Enterprise production process carbon emission metering system and method

Abstract

The invention belongs to the technical field of energy management, and particularly discloses an enterprise production process carbon emission metering system and method, and the system comprises a data collection module which is used for collecting a total power consumption data sample of an enterprise production process and a power consumption data sample of key equipment at the collection frequency of minutes as a training data set; the data processing module is used for preprocessing the training data set and completing model training; the model calculation module is used for performing prediction decomposition on newly input total electricity consumption data by utilizing the trained first neural network to obtain a prediction result, and calculating real-time carbon emission according to the prediction result by utilizing the trained second neural network; and the result display module is used for determining the dynamic state of the equipment performance and displaying the dynamic state of the equipment performance and the real-time carbon emission. According to the invention, the accuracy and real-time performance of carbon emission measurement can be improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of energy management, and more particularly relates to a carbon emission metering system and method for enterprise production processes. BACKGROUND

[0002] With the global emphasis on carbon neutrality, accurate metering of enterprise carbon emissions has become a key task. Current research on accurate estimation of carbon emissions using the relationship between electricity and carbon mostly only combines the relationship between electricity and carbon emissions, and when factors such as equipment power consumption and equipment efficiency are involved, the model's interpretability decreases significantly, often ignoring the complex nonlinear relationship in actual industrial processes, resulting in low accuracy and insufficient real-time performance of the estimation results.

[0003] Therefore, how to overcome the defects of poor carbon emission metering accuracy and insufficient real-time performance of the prior art is a technical problem that needs to be solved at present. SUMMARY

[0004] In view of the defects of the prior art, the purpose of the present application is to provide a carbon emission metering system and method for enterprise production processes, which aims to solve the problem of poor carbon emission metering accuracy and insufficient real-time performance of the prior art.

[0005] To achieve the above-mentioned purpose, in a first aspect, the present application provides a carbon emission metering system for enterprise production processes, comprising: a data acquisition module, a data processing module, a model calculation module and a result display module; The data acquisition module is configured to acquire total power consumption data samples and key equipment power consumption data samples of the enterprise production process as a training data set with a minute acquisition frequency; The data processing module is configured to preprocess the training data set and complete model training; The model calculation module is configured to use the trained first neural network to predict and decompose the total power consumption data to be estimated to obtain a prediction result, and use the trained second neural network to calculate real-time carbon emissions based on the prediction result; The result display module is configured to determine a device performance dynamic and display the device performance dynamic and real-time carbon emissions; the device performance dynamic includes device power consumption trend, operating state and prediction error.

[0006] Optionally, the first neural network is an informer neural network, and the second neural network is an MLP neural network; The informer neural network is configured to predict and decompose the preprocessed total power consumption data to obtain key equipment power consumption and the operating state of the direct emission system; The MLP neural network comprises a mapping relationship between the power consumption of the equipment and the carbon emission, and is configured to calculate the real-time carbon emission by the power consumption of the key equipment through the mapping relationship.

[0007] Optionally, the data processing module comprises a timestamp unification submodule, a gap filling submodule, and a fitting submodule. The timestamp unification submodule is configured to perform timestamp unification processing on the total power consumption data sample and the power consumption data sample of the key equipment. The gap filling submodule is configured to determine the to-be-improved data with missing values and outliers in the total power consumption data sample and the power consumption data sample of the key equipment, and fill the missing values in the to-be-improved data by using the Lagrange interpolation method. The fitting submodule is configured to fit the relationship between the data points by using a high-order polynomial.

[0008] Optionally, the data acquisition module further comprises a state representation submodule and a reference submodule. The state representation submodule is configured to represent the running state of the key equipment by using discrete numerical values; the first numerical value represents a shutdown state, the second numerical value represents a partial load state, and the third numerical value represents a full load state. The reference submodule is configured to acquire the fuel feed quantity, the low calorific value, the chemical raw material feed quantity, and the carbonate content as the basis for reference carbon emission calculation.

[0009] Optionally, the data processing module further comprises a feature mining submodule and a normalization submodule. The feature mining submodule is configured to mine features from the total power consumption data sample and the power consumption data sample of the key equipment, and extract trend features, seasonal features, and residual features to reflect the internal laws and change trends of the data. The normalization submodule is configured to construct a correlation function and an autocorrelation function to analyze the internal laws of the data, and convert the data features into the same scale by using a min-max normalization method.

[0010] Optionally, the informer neural network comprises a self-attention subnetwork and a feedforward subnetwork. The self-attention subnetwork is configured to calculate attention weights by using a probabilistic sparse self-attention mechanism. The feedforward subnetwork is configured to perform linear transformation and nonlinear activation on the preprocessed input data by using a fully connected layer.

[0011] Optionally, the result display module comprises a warning submodule, and the warning submodule is configured to trigger a threshold warning mechanism. The early warning mechanism includes a primary early warning response, an intermediate early warning response, and a high-level early warning response; the primary early warning response pushes a short message notification, the intermediate early warning response generates a diagnostic report, and the high-level early warning response links the DCS system to adjust the load.

[0012] The application provides an enterprise production process carbon emission metering method based on the enterprise production process carbon emission metering system. Collecting total power consumption data samples and key equipment power consumption data samples of a certain local area of an enterprise at a minute frequency; Preprocessing the training data set and completing model training; Using the trained first neural network to predict and decompose the total power consumption data to be estimated to obtain a prediction result, and using the trained second neural network to calculate real-time carbon emissions based on the prediction result; Determine the equipment performance dynamics, and display the equipment performance dynamics and real-time carbon emissions; the equipment performance dynamics include equipment power consumption trend, running state and prediction error.

[0013] Optionally, it further comprises: Building a multi-dimensional data acquisition network on the power supply line of the key equipment using smart meters and sensors, and importing historical operation data covering typical working conditions; Real-time data upload through the Internet of Things architecture, calculation of minute-level carbon emissions, and dynamic display through visual boards; Incremental training of the informer neural network and the MLP neural network based on the first time interval, and verification of the equipment accuracy based on the second time interval.

[0014] Optionally, the method comprises: Using an informer neural network to input the preprocessed total power consumption data, dynamically extracting key equipment power consumption through a probabilistic sparse self-attention mechanism, introducing a distillation operation to compress the attention layer output dimension, and outputting key equipment power consumption decomposition results and direct emission system running state; Inputting the key equipment power consumption, running state and fuel parameters into the MLP neural network, fitting the mapping relationship between equipment power consumption and carbon emissions, and obtaining carbon emission intensity prediction values.

[0015] In a third aspect, the present application provides an electronic device, comprising: at least one memory configured to store a program; and at least one processor configured to execute the program stored in the memory, wherein the processor is configured to execute the method described in the first aspect or any possible implementation manner of the first aspect when the program stored in the memory is executed.

[0016] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program, and the computer program, when executed on a processor, causes the processor to execute the method described in the first aspect or any possible implementation manner of the first aspect.

[0017] In a fifth aspect, the present application provides a computer program product, which, when executed on a processor, causes the processor to execute the method described in the first aspect or any possible implementation manner of the first aspect.

[0018] It can be understood that the beneficial effects of the above-mentioned second aspect to fifth aspect can be referred to the related description in the first aspect, which will not be repeated here.

[0019] Overall, compared with the prior art, the above technical solutions conceived by the present application have the following beneficial effects: (1) The present application collects total power consumption and key equipment power consumption data at the minute level, combines the preprocessing of the data processing module, ensures the accuracy and consistency of the input data, provides high-quality basic information for subsequent models to improve the accuracy of model recognition. The first neural network can deeply mine the time sequence features and device correlation in the total power consumption data, realize more accurate key equipment power consumption decomposition, effectively reduce the prediction error caused by the change of equipment running state, thereby improving the accuracy of carbon emission calculation. The second neural network directly calculates the real-time carbon emission based on the high-precision decomposition result combined with the mapping relationship, further improving the accuracy of the final result.

[0020] (2) The present application intuitively presents the advantages of high precision and real-time to the user through the real-time display and early warning mechanism of the result display module, improves the user experience and the convenience of system management operation.

[0021] (3) The present application ensures the efficiency of data processing and model inference through the minute-level data collection frequency and the optimized neural network structure, so that the real-time calculation and update of carbon emission is realized.

[0022] (4) The present application reduces the hardware deployment cost by fusing non-intrusive load monitoring technology and only needs to monitor the total circuit to decompose the key equipment power consumption. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1is a structural schematic diagram of an enterprise production process carbon emission metering system provided by an embodiment of the present application; Figure 2 is one of flow schematic diagrams of an enterprise production process carbon emission metering method provided by an embodiment of the present application; Figure 3 is another of flow schematic diagrams of an enterprise production process carbon emission metering method provided by an embodiment of the present application; Figure 4 is a framework schematic diagram of a load decomposition method based on a deep informer and an MLP model provided by an embodiment of the present application; Figure 5 is a flowchart of a microalgae carbon fixation process device provided by an embodiment of the present application; Figure 6 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0025] The term "and / or" in this document is a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that there are three cases of A alone, A and B together, and B alone. The symbol " / " in this document represents an or relationship of associated objects, for example, A / B represents A or B.

[0026] The terms "first" and "second" and the like in the description and claims herein are used to distinguish different objects, and are not used to describe a specific order of the objects. For example, the first response message and the second response message are used to distinguish different response messages, and are not used to describe a specific order of the response messages.

[0027] In the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or advantageous than other embodiments or design solutions. Rather, the use of "exemplary" or "for example" is intended to present concepts in a concrete manner.

[0028] In the description of the embodiments of the present application, unless otherwise specified, "a plurality of" means two or more, for example, a plurality of processing units means two or more processing units, and the like; a plurality of elements means two or more elements, and the like.

[0029] The embodiments of the present application will be described below with reference to the accompanying drawings.

[0030] With reference to Figure 1 The present application provides an enterprise production process carbon emission metering system, comprising a data acquisition module, a data processing module, a model calculation module and a result display module. The data acquisition module is configured to acquire total power consumption data samples and key equipment power consumption data samples of the enterprise production process as a training data set at a minute acquisition frequency. The data processing module is configured to preprocess the training data set and complete model training. The model calculation module is configured to use the trained first neural network to predict and decompose the total power consumption data to be estimated to obtain a prediction result, and use the trained second neural network to calculate real-time carbon emissions based on the prediction result. The result display module is configured to determine a device performance dynamic and display the device performance dynamic and real-time carbon emissions. The device performance dynamic includes device power consumption trend, operating state and prediction error.

[0031] Specifically, the data acquisition module is mainly composed of sensors and smart meters, and is deployed in relevant local areas of the enterprise to acquire total power consumption data samples and key equipment power consumption data samples of a certain local area of the enterprise in real time. The data acquisition frequency is minute level.

[0032] The data processing module is configured to perform uniform timestamp processing on the collected various types of data to ensure the synchronization and consistency of the data, and perform Lagrange interpolation on the parts with missing and abnormal data in the total power consumption and key equipment power consumption data to fill in the missing values. A high-order polynomial is used to fit the relationship between data points, so as to more accurately estimate the missing values.

[0033] Specifically, the data processing module performs uniform timestamp processing on the total power consumption data samples and key equipment power consumption data samples. The data processing module determines the to-be-improved data with missing and abnormal values in the total power consumption data samples and key equipment power consumption data samples, and uses Lagrange interpolation to fill in the missing values. A high-order polynomial is used to fit the relationship between data points.

[0034] The model calculation module includes a two-stage deep learning model. The first stage uses an informer neural network to decompose the total power consumption data, and outputs the key equipment power consumption and the operating state of the direct emission system. The second stage uses an MLP neural network to establish a mapping relationship between the equipment power consumption and the carbon emissions based on the output of the first stage, and outputs the real-time carbon emissions. The result display module is configured to display carbon emissions, equipment power consumption trend, operation state and prediction error in real time, and trigger a threshold warning mechanism.

[0035] Optionally, the first neural network is an informer neural network, and the second neural network is an MLP neural network. The informer neural network is configured to predict and decompose the preprocessed total power consumption data to obtain key equipment power consumption and operation state of a direct emission system. The MLP neural network includes a mapping relationship between equipment power consumption and carbon emissions, and is configured to calculate the real-time carbon emissions by calculating the key equipment power consumption through the mapping relationship.

[0036] Specifically, in the embodiment of the present application, the informer neural network module is responsible for predicting and decomposing the preprocessed total power consumption data. The module first receives the total power consumption data collected during the production process of the enterprise. After the data is preprocessed, it is input into the informer neural network.

[0037] It should be noted that informer is a deep learning model based on the Transformer architecture, and its core advantage is that it can efficiently process time series data and capture long-term sequential dependencies. Through the ProbSparse self-attention mechanism, the model reduces the computational complexity while preserving the key features in the power consumption data. After the prediction and decomposition of the informer module, the total power consumption is refined into the power consumption of each key equipment, and the operation state of the direct emission system is further analyzed. This process not only realizes the accurate decomposition of power consumption data, but also ensures the visualization of key equipment power consumption and direct emission system operation state, providing a reliable data foundation for subsequent carbon emission calculation.

[0038] The MLP (Multi-Layer Perceptron) neural network module is responsible for mapping the key equipment power consumption output by the informer module into real-time carbon emissions. By establishing a mapping relationship between equipment power consumption and carbon emissions, the module uses the efficient nonlinear fitting capability of the MLP neural network to convert power consumption data into carbon emissions. Through training, MLP can accurately capture the complex nonlinear relationship between power consumption and carbon emissions, thereby calculating the real-time carbon emissions of the enterprise. Not only does it improve the calculation speed, but also ensures the accuracy of carbon emission calculation.

[0039] The two modules work together to form a complete chain from electricity consumption prediction to carbon emission calculation. The informer module provides high-quality data input for the MLP module through accurate electricity consumption decomposition, and the MLP module converts electricity consumption data into real-time carbon emissions through efficient mapping relationship calculation. Through the cooperative working mode, the accuracy of the data is guaranteed, and the calculation efficiency is improved, so that the carbon emissions can be updated in real time and accurately reflect the actual emission situation of the enterprise, significantly improving the accuracy and real-time performance of carbon emission prediction.

[0040] Optionally, the state representation submodule is configured to represent the operating state of the key equipment by using discrete numerical values; the first numerical value represents the shutdown state, the second numerical value includes the partial load state, and the third numerical value includes the full load state. The reference submodule is configured to collect the fuel feed quantity, the low heat value, the chemical raw material feed quantity, and the carbonate content as the basis for calculating the reference carbon emission.

[0041] Specifically, the data acquisition module can collect the operating state of the equipment related to direct emission in real time through the state representation submodule, and represent the operating state by using discrete numerical values, for example, 0 represents the shutdown state, 0.17 represents the partial load state, and 1 represents the full load state, which accurately reflects the operating load of the equipment and provides key parameters for carbon emission estimation.

[0042] The data acquisition module can also collect relevant data such as fuel feed quantity, low heat value, and relevant data such as chemical raw material feed quantity and carbonate content through the reference submodule, which is an important basis for calculating the reference carbon emission, so that the reference value of carbon emission can be calculated more accurately, the accuracy of carbon emission monitoring can be improved, and the influence of material consumption in the production process on carbon emission can be considered comprehensively.

[0043] Optionally, the data processing module further includes a feature mining submodule and a normalization submodule; the feature mining submodule is configured to mine features from the total electricity consumption data sample and the electricity consumption data sample of the key equipment, extract trend features, seasonal features, and residual features to reflect the internal laws and trends of the data; and the normalization submodule is configured to construct a correlation function and an autocorrelation function to analyze the internal laws of the data, and convert the data features into the same scale by using the minimum and maximum value normalization method.

[0044] Specifically, the data processing module constructs a Lagrange interpolation polynomial function based on a plurality of known data points, so that the value of the function at the known data points is equal to the actual observed value, and then uses the polynomial function to calculate the missing value.

[0045] The basic form of the Lagrange interpolation polynomial constructed by the data processing module is as follows:

[0046] In the formula, The k-th interpolation base function of the j-th time of the independent variable x is represented as follows:

[0047] Wherein, (k=0, 1,..., n).

[0048] The data processing module deeply mines the collected data through a feature mining submodule, extracts features such as trends, seasonality and residuals, and these features can reflect the internal laws and trends of the data.

[0049] The data processing module constructs a correlation function and an autocorrelation function. For a given time series, the overall autocorrelation function for the discrete time series xt has a mean of The autocorrelation function ρ(k) at lag k is defined as:

[0050] Wherein, The total power consumption data at the t-th time point is represented as: The average power consumption is represented as: The total power consumption data at the k-th time point is represented as: N represents the length of the discrete time sequence.

[0051] The partial autocorrelation function is defined as:

[0052] The data processing module converts the data features into the same scale by using the minimum and maximum value normalization method through a normalization submodule. For N samples The normalized feature of each dimension x is:

[0053] Wherein, And The minimum and maximum values of the feature x on the sample are respectively.

[0054] Optionally, the informer neural network comprises a self-attention subnetwork and a feedforward subnetwork. The self-attention subnetwork is configured to calculate attention weights by using a probabilistic sparse self-attention mechanism. The feedforward subnetwork is configured to perform linear transformation and nonlinear activation on the preprocessed data input by a fully connected layer.

[0055] Specifically, in the model calculation module, the informer neural network adopts a ProbSparse self-attention mechanism, and the attention weight is calculated as follows: ; wherein, represent three matrices, respectively, a query matrix, a Key matrix and a Value matrix, is the dimension size of the Key, and Softmax is a normalized exponential function.

[0056] In the model calculation module, the informer neural network adopts a feedforward neural network to perform linear transformation and nonlinear activation on the input data through a fully connected layer, and the calculation formula is as follows:

[0057] wherein, W1 and W2 are weight matrices, and b1 and b2 are bias terms.

[0058] Optionally, the result display module comprises a warning submodule for triggering a threshold warning mechanism; wherein the warning mechanism comprises a primary warning response, an intermediate warning response and a high-level warning response; the primary warning response pushes a short message notification, the intermediate warning response generates a diagnosis report, and the high-level warning response links the DCS system to adjust the load.

[0059] With reference to Figure 2 The application provides an enterprise production process carbon emission metering method based on the enterprise production process carbon emission metering system. S101. Collecting total power consumption data samples and key equipment power consumption data samples of a certain local area of an enterprise at a minute frequency; S102. Preprocessing the training data set and completing model training; S103. Using the trained first neural network to predict and decompose the total power consumption data to be estimated to obtain a prediction result, and using the trained second neural network to calculate real-time carbon emissions based on the prediction result; S104. Determining the equipment performance dynamics and displaying the equipment performance dynamics and real-time carbon emissions; the equipment performance dynamics include equipment power consumption trend, running state and prediction error.

[0060] Optionally, it further comprises: A multi-dimensional data acquisition network is constructed by using smart meters and sensors on the power supply lines of key equipment, and historical operation data covering typical working conditions are imported; Through the Internet of Things architecture, data is uploaded in real time, minute-level carbon emissions are calculated, and dynamic display is realized through a visual board. Incremental training is performed on the informer neural network and the MLP neural network based on the first time interval, and device accuracy is verified based on the device accuracy at the second time interval.

[0061] Reference Figure 3 , Figure 3 Schematic diagram of a method for estimating carbon emissions during a production process according to an embodiment of the present application, including: Identify key equipment for enterprise carbon emissions, complete system hardware deployment, and import historical operating data; Set up a two-stage model to calculate minute-by-minute carbon emissions and dynamically display them through a visual dashboard; Incrementally train the model every quarter, verify equipment accuracy every month, and perform maintenance and optimization.

[0062] Specifically, the basic principle of the enterprise production process carbon emission measurement method provided in this application is to decouple the complex relationship between total electricity consumption and carbon emissions through a two-stage model, and improve the adaptability of dynamic scenarios by combining equipment operation state variables. The method includes the following steps: (1) Deployment and initialization: Complete system hardware deployment and data modeling infrastructure, install smart meters and sensors on key enterprise equipment, build a multi-dimensional data collection network, and import 12 months of historical operating data to cover typical operating conditions; (2) Real-time operation process: upload data in real time through the IoT architecture, calculate minute-level carbon emissions, and dynamically display them through a visual dashboard; (3) Maintenance and optimization: Incrementally train the model every quarter and verify the equipment accuracy every month to ensure that the load decomposition accuracy is ≥92%.

[0063] Optionally, the using the trained first neural network to perform predictive decomposition on the total electricity consumption data to be estimated to obtain a prediction result, and using the trained second neural network to calculate the real-time carbon emissions based on the prediction result includes: The informer neural network takes pre-processed total power consumption data as input, dynamically extracts the power consumption of key equipment through a probabilistic sparse self-attention mechanism, introduces a distillation operation to compress the output dimension of the attention layer, and outputs the power consumption decomposition results of key equipment and the operating status of the direct emission system; The power consumption, operating status and fuel parameters of key equipment are input into the MLP neural network to fit the mapping relationship between equipment power consumption and carbon emissions, and the predicted value of carbon emission intensity is obtained.

[0064] In particular, the above step (1) deployment and initialization is further optimized as follows: first, install an industrial-grade smart meter on the power supply line of the enterprise's key power-consuming equipment; for process systems with direct carbon emissions such as boilers and smelting furnaces, deploy temperature, pressure, and flow sensor groups simultaneously to form a multi-dimensional data acquisition network. All monitoring devices are connected to an edge computing server configured with a NILM algorithm through an industrial Internet of Things gateway, and a device-level power consumption feature database is constructed.

[0065] Preferably, during initialization in step (1), historical operation data covering 1-12 months need to be imported, including minute-level total load data recorded by the total power distribution room meter, independent meter readings of each sub-device, fuel / raw material consumption logs of direct emission systems, and device start / stop status, load rate changes, and other working condition label data.

[0066] The imported data in step (1) initialization is further optimized as follows: the data must include complete cycle records of typical working conditions such as device shutdown standby, abnormal jump, 30%-70% partial load, and 95%-100% full load.

[0067] The real-time running process in step (2) is further optimized as follows: the first stage of the informer neural network decomposition: input the cleaned total power consumption time series data, dynamically extract the key device power consumption features through the ProbSparse self-attention mechanism, and the formula is:

[0068] The output is the decomposition result of the key device power consumption (such as kiln tail fan, cooling fan) and the running state of the direct emission system.

[0069] Preferably, the output dimension of the distillation operation compression attention layer is introduced to reduce the memory occupation by 20% and the inference speed is improved to seconds.

[0070] The second stage of MLP carbon emission mapping: input the device power consumption, running state, and fuel parameters (such as raw coal low heat value) output by the first stage, fit the nonlinear relationship through a multilayer perceptron (hidden layer ≥3 layers, neuron number ≥128, activation function = ReLU), and output the minute-level carbon emission intensity prediction value with a mean square error (MSE) ≤0.05.

[0071] Preferably, batch normalization is used to accelerate model convergence, and the training efficiency is improved by 30%.

[0072] The visual board renders the carbon emission trend curve in real time, superimposes the model prediction value (blue) and the actual correction value (red) to form a double-line comparison graph; When the prediction deviation continuously exceeds the threshold value (>8% for more than 15 minutes), a hierarchical warning is triggered: Primary warning: Push short message to administrator, prompt abnormal carbon emission equipment location; Intermediate warning: Automatically generate PDF diagnosis report, including high energy consumption equipment energy efficiency analysis and emission reduction suggestions; Severe warning: Link DCS system to execute preset load adjustment strategy (such as reducing fan speed), and forcibly reduce carbon emission.

[0073] Preferably, an adaptive threshold algorithm is introduced to dynamically adjust the warning threshold (±5%~±10%) according to historical data, avoiding false positives.

[0074] Step (3) is further optimized to: extract 1% of real-time data every 15 minutes, cross-verify with a portable high-precision monitor (error <0.5%), and if the verification deviation is >1.5%, trigger model incremental fine-tuning and update MLP network weight parameters; Preferably, an online learning mechanism is used, with incremental training time ≤5 minutes, to ensure continuous optimization of the model.

[0075] The present application can achieve the following effects: High-precision monitoring: The error of carbon emission estimation is reduced by 30% compared with traditional methods, and the error fluctuation in sudden change scenarios (such as equipment start-stop) is <3%; Real-time response: Minute-level data update, warning response time <15 minutes; Cost optimization: NILM technology reduces hardware deployment cost by 45%, and operation and maintenance efficiency is improved by 60%; Interpretability: Intermediate variables (equipment power consumption, running state) clearly indicate the source of carbon emission, supporting accurate energy-saving decisions.

[0076] The present application is particularly suitable for high-energy-consuming industries such as cement and power, and through dynamic carbon footprint monitoring and intelligent warning, it helps enterprises to achieve green production and carbon neutralization goals.

[0077] The present application will be described in detail below in conjunction with specific embodiments: Example 1: The carbon emission estimation model is established based on the cement clinker section. The direct carbon emissions in the cement production process mainly include two parts: the first part is the process emission, which is the decomposition of carbonates in raw materials during calcination; the second part is the fuel combustion emission, which is the release of a large amount of CO2 during the combustion of raw coal to provide heat for the cement kiln. The power consumption of the firing system is mainly in the airflow conveying system, accounting for about 70% of the total power consumption of the firing system. Therefore, the key point of energy saving and consumption reduction in the clinker firing system is the airflow conveying system, i.e. the power consumption of each fan. The kiln tail high-temperature fan, kiln tail electric dust collection exhaust fan, cooling fan, primary fan, grate cooler cooling fan, kiln head exhaust fan, air cooler fan, Roots blower, kiln tail electric room transformer and kiln head electric room transformer are selected as variables. The power consumption of each device in the power system is monitored in real time by sensors and monitoring equipment. The carbon emission data of the clinker section under different working conditions in a long period of time, the total power consumption data and the minute-level power consumption data of each device are collected.

[0078] Referring to Figure 4 , a two-stage deep learning model is set up by combining the informer neural network with the MLP network. In order to improve the estimation accuracy and the interpretability of the model, the output of the informer network is defined as the power consumption of the key equipment in power generation and the running state of the clinker firing system. In the training stage, the actual value is used to train both stages of the model. However, in the verification stage, the estimated value of the intermediate variable generated by the trained first-stage model is input into the trained second-stage model, instead of the real data to estimate the carbon emission. The first stage inputs the total power consumption data into the informer network, trains the model to extract the power consumption data of each key equipment, and predicts the running state of the clinker firing system. The second stage takes the power consumption of the key equipment and the running state of the clinker firing system output by the first stage as input, and trains the deep learning model to predict the carbon dioxide emission.

[0079] The total power consumption and the power consumption data of each key equipment are preprocessed. For the parts with missing data, Lagrange interpolation method is used to fill in the missing values. A polynomial function is constructed based on multiple known data points, so that the value of the function at the known data points is equal to the actual observed value, and then the polynomial function is used to calculate the missing values. The basic form of Lagrange interpolation polynomial is as follows:

[0080] In the formula, denotes the k-th interpolation basis function of the independent variable x at the j-th time, and its specific form is as follows:

[0081] wherein (k = 0, 1,..., n). In addition, when the deep learning neural network is trained, it is difficult to learn from data with a too large numerical range, and it is not very sensitive to the scale characteristics of the input data, but is based on the analysis of the numerical size. When calculating the absolute error of the prediction result, the characteristics with larger numerical values will play a dominant role. Therefore, it is necessary to convert the data characteristics to the same scale. The minimum and maximum value normalization method is adopted, and its principle is as follows: assuming that there are N samples, for each dimension of the feature x, the normalized feature is:

[0082] wherein, and are the minimum and maximum values of the feature x on the sample, respectively.

[0083] At the same time, for several sudden change points of CO2 emission data, these mutations are caused by the conversion between the states of the clinker burning system. In order to improve the accuracy of the estimation, a three-level variable representing the clinker burning system is included, wherein 0 represents shutdown, 0.17 represents partial load operation, and 1 represents full load operation. This inclusion provides a more detailed view of the potential factors affecting CO2 emissions, enabling the model to adapt to the dynamic operating conditions of the plant.

[0084] Finally, the load data from n key devices and the clinker burning system operating state are selected and set as X1, X2,..., X n , Spearman correlation analysis is performed on n variables and carbon emissions Y, and the Spearman correlation coefficient is calculated. The process is to sort the variables X and Y from small to large and encode the rank, and use the rank R X and R Y to represent it. When sorting, the phenomenon of equal data causing the same rank is called tie, and the average rank of each data is taken as the rank of each data. The calculation formula of the Spearman correlation coefficient r s is as follows:

[0085] In order to avoid introducing irrelevant variables that may produce noise and reduce the estimation accuracy, 4-5 variables with the highest correlation are selected as intermediate variables.

[0086] In the first stage model, the informer neural network is involved to generate the estimated values of the key equipment power consumption and the burning system operating state with the total power consumption as the input, and to clarify the power consumption mode and distribution mode of the key equipment in the clinker section.

[0087] The informer neural network adopts ProbSparse self-attention mechanism and distillation operation, can adaptively select the key vector most important to the current query vector, greatly reduces the calculation amount and memory occupation, makes it can efficiently process long sequence time series data, especially beneficial to capture the long-term dependence and fluctuation of power consumption, so as to more accurately estimate the carbon emission. The informer model mainly involves the calculation of self-attention mechanism and feedforward neural network. The calculation formula of self-attention mechanism usually includes the calculation of query (Query), key (Key) and value (Value), as well as the calculation and weighted sum of attention weight. The input sequence is x, and the query Q, key K and value V are obtained after linear transformation. The calculation formula of the output of self-attention mechanism is:

[0088] Wherein, d k is the dimension size of the key, and Softmax is the normalization exponential function. The feedforward neural network performs linear transformation and nonlinear activation on the input data through the full connection layer, and the calculation formula is wherein, W1, W2 are weight matrices, and b1, b2 are bias terms.

[0089] In the second stage model, the MLP model is used to associate the power consumption of the key equipment and the running state of the cement clinker burning system with the CO2 emission. In the second stage, the actual value of the power consumption of the key equipment and the running state of the gas turbine are used as input, and the CO2 emission is generated by the model. In this stage, the MLP model is used to associate the power consumption of the key equipment and the running state of the cement clinker burning system with the CO2 emission. The MLP model contains multiple hidden layers, each hidden layer contains multiple neurons, and the neurons are connected in full connection mode and use activation function for nonlinear transformation. Input layer design: according to the characteristics of the collected data, the number of neurons in the input layer is determined. The parameters of the power consumption of the key equipment and the running state of the cement clinker burning system are used as the neurons of the input layer, and a total of 5-6 input neurons are set. Hidden layer design: adopt multi-layer hidden layer structure, the number of neurons in each hidden layer is determined according to the empirical formula or through test, and select appropriate activation function (such as ReLU function) to enhance the nonlinear expression ability of the model. Output layer design: the output layer is set with one neuron, corresponding to the predicted value of CO2 emission.

[0090] Finally, in the verification process, the two stages are integrated to form a complete structure. The total power consumption is used as input, which is processed by the informer network to generate the estimated value of the power consumption of the key equipment and the running state of the clinker burning system. Then these estimated values are input into the trained second stage MLP model to estimate the total carbon dioxide emission.

[0091] Example 2 A carbon emission estimation model is established for a biomass direct-fired power plant coupled with a microalgae cultivation system. The carbon capture capacity of the microalgae cultivation system is combined to quantify the carbon sequestration of microalgae, and a complete carbon accounting system is constructed. The carbon emissions in the biomass direct-fired power generation process mainly come from fuel combustion emissions. Biomass combustion releases a large amount of CO2, while the supporting microalgae cultivation system can capture CO2. Since the CO2 produced by biomass combustion is included in the agricultural and forestry carbon cycle, and combined with the high photosynthetic carbon sequestration capacity of microalgae, an active carbon sink can be achieved. The synergistic effect of the two makes the system as a whole present negative carbon emissions. The microalgae carbon sequestration process flow chart is shown in the figure.

[0092] This example takes the flue gas generated by the boiler as the starting point, which is pretreated by the cooler, water remover, and mixed gas filter, and then transported to the reactor unit by the flue gas fan and air fan. The reactor unit includes multiple photobioreactors and circulating tanks. The system synchronously integrates the water supply unit, sewage treated by the water supply pump, filter, and microorganism disinfection device, and the dosing unit to input the nutrient salt and wood ash dosing box and dosing pump to the reactor unit. The microalgae cultivation is realized in multiple photobioreactors, i.e., microalgae grow using CO2 in flue gas. The whole process is monitored in real time by gas flow sensors and CO2 sensors. The final harvesting unit separates, transports, and dries the product by centrifuge, peristaltic pump, and dryer, respectively, and stores the product as concentrated algae liquid and algae powder. The purified gas is discharged through the chimney unit. The system adopts modular design, covering the whole process of pretreatment, cultivation monitoring, nutrient supply, and product recovery, realizing the collaborative treatment of flue gas resource utilization and microalgae production.

[0093] The carbon sequestration efficiency of microalgae is affected by multiple key factors, including CO2 concentration, O2 concentration, light intensity, temperature, light uniformity, nutrient distribution, etc. Therefore, this case selects multiple key devices in the microalgae cultivation system, such as flue gas fan, air fan, water supply pump, dosing pump, circulating pump, LED lamp array, pressure transmitter, and temperature control equipment, as variables. First, the power consumption of each device in the microalgae cultivation system is monitored in real time by sensors and monitoring devices. The carbon sequestration amount, total power consumption, and minute-level power consumption data of the microalgae cultivation system under different working conditions over a long period of time are collected.

[0094] High-precision portable CO2 infrared sensors and gas flow sensors are temporarily deployed at the gas inlet and outlet of the system to collect CO2 concentration and gas flow rate data in real time for building a prediction model, and periodic data collection strategies are adopted to realize dynamic calibration and optimization of the model. The minute-level carbon sequestration amount calculation formula is as follows: Q: gas flow (L / min, standard state), C in , C out: Inlet / outlet CO2 concentration (ppm), 44 / 22.4: CO2 molar mass conversion coefficient, α: pH correction factor, β: temperature correction factor.

[0095] A two-stage deep learning model was set up, combining an informer neural network with an MLP network. To improve the accuracy of the estimation and the interpretability of the model, the output of the informer network was defined as the power consumption of key equipment in power generation. During the training phase, both stages of the model were trained using actual values. However, during the validation phase, the estimated values ​​of the intermediate variables generated by the trained first-stage model were input into the trained second-stage model instead of the real data to estimate carbon emissions. In the first stage, the total power consumption data was input into the informer network, and the training model extracted the power consumption data of each key equipment. In the second stage, the power consumption of key equipment output from the first stage was used as input to train the deep learning model to predict the carbon sequestration of microalgae.

[0096] The total power consumption and power consumption data of each key device are preprocessed. For any missing data, Lagrange interpolation is used to fill in the gaps. A high-order polynomial is used to fit the relationship between data points, thereby more accurately estimating the missing values. A polynomial function is constructed based on multiple known data points, so that the value of the function at the known data points is equal to the actual observed value. This polynomial function is then used to calculate the missing values. The basic form of the Lagrange interpolation polynomial is shown below:

[0097] Where, l k (x j ) indicates that the independent variable x takes the kth interpolation basis function at the jth time. The specific form is as follows:

[0098] (k=0, 1, ..., n).

[0099] In addition, deep learning neural networks have difficulty learning from data with a large numerical range during training. They are not very sensitive to the scale characteristics of the input data and analyze based on the numerical size. When calculating the absolute error of the forecast results, the features with larger numerical values ​​will play a dominant role. Therefore, it is necessary to convert the data features to the same scale. The minimum and maximum normalization method is used, and its principle is as follows: Assume that there are N samples , for each dimension feature x, the normalized feature is:

[0100] in, and are the minimum and maximum values ​​of feature x in the sample, respectively.

[0101] Finally, load data is selected from n key devices, set as X1, X2,..., X n Spearman correlation analysis is performed on n variables and carbon fixation amount Y, and Spearman correlation coefficient is calculated. The process is to sort the variables X and Y from small to large and encode the rank, and the rank R X and R Y are represented. When sorting, the phenomenon of equal data causing the same rank is called tie, and the average rank is taken as the rank of each data at this time. The calculation formula of Spearman correlation coefficient r s is as follows:

[0102] In order to avoid introducing irrelevant variables that may produce noise and reduce the estimation accuracy, 4-5 variables with the highest correlation are selected as intermediate variables.

[0103] In the first stage model, the Informer neural network is involved to generate the power consumption of key devices with total power consumption as input, and to clarify the power utilization mode and distribution mode of key devices in microalgae carbon fixation system.

[0104] The Informer neural network adopts ProbSparse self-attention mechanism and distillation operation, which can adaptively select the most important key vector for the current query vector, greatly reducing the computational complexity and memory occupation, so that it can efficiently process long sequence time series data, especially beneficial to capture the long-term dependence and fluctuations of power consumption, so as to more accurately estimate carbon emissions. The Informer model mainly involves the calculation of self-attention mechanism and feedforward neural network. The calculation formula of self-attention mechanism usually includes the calculation of query (Query), key (Key) and value (Value), as well as the calculation of attention weight and weighted sum. The input sequence is x, and the query Q, key K and value V are obtained after linear transformation. The calculation formula of the output of self-attention mechanism is:

[0105] d k is the dimension size of the key, and Softmax is the normalization exponential function. The feedforward neural network performs linear transformation and nonlinear activation on the input data through the fully connected layer, and the calculation formula is:

[0106] Where, W1, W2 are weight matrices, and b1, b2 are bias terms.

[0107] In the second stage model, the MLP model is used to link the key equipment power consumption and the microalgae carbon fixation amount. In the second stage, the actual value of the key equipment power consumption is used as the input, and the microalgae carbon fixation amount is generated by the model. In this stage, the MLP model is used to link the key equipment power consumption and the microalgae carbon fixation amount. The MLP model includes multiple hidden layers, each hidden layer includes multiple neurons, and the neurons are connected in a full connection manner and use an activation function for nonlinear transformation. Input layer design: according to the data characteristics collected, the number of neurons in the input layer is determined. The parameters of the key equipment power consumption are respectively used as the neurons of the input layer, and a total of 5-6 input neurons are set. Hidden layer design: a multi-layer hidden layer structure is used, the number of neurons in each hidden layer is determined according to an empirical formula or through experiments, and a suitable activation function (such as ReLU function) is selected to enhance the nonlinear expression ability of the model. Output layer design: the output layer is set with one neuron, corresponding to the predicted value of the microalgae carbon fixation amount.

[0108] Finally, in the verification process, the two stages are integrated to form a complete structure. The total power consumption is used as the input, which is processed by the Informer network to generate the estimated value of the key equipment power consumption. Then these estimated values are input into the trained second stage MLP model to estimate the microalgae carbon fixation amount.

[0109] Reference Figure 6 Based on the method in the above embodiment, the electronic device provided by the embodiment of the present application can include a processor (Processor) 610, a communication interface (Communications Interface) 620, a memory (Memory) 630, and a communication bus 640. The processor 610, the communication interface 620, and the memory 630 can communicate with each other through the communication bus 640. The processor 610 can call the logical instructions in the memory 630 to execute the method in the above embodiment.

[0110] In addition, the logical instructions in the memory 630 described above can be implemented in the form of a software function unit and sold or used as an independent product. When used, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application.

[0111] Based on the method in the above embodiments, the embodiments of the present application provide a computer readable storage medium, which stores a computer program. When the computer program is run on a processor, the processor executes the method in the above embodiments.

[0112] Based on the method in the above embodiments, the embodiments of the present application provide a computer program product, which, when run on a processor, causes the processor to execute the method in the above embodiments.

[0113] It can be understood that the processor in the embodiments of the present application can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor can be a microprocessor or any conventional processor.

[0114] The method steps in the embodiments of the present application can be implemented in the form of hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in a random access memory (RAM), a flash memory, a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically EPROM (EEPROM), a register, a hard disk, a mobile hard disk, a CD-ROM or any other form of storage medium well known in the art. An exemplary storage medium is coupled to the processor, so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and the storage medium can be located in an ASIC.

[0115] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in or transmitted by a computer readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media sets. The available media can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0116] It can be understood that various numerical numbers involved in the embodiments of the present application are only distinguished for convenience of description, and are not used to limit the scope of the embodiments of the present application.

[0117] Those skilled in the art easily understand that the above only describes the preferred embodiments of the present application and is not used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A carbon emission measurement system for an enterprise's production process, characterized by: include: Data acquisition module, data processing module, model calculation module and result display module; The data collection module is used to collect total power consumption data samples of the enterprise's production process and power consumption data samples of key equipment at a collection frequency of minutes as a training data set; The data processing module is used to preprocess the training data set to complete model training; The model calculation module is used to use the trained first neural network to predict and decompose the total electricity consumption data to be estimated to obtain a prediction result, and use the trained second neural network to calculate the real-time carbon emissions based on the prediction result; The result display module is used to determine the equipment performance dynamics and display the equipment performance dynamics and real-time carbon emissions; the equipment performance dynamics include equipment power consumption trends, operating status and prediction errors.

2. The enterprise production process carbon emission measurement system according to claim 1 is characterized in that: The first neural network is an informer neural network, and the second neural network is an MLP neural network; The informer neural network is used to predict and decompose the pre-processed total power consumption data to obtain the power consumption of key equipment and the operating status of the direct emission system; The MLP neural network includes a mapping relationship between equipment power consumption and carbon emissions, and is used to calculate the power consumption of the key equipment through the mapping relationship to obtain the real-time carbon emissions.

3. The enterprise production process carbon emission measurement system according to claim 2 is characterized in that: The informer neural network includes a self-attention subnetwork and a feedforward subnetwork; The self-attention subnetwork is used to calculate the attention weight using the probabilistic sparse self-attention mechanism; The feedforward subnetwork is used to perform linear transformation and nonlinear activation on the preprocessed input data through a fully connected layer.

4. The enterprise production process carbon emission measurement system according to claim 1 is characterized in that: The data processing module includes a timestamp unification submodule, a gap filling submodule and a fitting submodule; The timestamp unification submodule is used to perform unified timestamp processing on the total power consumption data samples and the power consumption data samples of key equipment; The gap filling submodule is used to determine the missing and abnormal data to be completed in the total power consumption data samples and the power consumption data samples of key equipment, and fill the gaps in the data to be completed using the Lagrange interpolation method; The fitting submodule is used to fit the relationship between data points using high-order polynomials.

5. The enterprise production process carbon emission measurement system according to claim 1 is characterized in that: The data acquisition module also includes a state representation submodule and a reference submodule; A state representation submodule is used to represent the operating state of the key equipment using discrete numerical values; the first numerical value represents the shutdown state, the second numerical value includes the partial load state, and the third numerical value includes the full load state; The benchmark submodule is used to collect fuel feed amount, low calorific value, chemical raw material feed amount and carbonate content as the basis for benchmark carbon emission calculation.

6. The enterprise production process carbon emission measurement system according to claim 1 is characterized in that: The data processing module also includes a feature mining submodule and a normalization submodule; The feature mining submodule is used to perform feature mining on total power consumption data samples and power consumption data samples of key equipment, extract trend features, seasonal features and residual features to reflect the inherent laws and change trends of the data; The normalization submodule is used to construct correlation functions and autocorrelation functions to analyze the inherent laws of the data, and uses the minimum and maximum normalization method to transform the data features into the same scale.

7. The enterprise production process carbon emission measurement system according to claim 1 is characterized in that: The result display module includes an early warning submodule, which is used to trigger a threshold early warning mechanism; The early warning mechanism includes primary early warning response, intermediate early warning response and advanced early warning response; The primary warning response pushes SMS notifications, the intermediate warning response generates diagnostic reports, and the advanced warning response links the DCS system to adjust the load.

8. A method for measuring carbon emissions from an enterprise production process implemented based on the enterprise production process carbon emission measurement system according to any one of claims 1 to 7, characterized in that: include: Collect total power consumption data samples of a certain local area of ​​the enterprise and power consumption data samples of key equipment at a frequency of minutes; Preprocessing the training data set to complete model training; Using the trained first neural network to perform predictive decomposition on the estimated total electricity consumption data to obtain a prediction result, and using the trained second neural network to calculate the real-time carbon emissions based on the prediction result; Determine equipment performance dynamics and display the equipment performance dynamics and real-time carbon emissions; the equipment performance dynamics include equipment power consumption trends, operating status and prediction errors.

9. The method for measuring carbon emissions from an enterprise's production process according to claim 8, characterized in that: Also includes: Build a multi-dimensional data collection network using smart meters and sensors on the power supply lines of key equipment, and import historical operating data covering typical operating conditions; Upload data in real time through the IoT architecture, calculate minute-by-minute carbon emissions, and dynamically display them through a visual dashboard; Incremental training is performed on the informer neural network and the MLP neural network based on the first time interval, and device accuracy is verified based on the device accuracy at the second time interval.

10. The method for measuring carbon emissions from an enterprise's production process according to claim 9, characterized in that: The method of using the trained first neural network to predict and decompose the estimated total electricity consumption data to obtain a prediction result, and using the trained second neural network to calculate the real-time carbon emissions based on the prediction result includes: The informer neural network takes pre-processed total power consumption data as input, dynamically extracts the power consumption of key equipment through a probabilistic sparse self-attention mechanism, introduces a distillation operation to compress the output dimension of the attention layer, and outputs the power consumption decomposition results of key equipment and the operating status of the direct emission system; The power consumption, operating status and fuel parameters of key equipment are input into the MLP neural network, and the mapping relationship between equipment power consumption and carbon emissions is fitted to obtain the predicted value of carbon emission intensity.

Images