LSTM at time step t, T is the sequence length, d is the dimension of the hidden layer, then the attention score of the i-th position in the sequence at time step t is expressed as: Among them, w a , W h , W s 、b a is the model parameter, tanh is the hyperbolic tangent activation function, s t-1 represents the hidden state of the previous time step; The attention score is converted into attention weight through the softmax function, and the expression formula is: Among them, α ti represents the attention weight of time step t on the i-th position in the sequence, e ti represents the attention score of time step t on the i-th position in the sequence; Use the attention weights to weight the hidden states of the LSTM to get the context vector c t The context vector c is used as input for subsequent layers. t The expression formula is:
8. According to claim 7, a method for analyzing enterprise pollution discharge behavior based on power data, It is characterized in that In step S7, the model training and parameter tuning specifically include the following steps: S71, normalizing the power data of the production equipment end and the pollution control equipment end that have undergone data cleaning processing; S72, dividing the data set obtained in step S71 into a training set, a validation set, and a test set in proportion; S73, input the training set into the production equipment and pollution control equipment operation status recognition model, select accuracy, precision, recall and F1-Score four indicators to evaluate the model performance, and use ten-fold cross validation to obtain the final indicator score of the model; S74. Set the model evaluation index standard, verify the model performance on the validation set, and continuously optimize and adjust the model parameters until the model output results meet the index requirements, and obtain a trained production equipment and pollution control equipment operation status recognition model; S75. Input the test set into the trained model for testing, output the operating status type information of the production equipment and pollution control equipment, and assist in pollution discharge supervision decision-making.
Citation Information
Patent Citations
Enterprise pollution discharge supervision method assisted by electric power model
CN110889088A
Power consumption anomaly detection method and system based on two-channel network and application
CN117272206A
Attention-CNN-LSTM industrial process fault diagnosis method based on improved grey wolf algorithm optimization
CN117421667A
Dynamic gesture recognition method based on hand key point and double-layer bidirectional LSTM network
CN117576783A
Cognitive workload recognition from temporal series information
WO2024015018A1
