An air concentration prediction method for an enterprise outfall

Abstract

The application discloses an air concentration prediction method for an enterprise discharge outlet, and comprises the following steps: inputting multiple groups of production parameters of a historical period into a production parameter prediction model to obtain multiple groups of predicted production parameters of a future period; wherein the production parameters comprise time parameters; inputting the time parameters of each group of the historical period and the future period into an air concentration preliminary prediction model to obtain preliminary predicted air concentration parameters of each group; and inputting the multiple groups of production parameters of the historical period, the air concentration parameters of a previous group, the preliminary predicted air concentration parameters, and the multiple groups of predicted production parameters of the future period and the preliminary predicted air concentration parameters into an air concentration final prediction model to obtain multiple groups of final predicted air concentration parameters of the future period. The air concentration prediction accuracy of the enterprise discharge outlet is improved.

Description

Technical Field

[0001] This application belongs to the field of environmental monitoring and artificial intelligence technology, and specifically relates to a method for predicting air concentration at enterprise sewage outlets. Background Technology

[0002] Predicting industrial gaseous pollutant emission concentrations is a core component of the environmental monitoring system. Accurately predicting the spatiotemporal trends of atmospheric pollutant concentrations is crucial for formulating and revising air pollution control measures and continuously improving air quality.

[0003] In recent years, with the development of deep learning technology, data-driven time series forecasting methods have been widely used in this field. However, existing technologies still have the following shortcomings in practical applications targeting enterprise wastewater discharge outlets: 1. Traditional pollutant prediction models treat production parameters and emission parameters as independent time series variables. In essence, they adopt a single-variable or finite-variable time series modeling approach, relying solely on historical concentration data for extrapolation and prediction. They ignore the nonlinear transmission effect of dynamic fluctuations in enterprise production process parameters (such as production volume, power consumption, etc.) on pollutant emission concentrations. When enterprise production processes are adjusted or equipment is started or stopped, the prediction accuracy decreases significantly. 2. When integrating multi-source data, traditional pollutant prediction models typically input production and emission parameters into the prediction model by simply concatenating features or applying fixed weights. This static fusion method ignores the dynamic coupling relationship between multi-source features—the weights of different features on emission concentrations at different times should change dynamically with the operating context. It cannot adaptively adjust the importance weights of each feature according to the time-series context, resulting in feature information at critical operating condition abrupt changes being submerged by a large amount of irrelevant noise, which restricts the prediction accuracy in complex industrial scenarios.

[0004] Therefore, there is an urgent need for a new technical solution to predict the air concentration at enterprise sewage outlets. Summary of the Invention

[0005] To address the aforementioned problems, this application provides a method for predicting air concentration at enterprise wastewater discharge outlets, thereby resolving the technical issues raised in the background section.

[0006] A method for predicting air concentration at enterprise wastewater discharge outlets includes: By inputting multiple sets of production parameters from historical periods into a production parameter prediction model, multiple sets of predicted production parameters for future periods are obtained; among these, production parameters include time parameters. The time parameters of each group in the historical period and the future period are input into the preliminary air concentration prediction model to obtain the preliminary predicted air concentration parameters of each group. By inputting multiple sets of production parameters from historical periods, the previous set of air concentration parameters, the preliminary predicted air concentration parameters, and multiple sets of predicted production parameters and preliminary predicted air concentration parameters from future periods into the final air concentration prediction model, multiple sets of final predicted air concentration parameters for future periods are obtained.

[0007] Furthermore, the production parameter prediction model includes multiple historical period production parameter prediction modules and multiple future period production parameter prediction modules connected in sequence. The first historical cycle production parameter prediction module is used to determine the predicted production parameters and status information for the next group based on the production parameters of the first group. The historical cycle production parameter prediction module, excluding the first one, is used to determine the predicted production parameters and status information of the next group based on the production parameters, predicted production parameters, and status information of the current group. The future cycle production parameter prediction module, excluding the last one, is used to determine the predicted production parameters and status information of the next group based on the predicted production parameters and status information of the current group. The last future cycle production parameter prediction module is used to determine the predicted production parameters for the next group based on the predicted production parameters and status information of the current group.

[0008] Furthermore, the historical cycle production parameter prediction module includes a first data fusion module and a first production parameter status prediction module; the future cycle production parameter prediction module includes a second data fusion module and a second production parameter status prediction module. The first data fusion module is used to weight and fuse the production parameters of the current group based on the attention mechanism to obtain the fused feature vector of the current group; The first production parameter state prediction module of the first historical cycle production parameter prediction module is used to predict the fused feature vector of the first group based on the Long Short-Term Memory Network (LSTM) to obtain the predicted production parameters and state information of the next group. The first production parameter state prediction module of the historical cycle production parameter prediction module, excluding the first one, is used to predict the fused feature vector, predicted production parameters and state information of the current group based on LSTM, so as to obtain the predicted production parameters and state information of the next group. The second data fusion module is used to perform weighted fusion of the predicted production parameters of the current group based on the attention mechanism to obtain the fusion feature vector of the current group. The second production parameter state prediction module of the last future cycle production parameter prediction module is used to predict the fused feature vector, predicted production parameters and state information of the current group based on LSTM to obtain the predicted production parameters of the next group. The second production parameter state prediction module, excluding the last one, is used to predict the fused feature vector, predicted production parameters, and state information of the current group based on LSTM, so as to obtain the predicted production parameters and state information of the next group.

[0009] Further, in the first data fusion module or the second data fusion module, the fused feature vector of the current group is calculated using the following formula: ; in, Indicates the first The fusion feature vector of the group , , They represent the first The initial weight matrix, key weight matrix, and value weight matrix of the group. Represents a constant. This indicates the normalization indicator function; in, In the first data fusion module, the first The production parameters of the group are represented in the second data fusion module. The group's predicted production parameters.

[0010] Furthermore, the final air concentration prediction model includes multiple historical period air concentration prediction modules and multiple future period air concentration prediction modules; The first historical cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the first group based on the production parameters of the first group, the air concentration parameters of the previous group, and the preliminary predicted air concentration parameters. The historical periodic air concentration prediction module, excluding the first one, is used to determine the final predicted air concentration parameters and status information of the current group based on the production parameters of the current group, the air concentration parameters of the previous group, the preliminary predicted air concentration parameters, the final predicted air concentration parameters of the previous group, and the status information.

[0011] Furthermore, the first future cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the current group based on the predicted production parameters of the current group, the air concentration parameters of the previous group, the preliminary predicted air concentration parameters, the final predicted air concentration parameters of the previous group, and the status information. The intermediate future cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the current group based on the predicted production parameters of the current group, the preliminary predicted air concentration parameters, and the final predicted air concentration parameters and status information of the previous group. The final future cycle air concentration prediction module is used to determine the final predicted air concentration parameters of the last group based on the predicted production parameters of the current group, the preliminary predicted air concentration parameters, and the final predicted air concentration parameters and status information of the previous group.

[0012] Furthermore, in the historical period air concentration prediction module or the future period air concentration prediction module, the final predicted air concentration parameters and status information for the current group are calculated using the following formula: ; in, Indicates the first The group's final predicted air concentration parameters, Indicates the first The group's final predicted air concentration parameters, Indicates the first Group status information, Indicates the first Group status information, Indicates the first Group forget gate information, Indicates the first Group input gate control information, Indicates the first The cell state of the group's input gate. Indicates the first Group output gate information; in, Indicates the first The weight matrix of the output gate of the group. Indicates the first The relevant weight matrix of the input gate of the group, Indicates the first The linear weight matrix of the input gate of the group. Indicates the first The weight matrix of the forget gate for the group; Indicates the first The bias vector of the output gate of the group. Indicates the first The relevant bias vector of the input gate of the group. Indicates the first The linear bias vector of the input gate of the group. Indicates the first The bias vector of the forget gate of the group; This represents the Sigmoid function. Indicates the activation function; in, In the historical periodic air concentration prediction module, the [number]th [period] is represented as [number]. Production parameters of the group, air concentration parameters of the previous group, and preliminary predicted air concentration parameters; In the first future cycle air concentration prediction module, it represents the... The predicted production parameters of the group, the air concentration parameters of the previous group, and the preliminary predicted air concentration parameters; Both the intermediate future cycle air concentration prediction module and the last future cycle air concentration prediction module indicate that the [number]th [period] is [the number]. The predicted production parameters, preliminary predicted air concentration parameters, and final predicted air concentration parameters of the previous group.

[0013] Furthermore, the number of groups in the historical period is greater than or equal to the number of groups in the future period.

[0014] Furthermore, the preliminary air concentration prediction model adopts the time series prediction model Prophet.

[0015] Furthermore, the preliminary air concentration prediction model, the production parameter prediction model, and the final air concentration prediction model are trained through the following steps: Based on the company's historical operating data, production parameters and air concentration parameters for each period are obtained to determine multiple training samples for the preliminary air concentration prediction model; each training sample of the preliminary air concentration prediction model includes time parameters and air concentration parameters. The preliminary air concentration prediction model is trained using multiple training samples to obtain the trained preliminary air concentration prediction model; based on the trained preliminary air concentration prediction model, the preliminary predicted air concentration parameters for each period are obtained. Multiple joint training samples are constructed based on the number of groups in the historical period and the number of groups in the future period to form a production parameter prediction model and an air concentration final prediction model. Each joint training sample of the production parameter prediction model and the air concentration final prediction model includes production parameters, air concentration parameters and preliminary predicted air concentration parameters for a sum of the number of groups in the historical period and the number of groups in the future period. The production parameter prediction model and the final air concentration prediction model are trained simultaneously using multiple joint training samples to obtain the trained production parameter prediction model and the final air concentration prediction model.

[0016] The present application discloses a method for predicting air concentration at enterprise wastewater discharge outlets. This method uses multiple sets of production parameters based on historical periods to obtain multiple sets of predicted production parameters for future periods using a production parameter prediction model. These predicted production parameters for future periods are then used as prior information and input into the final air concentration prediction model, thereby improving the accuracy of air concentration prediction at enterprise wastewater discharge outlets.

[0017] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A schematic flowchart of an air concentration prediction method for enterprise sewage outlets according to an embodiment of this application is shown. Figure 2 A schematic diagram of the structure of a production parameter prediction model according to an embodiment of this application is shown; Figure 3 A schematic diagram of the structure of a first data fusion module or a second data fusion module according to an embodiment of this application is shown; Figure 4 A schematic diagram of the structure of a first production parameter state prediction module or a second production parameter state prediction module according to an embodiment of this application is shown. Figure 5 A schematic diagram of the structure of the final air concentration prediction model according to an embodiment of this application is shown; Figure 6 A schematic diagram of the structure of a historical period air concentration prediction module or a future period air concentration prediction module according to an embodiment of this application is shown. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Figure 1 A flowchart illustrating an air concentration prediction method for enterprise wastewater outlets according to an embodiment of this application is shown, as follows: Figure 1 As shown, the air concentration prediction method includes: Step S101: Input multiple sets of production parameters from historical periods into the production parameter prediction model to obtain multiple sets of predicted production parameters for future periods; wherein, the production parameters include time parameters. Step S102: Input the time parameters of each group of historical and future periods into the preliminary air concentration prediction model to obtain the preliminary predicted air concentration parameters of each group. Step S103: Input multiple sets of production parameters from historical periods, the previous set of air concentration parameters, the preliminary predicted air concentration parameters, and multiple sets of production parameters and the preliminary predicted air concentration parameters from future periods into the final air concentration prediction model to obtain multiple sets of final predicted air concentration parameters for future periods.

[0022] Specifically, the period can be set according to the company's actual production needs, and the unit can be minutes, hours, days, weeks, months, or years. For example, with a period of days, the data for the next 5 days can be predicted using the historical data of the past 5 days in the company's operational data.

[0023] Specifically, production parameters include time parameters, as well as one or more of the following: production volume, power consumption, production material consumption, and power consumption of environmental protection equipment. Air concentration parameters include one or more of the following: sulfur oxide concentration, nitrogen oxide concentration, carbon oxide concentration, ozone concentration, volatile organic compound concentration, greenhouse gas concentration, and inorganic gas concentration. It can be understood that by using time parameters to represent the period to which the data belongs, the production parameter prediction model, the preliminary air concentration prediction model, and the final air concentration prediction model can capture patterns in the time dimension.

[0024] Specifically, each set of production parameters and air concentration parameters in the historical cycle can be obtained through the company's historical operating data, while each set of production parameters and air concentration parameters in the future cycle are unknown data and need to be predicted.

[0025] Specifically, each set of data in the historical period includes production parameters, the previous set of air concentration parameters, and the preliminary predicted air concentration parameters. The production parameters and the preliminary predicted air concentration parameters are the data corresponding to the current set.

[0026] Specifically, the production parameters and air concentration parameters in each set of data for future cycles are all predicted data for the current set.

[0027] Specifically, such as Figure 1 As shown, in step S101, multiple sets of production parameters for a historical period are obtained from the enterprise's historical operating data. For example, the historical period is from day 1 to day 5, and the period is in days. Each day includes one set of production parameters, so the multiple sets of production parameters for the historical period include five sets of production parameters from day 1 to day 5.

[0028] Specifically, the multiple sets of production parameters for the future period in step S101 represent the production parameters for multiple periods that need to be predicted. For example, if it is necessary to predict the air concentration for the next 5 days, and the future period is from the 6th to the 10th day, then 5 sets of production parameters for the next 5 days are predicted in step S101.

[0029] Specifically, in step S101, after inputting the five sets of production parameters from day 1 to day 5 into the production parameter prediction model, the production parameter prediction model outputs the predicted production parameters from day 6 to day 10.

[0030] Figure 2 A schematic diagram of the structure of a production parameter prediction model according to an embodiment of this application is shown.

[0031] Preferably, the production parameter prediction model includes multiple historical period production parameter prediction modules and multiple future period production parameter prediction modules connected in sequence. The first historical cycle production parameter prediction module is used to determine the predicted production parameters and status information for the next group based on the production parameters of the first group. The historical cycle production parameter prediction module, excluding the first one, is used to determine the predicted production parameters and status information of the next group based on the production parameters, predicted production parameters, and status information of the current group. The future cycle production parameter prediction module, excluding the last one, is used to determine the predicted production parameters and status information of the next group based on the predicted production parameters and status information of the current group. The last future cycle production parameter prediction module is used to determine the predicted production parameters for the next group based on the predicted production parameters and status information of the current group.

[0032] For example, such as Figure 2 As shown, the production parameter prediction model includes five historical period production parameter prediction modules and four future period production parameter prediction modules connected in sequence.

[0033] like Figure 2 As shown, the production parameters of group 1 The predicted production parameters and status information for the second group are obtained by inputting them into the production parameter prediction module for the first historical cycle. The production parameters for the second group are then... The predicted production parameters and status information are input into the production parameter prediction module for the second historical cycle to obtain the predicted production parameters and status information for the third group. The production parameters of the third group are then... The predicted production parameters and status information are input into the production parameter prediction module of the third historical cycle to obtain the predicted production parameters and status information of the fourth group. The production parameters of the fourth group are then... The predicted production parameters and status information are input into the production parameter prediction module of the 4th historical cycle to obtain the predicted production parameters and status information of the 5th group. The production parameters of the 5th group are then... The predicted production parameters and status information are input into the production parameter prediction module of the 5th historical cycle to obtain the predicted production parameters of the 6th group. Status information .

[0034] like Figure 2 As shown, the predicted production parameters for group 6 are... and status information The predicted production parameters for the 7th group are obtained by inputting them into the production parameter prediction module for the first future cycle. and status information The predicted production parameters of Group 7 and status information The predicted production parameters for the 8th group are obtained by inputting them into the production parameter prediction module for the 2nd future cycle. and status information The predicted production parameters of Group 8 and status information The predicted production parameters for the 9th set are obtained by inputting them into the production parameter prediction module for the 3rd future cycle. and status information The predicted production parameters of Group 9 and status information The predicted production parameters for the 10th set are obtained by inputting them into the production parameter prediction module for the 4th future cycle. .

[0035] Preferably, such as Figure 2 As shown, the historical cycle production parameter prediction module includes a first data fusion module and a first production parameter status prediction module; the future cycle production parameter prediction module includes a second data fusion module and a second production parameter status prediction module. The first data fusion module is used to weight and fuse the production parameters of the current group based on the attention mechanism to obtain the fused feature vector of the current group; The first production parameter state prediction module of the first historical cycle production parameter prediction module is used to predict the fused feature vector of the first group based on the Long Short-Term Memory Network (LSTM) to obtain the predicted production parameters and state information of the next group. The first production parameter state prediction module of the historical cycle production parameter prediction module, excluding the first one, is used to predict the fused feature vector, predicted production parameters and state information of the current group based on LSTM, so as to obtain the predicted production parameters and state information of the next group. The second data fusion module is used to perform weighted fusion of the predicted production parameters of the current group based on the attention mechanism to obtain the fusion feature vector of the current group. The second production parameter state prediction module of the last future cycle production parameter prediction module is used to predict the fused feature vector, predicted production parameters and state information of the current group based on LSTM to obtain the predicted production parameters of the next group. The second production parameter state prediction module, excluding the last one, is used to predict the fused feature vector, predicted production parameters, and state information of the current group based on LSTM, so as to obtain the predicted production parameters and state information of the next group.

[0036] For example, such as Figure 2 As shown, each historical cycle production parameter prediction module includes a first data fusion module and a first production parameter status prediction module, and each future cycle production parameter prediction module includes a second data fusion module and a second production parameter status prediction module.

[0037] Specifically, in the first data fusion module, the production parameters of the current group are weighted and fused using an attention-based mechanism to obtain the fused feature vector of the current group; in the second data fusion module, the predicted production parameters of the current group are weighted and fused using an attention-based mechanism to obtain the fused feature vector of the current group.

[0038] like Figure 2 As shown, the production parameters for group 1 The weighted fusion is performed to obtain the fusion feature vector of the first group. For the production parameters of Group 2 The weighted fusion is performed to obtain the fusion feature vector of the second group. For the production parameters of Group 3 The weighted fusion is performed to obtain the fusion feature vector of the third group. For the production parameters of Group 4 The weighted fusion is performed to obtain the fusion feature vector of the 4th group. For the production parameters of Group 5 The weighted fusion is performed to obtain the fusion feature vector of the 5th group. ; Forecast production parameters for Group 6 The weighted fusion is performed to obtain the fusion feature vector of the 6th group. For the predicted production parameters of Group 7 The weighted fusion is performed to obtain the fusion feature vector of the 7th group. For the predicted production parameters of Group 8 The weighted fusion is performed to obtain the fusion feature vector of the 8th group. For the predicted production parameters of Group 9 The weighted fusion is performed to obtain the fusion feature vector of the 9th group. .

[0039] Preferably, the fused feature vector of the current group is calculated in the first data fusion module or the second data fusion module using the following formula: ; in, Indicates the first The fusion feature vector of the group , , They represent the first The initial weight matrix, key weight matrix, and value weight matrix of the group. Represents a constant. This indicates the normalization indicator function; in, In the first data fusion module, the first The production parameters of the group are represented in the second data fusion module. The group's predicted production parameters.

[0040] Figure 3 A schematic diagram of the structure of a first data fusion module or a second data fusion module according to an embodiment of this application is shown.

[0041] like Figure 3 As shown, when representing the first data fusion module, Indicates the first The production parameters of the group, when representing the second data fusion module, Indicates the first The group's predicted production parameters.

[0042] For example, such as Figure 2 As shown, the data input into the first data fusion module is (i=1,2,3,4,5), including time parameters Production volume ,power consumption Production material consumption Power consumption of environmental protection devices ,Right now The data input into the second data fusion module is (i=6, 7, 8, 9), including time parameters Forecasted production volume Predicted power consumption Predicting the consumption of production materials Predicting the power consumption of environmental protection devices .

[0043] For example, in the first data fusion module or the second data fusion module, the initial weight matrix Key weight matrix Value weight matrix This is the core of the self-attention mechanism. Construct the initial weight matrix. Combined with production parameters or predicted production parameters, to obtain Similarly, construct the key weight matrix. Value weight matrix ,get and .

[0044] Specifically, constants This represents the feature dimension, which is 1. It is used to scale the attention score and avoid numerical overflow. The normalized exponential function "compresses" a multidimensional vector containing arbitrary real numbers into another multidimensional real vector, such that each element is in the range (0,1) and the sum of all elements is 1, thus constructing an effective probability distribution.

[0045] The fused feature vectors of groups 1 to 9 are obtained through the first data fusion module or the second data fusion module.

[0046] Specifically, such as Figure 2 As shown, both the first and second production parameter state prediction modules are based on the Long Short-Term Memory (LSTM) network to fuse feature vectors. Make predictions to obtain the predicted production parameters for the next set. and status information.

[0047] Figure 4 A schematic diagram of the structure of a first production parameter state prediction module or a second production parameter state prediction module according to an embodiment of this application is shown.

[0048] Specifically, such as Figure 4 As shown, in the first production parameter status prediction module of the first historical cycle production parameter prediction module, there is no predicted production parameter and status information of the previous group. The zero vector is used to participate, and the fusion feature vector of the first group is predicted based on LSTM to obtain the predicted production parameter and status information of the second group.

[0049] In the first production parameter status prediction module of the second historical cycle production parameter prediction module, the fused feature vector of the second group, the predicted production parameters and status information of the second group are predicted based on LSTM to obtain the predicted production parameters and status information of the third group; similarly, the predicted production parameters and status information of the fourth group and the predicted production parameters and status information of the fifth group are obtained.

[0050] Specifically, such as Figure 4 As shown, in the second production parameter state prediction module of the production parameter prediction module for the last future cycle, there is no need to output the state information of the next group. It is only necessary to predict the fused feature vector, predicted production parameters and state information of the 9th group based on LSTM to obtain the predicted production parameters of the 10th group.

[0051] In the second production parameter state prediction module of the future cycle production parameter prediction module (excluding the last one), the fused feature vector, predicted production parameters and state information of the current group are predicted based on LSTM to obtain the predicted production parameters and state information of the next group.

[0052] like Figure 4 As shown, exemplarily, For the first The group's state information is initialized to a zero vector at the start time (i.e., t=0). For the first The predicted production parameters of the group, i.e., the predicted first... Group production volume, power consumption, production material consumption, and power consumption of environmental protection equipment. , For the first The fusion feature vector of the group; similarly, For the first Group status information, For the first The group's predicted production parameters.

[0053] In either the first or second production parameter status prediction module, the predicted production parameters and status information for the next group are calculated using the following formula: ; in, To output gate information, Here is the weight matrix of the output gate. This is the bias vector for the output gate. The cell state for the input gate. The relevant weight matrix of the input gate, This is the relevant bias vector of the input gate. For inputting door information, Let be the linear weight matrix of the input gate. Let be the linear bias vector of the input gate. Forgotten Gate information, Here is the weight matrix for the forget gate. This is the bias vector for the forget gate.

[0054] Understandably, in the production parameter prediction model, a self-attention mechanism is used to achieve dynamic weight allocation of production parameters, explore multi-feature correlations, and introduce an iterative prediction mechanism to dynamically compensate for future missing data using previous prediction values.

[0055] It is worth noting that by using historical production data from the past to predict future production data through a production parameter prediction model, and then using this data as prior information input into the final air concentration prediction model, the prediction accuracy of air concentration under complex operating conditions can be improved.

[0056] Specifically, such as Figure 1 As shown, in step S102, the time parameters of each group of historical and future periods are input into the preliminary air concentration prediction model to obtain the preliminary predicted air concentration parameters for each group. For example, inputting the time parameters of 5 historical periods and 5 future periods into the preliminary air concentration prediction model yields 10 preliminary predicted air concentration parameters, corresponding to 10 periods.

[0057] Preferably, the preliminary air concentration prediction model adopts the time series prediction model Prophet.

[0058] Specifically, Prophet is used as a static feature extractor to predict the corresponding preliminary air concentration parameter for each time parameter.

[0059] Specifically, such as Figure 1 As shown, in step S103, the multiple sets of predicted production parameters for the future period obtained in step S101, the multiple sets of preliminary predicted air concentration parameters for the historical period obtained in step S102, the multiple sets of preliminary predicted air concentration parameters for the future period, the multiple sets of production parameters for the historical period, and the previous set of air concentration parameters are input into the final air concentration prediction model to obtain the multiple sets of final predicted air concentration parameters for the future period.

[0060] For example, five sets of data from the historical period, including production parameters, the previous set of air concentration parameters, and the preliminary predicted air concentration parameters, as well as five sets of data from the future period, including predicted production parameters and the preliminary predicted air concentration parameters, are all input into the final air concentration prediction model to obtain five sets of final predicted air concentration parameters for the future period.

[0061] Preferably, the final air concentration prediction model includes multiple historical period air concentration prediction modules and multiple future period air concentration prediction modules; The first historical cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the first group based on the production parameters of the first group, the air concentration parameters of the previous group, and the preliminary predicted air concentration parameters. The historical periodic air concentration prediction module, excluding the first one, is used to determine the final predicted air concentration parameters and status information of the current group based on the production parameters of the current group, the air concentration parameters of the previous group, the preliminary predicted air concentration parameters, the final predicted air concentration parameters of the previous group, and the status information.

[0062] Preferably, the first future cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the current group based on the predicted production parameters of the current group, the air concentration parameters of the previous group, the preliminary predicted air concentration parameters, the final predicted air concentration parameters of the previous group, and the status information. The intermediate future cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the current group based on the predicted production parameters of the current group, the preliminary predicted air concentration parameters, and the final predicted air concentration parameters and status information of the previous group. The final future cycle air concentration prediction module is used to determine the final predicted air concentration parameters of the last group based on the predicted production parameters of the current group, the preliminary predicted air concentration parameters, and the final predicted air concentration parameters and status information of the previous group.

[0063] Figure 5 A schematic diagram of the structure of the final air concentration prediction model according to an embodiment of this application is shown.

[0064] like Figure 5 As shown, the final air concentration prediction model includes 5 historical period air concentration prediction modules and 5 future period air concentration prediction modules. The 5 future period air concentration prediction modules are used to obtain 5 sets of final predicted air concentration parameters.

[0065] Specifically, in the air concentration prediction module of the first historical cycle, the final predicted air concentration parameters and status information of the first group are determined based on the production parameters of the first group, the air concentration parameters of the 0th group, and the preliminary predicted air concentration parameters of the first group.

[0066] In the air concentration prediction module for the second historical cycle, the final predicted air concentration parameters and status information for the second group are determined based on the production parameters of the second group, the air concentration parameters of the first group, the preliminary predicted air concentration parameters of the first group, and the final predicted air concentration parameters and status information of the first group. Similarly, the final predicted air concentration parameters and status information for the third, fourth, and fifth groups are obtained.

[0067] Specifically, in the air concentration prediction module for the first future cycle, the final predicted air concentration parameters and status information for the sixth group are determined based on the predicted production parameters of the sixth group, the air concentration parameters of the fifth group, the preliminary predicted air concentration parameters of the sixth group, and the final predicted air concentration parameters and status information of the fifth group.

[0068] In the second future cycle air concentration prediction module, the final predicted air concentration parameters and status information of the 7th group are determined based on the predicted production parameters of the 7th group, the preliminary predicted air concentration parameters of the 7th group, the final predicted air concentration parameters of the 6th group, and the status information; similarly, the final predicted air concentration parameters and status information of the 8th, 9th, and 10th groups are obtained.

[0069] like Figure 5 As shown, ( i =1,2,3,4,5), from composition, Including the i Group production volume ,power consumption Production material consumption Power consumption of environmental protection devices and the i -1 group of air concentration parameters , Indicates the first i The time parameter of the group, the time parameter The first one is obtained through Prophet. i Preliminary predicted air concentration parameters of the group .

[0070] For example, ,Depend on Composition; when i When =6, Including the i -1 group of air concentration parameters and the i Group's projected production volume Predicted power consumption Predicting the consumption of production materials Predicting the power consumption of environmental protection devices ;when i =7, 8, 9, 10 Including the i Group's projected production volume Predicted power consumption Predicting the consumption of production materials Predicting the power consumption of environmental protection devices and the i-1 The group's final predicted air concentration parameters.

[0071] in, Indicates the first i The group's final predicted air concentration parameters.

[0072] Figure 6A schematic diagram of the structure of a historical period air concentration prediction module or a future period air concentration prediction module according to an embodiment of this application is shown.

[0073] Preferably, such as Figure 6 As shown, the final predicted air concentration parameters and status information for the current group are calculated using the following formula in the historical period air concentration prediction module or the future period air concentration prediction module: ; in, Indicates the first The group's final predicted air concentration parameters, Indicates the first The group's final predicted air concentration parameters, Indicates the first Group status information, Indicates the first Group status information, Indicates the first Group forget gate information, Indicates the first Group input gate control information, Indicates the first The cell state of the group's input gate. Indicates the first Group output gate information; in, Indicates the first The weight matrix of the output gate of the group. Indicates the first The relevant weight matrix of the input gate of the group, Indicates the first The linear weight matrix of the input gate of the group. Indicates the first The weight matrix of the forget gate for the group; Indicates the first The bias vector of the output gate of the group. Indicates the first The relevant bias vector of the input gate of the group. Indicates the first The linear bias vector of the input gate of the group. Indicates the first The bias vector of the forget gate of the group; This represents the Sigmoid function. Indicates the activation function; in, In the historical periodic air concentration prediction module, the [number]th [period] is represented as [number]. Production parameters of the group, air concentration parameters of the previous group, and preliminary predicted air concentration parameters; In the first future cycle air concentration prediction module, it represents the... The predicted production parameters of the group, the air concentration parameters of the previous group, and the preliminary predicted air concentration parameters; Both the intermediate future cycle air concentration prediction module and the last future cycle air concentration prediction module indicate that the [number]th [period] is [the number]. The predicted production parameters, preliminary predicted air concentration parameters, and final predicted air concentration parameters of the previous group.

[0074] Preferably, the number of groups in the historical period is greater than or equal to the number of groups in the future period.

[0075] For example, using five historical data sets, predict five or fewer future data sets.

[0076] Preferably, the preliminary air concentration prediction model, the production parameter prediction model, and the final air concentration prediction model are trained through the following steps: Based on the company's historical operating data, production parameters and air concentration parameters for each period are obtained to determine multiple training samples for the preliminary air concentration prediction model; each training sample of the preliminary air concentration prediction model includes time parameters and air concentration parameters. The preliminary air concentration prediction model is trained using multiple training samples to obtain the trained preliminary air concentration prediction model; based on the trained preliminary air concentration prediction model, the preliminary predicted air concentration parameters for each period are obtained. Multiple joint training samples are constructed based on the number of groups in the historical period and the number of groups in the future period to form a production parameter prediction model and an air concentration final prediction model. Each joint training sample of the production parameter prediction model and the air concentration final prediction model includes production parameters, air concentration parameters and preliminary predicted air concentration parameters for a sum of the number of groups in the historical period and the number of groups in the future period. The production parameter prediction model and the final air concentration prediction model are trained simultaneously using multiple joint training samples to obtain the trained production parameter prediction model and the final air concentration prediction model.

[0077] Specifically, when training the preliminary air concentration prediction model, the production parameter prediction model, and the final air concentration prediction model, the preliminary air concentration prediction model is trained first, and the preliminary air concentration prediction model is trained independently, while the production parameter prediction model and the final air concentration prediction model are trained jointly.

[0078] When training the preliminary air concentration prediction model, each training sample includes a time parameter and an air concentration parameter. Supervised learning training is performed with the air concentration parameter as the target variable, and the time parameter is input into the preliminary air concentration prediction model as a time explanatory variable. The supervised learning objective is constructed with the air concentration parameter, and the parameter training and validation of the preliminary air concentration prediction model are completed through maximum likelihood estimation.

[0079] After training the initial air concentration prediction model, the production parameter prediction model and the final air concentration prediction model are then jointly trained.

[0080] For example, data from the past 5 years is collected, including date t, production volume a, power consumption b, production material consumption c, power consumption of environmental protection devices d, and air concentration u. The preliminarily predicted air concentration for each time t is obtained using a trained preliminary air concentration prediction model.

[0081] Assuming that the prediction of the next 5 days' data is based on the historical data of the past 5 days, each joint training sample includes data for 10 days. The data for each day includes date t, production volume a, power consumption b, production material consumption c, power consumption of environmental protection equipment d, air concentration u, and preliminary predicted air concentration.

[0082] When jointly training the production parameter prediction model and the final air concentration prediction model, the preliminary air concentration prediction model serves as a static feature extractor, and its model parameters remain fixed during training. The model parameters of the production parameter prediction model and the final air concentration prediction model are updated synchronously through the backpropagation algorithm.

[0083] For example, when training the preliminary air concentration prediction model, the production parameter prediction model, and the final air concentration prediction model, the cross-entropy loss function can be used to calculate the loss value for each training session.

[0084] Specifically, such as Figure 1 As shown, the multiple sets of final predicted air concentration parameters for the future period obtained in step S103 are used as the final air concentration values.

[0085] It is worth noting that by integrating the output data of the production parameter prediction model and the preliminary air concentration prediction model into the final air concentration prediction model, a dynamic correlation of multiple variables is constructed, which further improves the prediction accuracy of air concentration.

[0086] The present application provides a method for predicting air concentration at enterprise sewage outlets. This method uses a production parameter prediction model to obtain multiple sets of predicted production parameters for future periods based on multiple sets of production parameters from historical periods. These predicted production parameters for future periods are then used as prior information and input into the final air concentration prediction model, thereby improving the accuracy of air concentration prediction at enterprise sewage outlets.

[0087] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for predicting air concentration at enterprise wastewater discharge outlets, characterized in that, The air concentration prediction method includes: By inputting multiple sets of production parameters from historical periods into a production parameter prediction model, multiple sets of predicted production parameters for future periods are obtained; among these, production parameters include time parameters. The time parameters of each group in the historical period and the future period are input into the preliminary air concentration prediction model to obtain the preliminary predicted air concentration parameters of each group. By inputting multiple sets of production parameters from historical periods, the previous set of air concentration parameters, the preliminary predicted air concentration parameters, and multiple sets of predicted production parameters and preliminary predicted air concentration parameters from future periods into the final air concentration prediction model, multiple sets of final predicted air concentration parameters for future periods are obtained.

2. The air concentration prediction method according to claim 1, characterized in that, The production parameter prediction model includes multiple historical period production parameter prediction modules and multiple future period production parameter prediction modules connected in sequence. The first historical cycle production parameter prediction module is used to determine the predicted production parameters and status information for the next group based on the production parameters of the first group. The historical cycle production parameter prediction module, excluding the first one, is used to determine the predicted production parameters and status information of the next group based on the production parameters, predicted production parameters, and status information of the current group. The future cycle production parameter prediction module, excluding the last one, is used to determine the predicted production parameters and status information of the next group based on the predicted production parameters and status information of the current group. The last future cycle production parameter prediction module is used to determine the predicted production parameters for the next group based on the predicted production parameters and status information of the current group.

3. The air concentration prediction method according to claim 2, characterized in that, The historical cycle production parameter prediction module includes a first data fusion module and a first production parameter status prediction module; the future cycle production parameter prediction module includes a second data fusion module and a second production parameter status prediction module. The first data fusion module is used to weight and fuse the production parameters of the current group based on the attention mechanism to obtain the fused feature vector of the current group; The first production parameter state prediction module of the first historical cycle production parameter prediction module is used to predict the fused feature vector of the first group based on the Long Short-Term Memory Network (LSTM) to obtain the predicted production parameters and state information of the next group. The first production parameter state prediction module of the historical cycle production parameter prediction module, excluding the first one, is used to predict the fused feature vector, predicted production parameters and state information of the current group based on LSTM, so as to obtain the predicted production parameters and state information of the next group. The second data fusion module is used to perform weighted fusion of the predicted production parameters of the current group based on the attention mechanism to obtain the fusion feature vector of the current group. The second production parameter state prediction module of the last future cycle production parameter prediction module is used to predict the fused feature vector, predicted production parameters and state information of the current group based on LSTM to obtain the predicted production parameters of the next group. The second production parameter state prediction module, excluding the last one, is used to predict the fused feature vector, predicted production parameters, and state information of the current group based on LSTM, so as to obtain the predicted production parameters and state information of the next group.

4. The air concentration prediction method according to claim 3, characterized in that, The fused feature vector of the current group is calculated in the first or second data fusion module using the following formula: ； in, Indicates the first The fusion feature vector of the group , , They represent the first The initial weight matrix, key weight matrix, and value weight matrix of the group. Represents a constant. This indicates the normalization indicator function; in, In the first data fusion module, the first The production parameters of the group are represented in the second data fusion module. The group's predicted production parameters.

5. The air concentration prediction method according to claim 1, characterized in that, The final air concentration prediction model includes multiple historical period air concentration prediction modules and multiple future period air concentration prediction modules. The first historical cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the first group based on the production parameters of the first group, the air concentration parameters of the previous group, and the preliminary predicted air concentration parameters. The historical periodic air concentration prediction module, excluding the first one, is used to determine the final predicted air concentration parameters and status information of the current group based on the production parameters of the current group, the air concentration parameters of the previous group, the preliminary predicted air concentration parameters, the final predicted air concentration parameters of the previous group, and the status information.

6. The air concentration prediction method according to claim 5, characterized in that, The first future cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the current group based on the predicted production parameters of the current group, the air concentration parameters of the previous group, the preliminary predicted air concentration parameters, the final predicted air concentration parameters of the previous group, and the status information. The intermediate future cycle air concentration prediction module is used to determine the final predicted air concentration parameters and status information of the current group based on the predicted production parameters of the current group, the preliminary predicted air concentration parameters, and the final predicted air concentration parameters and status information of the previous group. The final future cycle air concentration prediction module is used to determine the final predicted air concentration parameters of the last group based on the predicted production parameters of the current group, the preliminary predicted air concentration parameters, and the final predicted air concentration parameters and status information of the previous group.

7. The air concentration prediction method according to claim 6, characterized in that, In the historical period air concentration prediction module or the future period air concentration prediction module, the final predicted air concentration parameters and status information for the current group are calculated using the following formula: ； in, Indicates the first The group's final predicted air concentration parameters, Indicates the first The group's final predicted air concentration parameters, Indicates the first Group status information, Indicates the first Group status information, Indicates the first Group forget gate information, Indicates the first Group input gate control information, Indicates the first The cell state of the group's input gate. Indicates the first Group output gate information; in, Indicates the first The weight matrix of the output gate of the group. Indicates the first The relevant weight matrix of the input gate of the group, Indicates the first The linear weight matrix of the input gate of the group. Indicates the first The weight matrix of the forget gate for the group; Indicates the first The bias vector of the output gate of the group. Indicates the first The relevant bias vector of the input gate of the group. Indicates the first The linear bias vector of the input gate of the group. Indicates the first The bias vector of the forget gate of the group; This represents the Sigmoid function. Indicates the activation function; in, In the historical periodic air concentration prediction module, the [number]th [period] is represented as [number]. Production parameters of the group, air concentration parameters of the previous group, and preliminary predicted air concentration parameters; In the first future cycle air concentration prediction module, it represents the... The predicted production parameters of the group, the air concentration parameters of the previous group, and the preliminary predicted air concentration parameters; Both the intermediate future cycle air concentration prediction module and the last future cycle air concentration prediction module indicate that the [number]th [period] is [the number]. The predicted production parameters, preliminary predicted air concentration parameters, and final predicted air concentration parameters of the previous group.

8. The air concentration prediction method according to any one of claims 1-7, characterized in that, The number of groups in the historical period is greater than or equal to the number of groups in the future period.

9. The air concentration prediction method according to any one of claims 1-7, characterized in that, The preliminary air concentration prediction model uses the Prophet time series prediction model.

10. The air concentration prediction method according to any one of claims 1-7, characterized in that, The preliminary air concentration prediction model, the production parameter prediction model, and the final air concentration prediction model are trained using the following steps: Based on the company's historical operating data, production parameters and air concentration parameters for each period are obtained to determine multiple training samples for the preliminary air concentration prediction model; each training sample of the preliminary air concentration prediction model includes time parameters and air concentration parameters. The preliminary air concentration prediction model is trained using multiple training samples to obtain the trained preliminary air concentration prediction model; based on the trained preliminary air concentration prediction model, the preliminary predicted air concentration parameters for each period are obtained. Multiple joint training samples are constructed based on the number of groups in the historical period and the number of groups in the future period to form a production parameter prediction model and an air concentration final prediction model. Each joint training sample of the production parameter prediction model and the air concentration final prediction model includes production parameters, air concentration parameters and preliminary predicted air concentration parameters for a sum of the number of groups in the historical period and the number of groups in the future period. The production parameter prediction model and the final air concentration prediction model are trained simultaneously using multiple joint training samples to obtain the trained production parameter prediction model and the final air concentration prediction model.

Images