Environmental protection data intelligent monitoring system

Abstract

The invention provides an environmental protection data intelligent monitoring system, which comprises a data acquisition module, a data transmission module and a server unit, and is characterized in that the server unit is configured to execute the following operations: receiving and fusing data of at least two heterogeneous data sources; based on an association relationship between the historical environmental protection monitoring data and the historical production operation data, establishing a dynamic reference model, the dynamic reference model being used for dynamically calculating a reference prediction value of the corresponding environmental protection monitoring data in the current production operation state according to the real-time production operation data; comparing the newly received real-time environmental protection monitoring data with a reference predicted value calculated by the dynamic reference model according to the real-time production operation data at the same moment to generate a real-time deviation value; and generating predictive risk early warning based on the change trend of the real-time deviation value or a comparison result of the real-time deviation value and a preset deviation threshold value. The production data and the environmental protection data can be fused, and the environmental risk can be intelligently predicted.

Description

Technical Field

[0001] This invention relates to an enterprise monitoring system, specifically an environmental data intelligent monitoring system, belonging to the field of environmental data monitoring technology. Background Technology

[0002] With increasingly stringent environmental regulations and heightened corporate social responsibility awareness, effective monitoring of environmental data such as waste gas, wastewater, and noise from industrial production processes has become a necessity for compliant operations. Traditional environmental monitoring systems typically employ static threshold alarm mechanisms, triggering an alarm when pollutant concentrations exceed national or local legal limits. However, this traditional monitoring model is reactive and lacks foresight, only issuing alerts when environmental indicators have already exceeded or are about to exceed limits. By this time, substantial environmental impacts may have already occurred, or the window for corrective action may be extremely short, leading to a passive response and potentially even administrative penalties. A high emission value may be within the normal fluctuation range when the production line is running at full capacity; however, when the production line is deloaded or shut down, the same emission value may indicate serious equipment failure or management oversight. Traditional monitoring systems cannot distinguish between these two situations, making it impossible to identify abnormal states inconsistent with production conditions in their early stages. Summary of the Invention

[0003] Based on the above background, the purpose of this invention is to provide an intelligent environmental data monitoring system that can integrate production data and environmental data, intelligently predict environmental risks, realize proactive and forward-looking environmental management, and overcome the problems of passive response and detachment from operating conditions in existing environmental monitoring systems, which can easily lead to misjudgments.

[0004] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0005] An environmental protection data intelligent monitoring system includes a data acquisition module, a data transmission module, and a server unit, wherein the server unit is configured to perform the following operations:

[0006] The system receives and integrates data from at least two heterogeneous data sources, including real-time environmental monitoring data from the data acquisition module and real-time production and operation data from an external enterprise information system.

[0007] Based on the correlation between historical environmental monitoring data and historical production and operation data, a dynamic benchmark model is established. The dynamic benchmark model is used to dynamically calculate the benchmark predicted value of environmental monitoring data corresponding to the current production and operation status based on real-time production and operation data.

[0008] For newly received real-time environmental monitoring data, compare it with the benchmark prediction value calculated by the dynamic benchmark model based on the real-time production and operation data at the same time to generate a real-time deviation value;

[0009] Based on the changing trend of the real-time deviation value or the comparison result with the preset deviation threshold, a predictive risk warning is generated before the real-time environmental monitoring data reaches the statutory alarm threshold.

[0010] Preferably, the server unit is further configured to construct the real-time production operation data into a production state vector, the dimensions of which include equipment operating power for characterizing production load, production line start-up and shutdown status for characterizing production cycle time, and energy medium usage flow rate for characterizing material balance; the real-time environmental monitoring data includes the concentration of exhaust pollutants that have a potential causal relationship with the production state vector.

[0011] Preferably, the dynamic benchmark model is a hybrid spatiotemporal attention neural network model, the architecture of which includes:

[0012] Long Short-Term Memory network layer used to process the time series data of each dimension in the production state vector;

[0013] A one-dimensional convolutional neural network layer configured in parallel with the long short-term memory network layer and used to extract local combined features of the production state vector; and,

[0014] The attention mechanism layer receives the outputs of the long short-term memory network layer and the one-dimensional convolutional neural network layer. It adopts a scaled dot product attention mechanism, learns the dynamic weights of the impact of different production and operation data dimensions on environmental monitoring data at different time points, and performs weighted fusion on the outputs of the long short-term memory network layer and the one-dimensional convolutional neural network layer to finally output the baseline prediction value.

[0015] Preferably, the server unit is also configured to generate a normalized deviation index as the real-time deviation value; the dynamic benchmark model outputs a prediction confidence interval and a prediction standard deviation while outputting the benchmark prediction value, and the normalized deviation index is obtained by calculating the ratio of the difference between the real-time environmental monitoring data and the benchmark prediction value to the prediction standard deviation.

[0016] Preferably, the server unit is further configured to perform first and second derivative calculations on the normalized deviation index within a preset sliding time window to obtain the rate of change and acceleration of the normalized deviation index; and when the rate of change continuously exceeds a preset rate threshold and the acceleration is positive, the predictive risk warning is generated.

[0017] Preferably, the intelligent environmental data monitoring system also includes a user terminal, and the server unit is further configured to provide a visual early warning interface to the user terminal. The visual early warning interface is used to display the predicted trajectory safety cone. The upper and lower boundaries of the predicted trajectory safety cone are dynamically formed by the prediction confidence interval output by the dynamic benchmark model. The visual early warning is also used to extrapolate and predict the future short-term trajectory of the real-time environmental monitoring data according to its current trend, and to highlight the point where the predicted trajectory will cross the safety cone.

[0018] Preferably, the server unit is also configured to perform optimization operations based on model performance degradation. The optimization operations include: continuously monitoring the long-term average and variance of the normalized bias index, and triggering a retraining process for the dynamic benchmark model when the long-term average is detected to have a persistent drift or the variance has increased significantly and exceeded a preset model confidence range.

[0019] Preferably, the server unit is further configured to perform time-series data alignment and feature engineering operations before establishing a dynamic benchmark model. The time-series data alignment and feature engineering operations include: resampling the environmental monitoring data and the production and operation data with different sampling frequencies to a unified time benchmark through interpolation or aggregation; creating multiple time-lag features for the production and operation data at each time point to use historical production and operation status as one of the input features at the current moment, so that the dynamic benchmark model can learn the delayed effect between production activities and environmental impacts.

[0020] Compared with the prior art, the present invention has the following advantages:

[0021] The present invention provides an intelligent monitoring system for environmental data. By constructing a dynamic benchmark model based on production conditions, the monitoring logic is changed from judging whether the standard is exceeded to judging whether there is an anomaly. Combined with deviation trend analysis, it can issue risk warnings in advance when environmental indicators are far from reaching the legal limits, giving enterprises time to respond.

[0022] This invention deeply integrates isolated environmental data with production and operation data, improving the accuracy of early warning judgments and providing a data foundation for subsequent risk tracing and decision-making.

[0023] The hybrid spatiotemporal attention neural network model used in this invention can comprehensively capture the long-term trends, local patterns and key driving factors of data, better fit complex nonlinear relationships, and improve the accuracy of dynamic benchmark prediction.

[0024] The optimization mechanism based on model performance degradation used in this invention improves the system's adaptability to gradual changes in production processes. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0026] Figure 1 This is a logical architecture diagram of an environmental data intelligent monitoring system according to the present invention;

[0027] Figure 2 This is a schematic diagram of the principle of an intelligent environmental data monitoring system according to the present invention;

[0028] Figure 3 This is a structural diagram of the hybrid spatiotemporal attention neural network model in this invention;

[0029] Figure 4 This is a diagram showing the interface of the application layer in this invention;

[0030] Figure 5 This is a data display diagram of the application layer in this invention;

[0031] Figure 6 This is another data display diagram of the application layer in this invention. Detailed Implementation

[0032] The technical solution of the present invention will be further described in detail below through specific embodiments and in conjunction with the accompanying drawings. It should be understood that the implementation of the present invention is not limited to the following embodiments, and any modifications and / or alterations made to the present invention will fall within the protection scope of the present invention.

[0033] In this invention, unless otherwise specified, all parts and percentages are by weight, and the equipment and raw materials used are commercially available or commonly used in the art. Unless otherwise specified, the methods in the following embodiments are conventional methods in the art. Unless otherwise specified, the components or equipment in the following embodiments are general standard parts or components known to those skilled in the art, and their structures and principles can be learned by those skilled in the art through technical manuals or conventional experimental methods.

[0034] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings. In this detailed description, numerous specific details are set forth to facilitate explanation and provide a thorough understanding of the embodiments of the present invention. However, one or more embodiments may be practiced by those skilled in the art without these specific details.

[0035] An embodiment of the present invention discloses an intelligent monitoring system for environmental protection data, such as... Figure 1 and Figure 2 As shown, the logical architecture of this system includes a data source layer, a server unit, and an application layer.

[0036] The data source layer includes at least two heterogeneous data sources. One is the environmental monitoring data source, collected by data acquisition modules deployed on-site within the enterprise. Examples include continuous emission monitoring systems installed on chimneys to collect air pollutant concentrations, online water quality analyzers installed at discharge outlets to collect wastewater pollutant indicators, and noise sensors at the plant boundary. These sensors transmit real-time monitoring data to the server unit via data transmission modules. The other is the production and operation data source, originating from external enterprise information systems such as Manufacturing Execution Systems (MES), Enterprise Resource Planning (ERP), and Distributed Control Systems (DCS). The server unit obtains real-time production and operation data from these systems via API interfaces or direct database connections.

[0037] The server unit is deployed in the cloud or in the enterprise's on-premises data center. The server unit is configured to perform the following operations:

[0038] It receives and integrates data from at least two heterogeneous data sources, including real-time environmental monitoring data from the data acquisition module and real-time production and operation data from external enterprise information systems.

[0039] Based on the correlation between historical environmental monitoring data and historical production and operation data, a dynamic benchmark model is established. The dynamic benchmark model is used to dynamically calculate the benchmark predicted value of environmental monitoring data corresponding to the current production and operation status based on real-time production and operation data.

[0040] For newly received real-time environmental monitoring data, compare it with the benchmark prediction value calculated by the dynamic benchmark model based on the real-time production and operation data at the same time to generate a real-time deviation value;

[0041] Based on the changing trend of real-time deviation values ​​or the comparison results with preset deviation thresholds, predictive risk warnings are generated before real-time environmental monitoring data reach the statutory alarm threshold.

[0042] The application layer refers to the user terminal, which can be the terminal PC in the monitoring center, the engineer's laptop, or the mobile device of the administrator. For example... Figure 4-6 As shown, users can access the services provided by the server unit through a browser or a dedicated client to monitor environmental data, receive early warning information, and perform related management operations.

[0043] The system's workflow is divided into two stages: an offline model training stage and an online real-time monitoring stage.

[0044] The purpose of the offline model training phase is to establish a dynamic benchmark model.

[0045] The server unit first extracts environmental monitoring data and production operation data spanning a long period from the historical database.

[0046] Because different data are sampled at different frequencies—for example, exhaust gas data is recorded every five minutes, while equipment power data is recorded every second—this system uses resampling technology to align them to a unified time base. For low-frequency data, linear interpolation is used to complete the data; for high-frequency data, mean or maximum value aggregation is used.

[0047] The aligned production and operation data, such as equipment operating power, production line start-up and shutdown status, and energy medium usage flow rate, are used to construct a multi-dimensional production state vector X(t).

[0048] The environmental impact of production activities often has a delay. To capture this effect, the system additionally appends production state vectors from several past time points (e.g., t-1, t-5, t-10 minutes) to the vector X(t) at each time point t, forming an augmented feature vector. This allows the model to take into account the impact of historical production states when predicting environmental data at time t.

[0049] The processed historical production state vector is used as input, and the corresponding historical environmental monitoring data is used as output to train a hybrid spatiotemporal attention neural network model through supervised learning. Figure 3 As shown, the specific architecture of the hybrid spatiotemporal attention neural network model is as follows.

[0050] The input data is simultaneously fed into a Long Short-Term Memory (LSTM) layer and a one-dimensional convolutional neural network (1D-CNN) layer. The LSTM layer is used to learn the long-term dependencies of each dimension of the production state vector over time. The 1D-CNN layer, on the other hand, extracts local combined features of the production state within a short time window using multiple convolutional kernels of different sizes.

[0051] The outputs of the two layers are concatenated and then fed into the attention mechanism layer. This layer uses a scaled dot product attention mechanism, the calculation of which can be represented by the following formula:

[0052]

[0053] In the formula, Q, K, and V are three matrices learned from the concatenated feature vectors through independent linear transformation layers. Q represents the query information that needs to be focused on, K represents all available query key information, and V represents the value information corresponding to the key information. T d represents the transpose of matrix K. KThe dimension representing K is scaled using its square root to prevent the dot product from becoming too large when its value is large, thus avoiding the softmax function from entering the region of minimum gradient. The softmax function normalizes the calculated attention scores so that their sum is 1, forming the weight distribution.

[0054] Finally, the attention-weighted features are fed into a fully connected layer, outputting two results: the baseline prediction value Y. pred (t) and the prediction standard deviation σ pred (t). The standard deviation represents the degree of uncertainty of the model regarding the current predicted value.

[0055] After the model is trained and deployed, the system enters the online real-time monitoring phase.

[0056] The server unit receives environmental monitoring data in real time. real (t) and production operation data. The production operation data is constructed in real time into a production state vector X(t).

[0057] Input X(t) into the trained dynamic benchmark model to calculate the benchmark prediction value Y in real time. pred (t), prediction confidence interval and prediction standard deviation σ pred (t). The predicted confidence interval is determined by Y. pred (t)±nσ pred As defined by (t), n represents the confidence level coefficient.

[0058] Next, deviation analysis and early warning generation are performed.

[0059] The normalized deviation index I is calculated using the following formula. dev (t):

[0060]

[0061] Within a preset sliding time window, the system performs I... dev The time series of (t) is analyzed, and its first derivative (rate of change v) and second derivative (acceleration a) are calculated by numerical methods such as finite difference.

[0062] The conditions for the system to trigger predictive risk warnings are: I dev The absolute value of (t) exceeds the preset deviation threshold, or v continuously exceeds the rate threshold and a is positive. This warning is issued much earlier than Y. real (t) The statutory alarm threshold is reached.

[0063] On the user terminal's visual early warning interface, the system presents the monitoring results graphically. The graph displays real-time environmental monitoring data curves, baseline predicted value curves, and a predicted trajectory safety cone composed of the upper and lower limits of the predicted confidence interval, which varies with operating conditions.

[0064] When the warning conditions may be triggered, the system will also consider the current Y. real The system extrapolates the trend of change (t) (i.e., the rate of change v) into the future, forming a short-term trajectory prediction line. If this prediction line is about to intersect the boundary of the safety cone, the system will highlight the crossing point and the estimated remaining time, providing the operator with strong predictability.

[0065] The server unit continuously calculates the normalization deviation index I in the background. dev (t) Statistical characteristics over a longer time period of weeks or months. Under perfectly ideal conditions, I dev (t) should be normally distributed with a mean of 0. When the system detects that its long-term mean continuously deviates from 0, or its variance increases significantly beyond the preset model confidence range, it means that the actual production-environmental protection relationship may have changed, and the performance of the original model is degrading. At this time, the system triggers a retraining process to update the model using the latest historical data to ensure its long-term effectiveness.

[0066] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. An environmental data intelligent monitoring system, characterized in that: The environmental data intelligent monitoring system includes a data acquisition module, a data transmission module, and a server unit, which is configured to perform the following operations: The system receives and integrates data from at least two heterogeneous data sources, including real-time environmental monitoring data from the data acquisition module and real-time production and operation data from an external enterprise information system. Based on the correlation between historical environmental monitoring data and historical production and operation data, a dynamic benchmark model is established. The dynamic benchmark model is used to dynamically calculate the benchmark predicted value of environmental monitoring data corresponding to the current production and operation status based on real-time production and operation data. For newly received real-time environmental monitoring data, compare it with the benchmark prediction value calculated by the dynamic benchmark model based on the real-time production and operation data at the same time to generate a real-time deviation value; Based on the changing trend of the real-time deviation value or the comparison result with the preset deviation threshold, a predictive risk warning is generated before the real-time environmental monitoring data reaches the statutory alarm threshold.

2. The environmental data intelligent monitoring system according to claim 1, characterized in that: The server unit is also configured to construct the real-time production operation data into a production state vector, the dimensions of which include equipment operating power to characterize production load, production line start-up and shutdown status to characterize production cycle time, and energy medium usage flow rate to characterize material balance; the real-time environmental monitoring data includes the concentration of exhaust pollutants that have a potential causal relationship with the production state vector.

3. The environmental protection data intelligent monitoring system according to claim 1, characterized in that: The dynamic benchmark model is a hybrid spatiotemporal attention neural network model, the architecture of which includes: Long Short-Term Memory network layer used to process the time series data of each dimension in the production state vector; A one-dimensional convolutional neural network layer configured in parallel with the long short-term memory network layer and used to extract local combined features of the production state vector; and, The attention mechanism layer receives the outputs of the long short-term memory network layer and the one-dimensional convolutional neural network layer. It adopts a scaled dot product attention mechanism, learns the dynamic weights of the impact of different production and operation data dimensions on environmental monitoring data at different time points, and performs weighted fusion on the outputs of the long short-term memory network layer and the one-dimensional convolutional neural network layer to finally output the baseline prediction value.

4. The environmental protection data intelligent monitoring system according to claim 1, characterized in that: The server unit is also configured to generate a normalized deviation index as the real-time deviation value; the dynamic benchmark model outputs a prediction confidence interval and a prediction standard deviation while outputting the benchmark prediction value, and the normalized deviation index is obtained by calculating the ratio of the difference between the real-time environmental monitoring data and the benchmark prediction value to the prediction standard deviation.

5. The environmental protection data intelligent monitoring system according to claim 4, characterized in that: The server unit is also configured to perform first and second derivative calculations on the normalized deviation index within a preset sliding time window to obtain the rate of change and acceleration of the normalized deviation index. Furthermore, when the rate of change continuously exceeds a preset rate threshold and the acceleration is positive, the predictive risk warning is generated.

6. The environmental data intelligent monitoring system according to claim 4, characterized in that: The intelligent environmental data monitoring system also includes a user terminal. The server unit is configured to provide a visual early warning interface to the user terminal. The visual early warning interface is used to display the predicted trajectory safety cone. The upper and lower boundaries of the predicted trajectory safety cone are dynamically formed by the prediction confidence interval output by the dynamic benchmark model. The visual early warning is also used to extrapolate and predict the future short-term trajectory of the real-time environmental monitoring data according to its current trend, and to highlight the point where the predicted trajectory will cross the safety cone.

7. The environmental protection data intelligent monitoring system according to claim 4, characterized in that: The server unit is also configured to perform optimization operations based on model performance degradation. The optimization operations include: continuously monitoring the long-term average and variance of the normalized bias index, and triggering a retraining process for the dynamic benchmark model when the long-term average is detected to have a persistent drift or the variance is detected to have increased significantly and exceeded a preset model confidence range.

8. The environmental protection data intelligent monitoring system according to claim 1, characterized in that: The server unit is also configured to perform time-series data alignment and feature engineering operations before establishing a dynamic baseline model. The time-series data alignment and feature engineering operations include: resampling the environmental monitoring data and the production and operation data with different sampling frequencies to a unified time baseline by interpolation or aggregation; creating multiple time-lag features for the production and operation data at each time point to use the historical production and operation status as one of the input features at the current moment, so that the dynamic baseline model can learn the delayed effect between production activities and environmental impacts.

Images