Carbon emission data anomaly detection system and method based on event triggering

Abstract

The invention discloses a carbon emission data anomaly detection system and method based on event triggering, and the method comprises the steps: S1, collecting the data of a plurality of sensors, and obtaining a carbon emission database; s2, judging whether the sensor is abnormal or not according to the carbon emission database; if yes, obtaining an abnormal sensor and sending an equipment abnormal signal; if not, skipping to S4; s3, predicting a normal value of the abnormal sensor by using a data filling method, and updating the carbon emission database; s4, obtaining the predicted total carbon emission at the current time by using the carbon emission prediction model, and obtaining the actual total carbon emission from the carbon emission database; s5, whether the difference value between the predicted total carbon emission and the actual total carbon emission is larger than a preset difference value threshold value or not is judged; if yes, sending a standard exceeding signal; and if not, sending a normal signal. The invention relates to the technical field of factory carbon emission detection, and solves the technical problems that the data quality accuracy cannot be guaranteed and the data mining of a carbon emission prediction model is single in the existing method.

Description

Technical Field

[0001] The present invention belongs to the field of carbon emission data monitoring, and relates to data filling and anomaly detection technology, and specifically to a carbon emission data anomaly detection system and method based on event triggering. Background Art

[0002] Carbon emission anomaly detection plays a vital role in factory operations. It is not only a key strategy for enterprises to achieve environmentally friendly production and actively respond to the challenges of global climate change, but also an important basis for enhancing brand image and participating in the carbon trading market. First of all, the core value of carbon emission anomaly detection lies in its real-time monitoring and early warning functions. By continuously monitoring carbon emission data during the factory production process, the system can instantly identify and report any abnormal emissions, enabling management to respond quickly and take necessary adjustment measures to effectively curb potential excessive emissions, thereby achieving the goal of energy conservation and emission reduction. It not only helps factories optimize resource allocation and reduce unnecessary energy consumption and emissions, but is also directly related to the shaping of the company's environmental image, and is a direct manifestation of enhancing corporate social responsibility.

[0003] However, despite the importance of carbon emission anomaly detection, existing technologies still face many challenges in practical applications. A notable problem is that current technologies are often limited to isolated analysis of carbon emission data, and fail to fully consider the interference that external factors such as factory equipment failure or data transmission errors may cause to data accuracy. When equipment fails or sensor data is abnormal, carbon emission data may show unrealistic fluctuations, which will directly affect the accuracy and reliability of anomaly detection. In addition, most of the existing carbon emission prediction models are built only based on historical carbon emission data, and the dimensions of data mining are relatively limited. They fail to fully integrate multi-dimensional information such as production processes, equipment status, and energy efficiency, which limits the model's prediction capabilities and the accuracy of anomaly detection.

[0004] For example, the invention patent with publication number CN117150407B discloses "a method for detecting anomalies in industrial carbon emission data". This method performs anomaly detection based on the company's historical carbon emission data, but does not fully consider other actual production and operation data related to carbon emissions. The prediction effect in practical applications is easily divorced from the actual production conditions. For example, when production equipment fails or process parameters change, carbon emission data may be significantly affected, affecting the accuracy and timeliness of anomaly detection. Therefore, it is difficult for a single source of carbon emission data to provide sufficient information to accurately identify anomalies. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a carbon emission data anomaly detection system and method based on event triggering, which is used to solve the data quality problems of the existing carbon emission anomaly detection method caused by equipment failure, human operating errors or other factors, as well as the problem of insufficient prediction accuracy caused by the single data mining of the existing carbon emission prediction model.

[0006] To achieve the above object, a first aspect of the present invention provides a carbon emission data anomaly detection system based on event triggering, comprising:

[0007] Data collection module: used to collect data from several sensors and total carbon emissions using the monitoring center, and store the data from several sensors and total carbon emissions in a database according to sensor ID and time sequence to obtain a carbon emissions database;

[0008] Equipment monitoring module: used to determine whether the sensor is abnormal based on the historical data in the carbon emission database; if yes, obtain the abnormal sensor and abnormal sensor ID, and send an equipment abnormality signal; if no, start the carbon emission detection module;

[0009] Data filling module: used to predict the normal value of the abnormal sensor by using the carbon emission data filling method based on multi-source data fusion, obtain the predicted normal value sequence, and update the carbon emission database according to the abnormal sensor ID;

[0010] Carbon emission detection module: used to obtain the predicted total carbon emissions at the current time using the carbon emission prediction model built based on factory operation data, and obtain the actual total carbon emissions at the current time from the carbon emission database, and determine whether the difference between the predicted total carbon emissions and the actual total carbon emissions is greater than the preset difference threshold; if yes, send an over-standard signal; if not, send a normal signal.

[0011] Furthermore, judging whether the sensor is abnormal based on historical data in the carbon emission database includes:

[0012] A1, based on the sensor ID and the carbon emission database, obtain the carbon emission sequence of a single sensor between the target time T1 and the current time T2, and obtain several carbon emission sequences Ci(t); where t represents time, and t∈[T1,T2], i=1,2,3,…,N, and N represents the number of sensors;

[0013] A3, using the sliding window method to determine whether the sensor is abnormal based on Ci(t).

[0014] Furthermore, the use of the sliding window method to determine whether the sensor is abnormal according to Ci(t) includes:

[0015] A3-1, divide Ci(t) according to the window size w and the sliding step size s to obtain several windows;

[0016] A3-2, for a single window, the correlation coefficient ρij of sensor i and sensor j is calculated according to the formula ρij={∑[(Ci(t')-ACi)×(Cj(t')-ACj)]} / [√∑(Ci(t')-ACi)^2×∑(Cj(t')-ACj)^2]; where Ci(t') and Cj(t') represent the carbon emission sequences of sensors i and j in the current window, respectively, and the summation range is from the start time to the end time of the current window, ACi and ACj represent the average values ​​of the carbon emission data of sensors i and j in the current window, respectively, and i,j=1,2,3,…,N;

[0017] A3-3, construct an n×n similarity matrix S, with element Sij=ρij, and count the number mi of elements in the i-th row of S that are less than a preset correlation threshold;

[0018] A3-4, determine whether mi is greater than or equal to a preset threshold; if yes, mark sensor i as an abnormal sensor, and mark the start time of the current window as the fault time T; if no, jump to A3-5;

[0019] A3-5, determine whether the traversal of several windows has been completed; if yes, mark sensor i as a normal sensor; if no, repeat A3-2 to A3-5.

[0020] Reasonable time window division of carbon emission sequence provides a basis for the subsequent analysis of the changing rules of sensor data within a local time range. In the factory, since carbon emission data is closely linked to order tasks, the carbon emission data monitored by each sensor is related to the total task. Therefore, under normal circumstances, the data collected by each sensor often show significant correlation with each other. By constructing a correlation matrix, it is possible to determine whether a sensor is faulty based on whether the correlation between a certain sensor data and other sensor data deviates from the normal range, thereby timely discovering abnormal sensors and avoiding carbon emission data errors caused by sensor failures, which mislead production task evaluation and environmental protection strategy formulation.

[0021] Furthermore, the method of predicting the normal value of the abnormal sensor by using the carbon emission data filling method based on multi-source data fusion includes:

[0022] B1, using the pre-trained time series analysis model to predict the normal value of the abnormal sensor in [T, T2], and obtain the first prediction value sequence C1;

[0023] B2, using the multivariate regression model to predict the normal value of the abnormal sensor in [T, T2], and obtain the second prediction value sequence C2;

[0024] B3, perform weighted summation of C1 and C2 to obtain the predicted normal value sequence C; wherein the weight coefficient of the weighted summation is determined based on historical experience, and the sum of the weight coefficients of the weighted summation is 1.

[0025] The normal values ​​of abnormal sensors are predicted by using time series analysis model and multiple regression model respectively, and then the weighted sum is used to obtain the final prediction value sequence. The time series analysis model can capture the trend and regularity of data in the time dimension, and the multiple regression model can use the correlation between other normal sensor data and abnormal sensor data. The combination of the two can integrate multiple aspects of information, make up for the limitations of a single model, make the prediction results closer to the true value, and thus improve the accuracy of the prediction of the normal value of the abnormal sensor.

[0026] Furthermore, the construction method of the pre-trained time series analysis model includes:

[0027] B1-11, based on the carbon emission database, obtain the total carbon emission sequence Ct in the normal historical data, and use the stationarity test method to determine whether several carbon emission sequences Ct are stationary; if yes, obtain the stationary Ct; if not, perform a differential operation on Ct to obtain the stationary Ct;

[0028] B1-12, based on the autocorrelation function graph and partial autocorrelation function graph of the stabilized Ct, the order (p, d, q) is obtained, and the ARIMA model is constructed with (p, d, q) to obtain the time series analysis model; where p represents the autoregressive order, q represents the moving average order, and d represents the difference order;

[0029] B1-13, dividing Ct into a first training set, a first validation set and a first test set according to a preset ratio, and iteratively training and validating the time series analysis model using the first training set and the first validation set to obtain an analysis model with the best validation result;

[0030] B1-14, use the first test set to test and verify the analysis model with the best results, and compare whether the test results reach the first test threshold; if yes, obtain the pre-trained time series analysis model; if not, adjust the parameters of the time series analysis and repeat B2 to B4.

[0031] The order of the model is determined by using the autocorrelation function graph and the partial autocorrelation function graph, so as to construct a suitable ARIMA model structure according to the correlation characteristics of the data itself, so that the model can identify the changing pattern of carbon emission data in the time dimension, and discover the correlation strength between the current carbon emission data and the data of the previous few hours or days, and then be used to accurately predict the future trend of carbon emission data.

[0032] Furthermore, the method of using the pre-trained time series analysis model to predict the normal value of the abnormal sensor within [T, T2] includes:

[0033] B1-21, obtain the total carbon emission sequence C of [T1,T] based on the carbon emission database;

[0034] B1-22, input C(t) into the pre-trained time series analysis model to obtain the total carbon emission prediction sequence Pre of [T, T2];

[0035] B1-23, based on the carbon emission database, obtain the abnormal sensor carbon emission sequence C' of [T1, T], and calculate the average proportion p of the abnormal sensor data to the total carbon emission data according to the formula p = [∑C'(a) / C(a)] / m'; where the summation range is the data points from T1 to T in time a, and m' represents the number of total carbon emission data in [T1, T];

[0036] B1-24, multiply p by the elements in Pre to obtain the first prediction value sequence C1.

[0037] In a factory with multiple production links, the carbon emission data monitored by different sensors contribute differently to the overall carbon emissions. By calculating the average proportion p of the data of abnormal sensors in the total carbon emission data and multiplying it with the predicted total carbon emission sequence Pre to obtain the first predicted value sequence C1, the predicted value can be adjusted according to the relative importance of the abnormal sensor in the entire carbon emission system to ensure that the prediction result can accurately reflect the data situation when the sensor is working normally.

[0038] Furthermore, the multivariate regression model is constructed in the following manner:

[0039] B2-11, based on the carbon emission database, obtain the carbon emission sequence data of several sensors in the normal historical data to obtain the historical carbon emission sequence;

[0040] B2-12, construct a multivariate linear regression model: y = α0 + α1x1 + α2x2 + ... + αnxn; where y represents abnormal sensor data, x1, x2, ..., xn represent other normal sensor data, and α0, α1, ..., αn are regression coefficients;

[0041] B2-13, use the regression parameter estimation method to train the multivariate linear regression model of the historical carbon emission series to obtain the estimated value of the regression coefficient;

[0042] B2-14, substitute the estimated values ​​of the regression coefficients into the multiple linear regression model to obtain the multiple regression model.

[0043] In a factory, since the production process is an organic whole, the carbon emission data monitored by different sensors are often correlated. By constructing a multivariate linear regression model, the correlation between abnormal sensor data and other normal sensor data can be quantified.

[0044] Furthermore, the method of using a multivariate regression model to predict a normal value of an abnormal sensor within [T, T2] includes:

[0045] B2-21, based on the carbon emission database, obtain the carbon emission sequences of several other normal sensors in [T, T2] to obtain the normal carbon emission sequence;

[0046] B2-22, input the normal carbon emission sequence into the multivariate regression model to obtain the second prediction value sequence C2.

[0047] Furthermore, the process of constructing the carbon emission prediction model includes:

[0048] C1, collecting historical factory operation data at a preset time interval, and preprocessing the historical factory operation data to obtain preprocessed operation data; wherein the historical factory operation data includes the factory power consumption, the working time of the factory machines, the total driving time of the cars in the factory, the number of workers, the quality of the processed materials, the carbon content coefficient of the processed materials, and the conversion efficiency of the processed materials within the preset time interval;

[0049] C2, annotate the preprocessed operation data according to the total carbon emissions in the corresponding time interval to obtain annotated data, and combine the preprocessed operation data and the annotated data into a data set to obtain a model data set;

[0050] C3, build a neural network model based on the deep learning algorithm to obtain the initial network model;

[0051] C4, dividing the model data set into a second training set, a second validation set, and a second test set according to a preset ratio, and using the second training set and the second validation set to iteratively train and validate the initial network model to obtain a network model with the best validation result;

[0052] C5, use the second test set to test the network model with the best verification result to determine whether the test result reaches the preset second test threshold; if yes, obtain the carbon emission prediction model; if not, adjust the parameters of the initial network model and repeat C3 to C5.

[0053] In the factory operation data, electricity consumption, machine working time, total vehicle driving time, number of workers, quality of processed materials, carbon content coefficient and conversion efficiency are closely related to carbon emissions. High electricity consumption, increased machine working time, longer vehicle driving time and increased number of workers often lead to increased energy consumption and increased carbon emissions; high material quality and high carbon content coefficient will increase carbon emissions during processing, while high material conversion efficiency will help reduce carbon emissions. These data provide key basis for building carbon emission prediction models and help accurately predict carbon emission levels.

[0054] A second aspect of the present invention provides a method for detecting anomalies in carbon emission data based on event triggering, comprising:

[0055] S1, collect data from several sensors, and store the data from several sensors in a database according to sensor ID and time sequence to obtain a carbon emission database;

[0056] S2, judging whether the sensor is abnormal based on the historical data in the carbon emission database; if yes, obtaining the abnormal sensor and abnormal sensor ID, and sending a device abnormal signal; if no, jumping to S4;

[0057] S3, using the carbon emission data filling method based on multi-source data fusion to predict the normal value of the abnormal sensor, obtain the predicted normal value sequence, and update the carbon emission database according to the abnormal sensor ID;

[0058] S4, using a carbon emission prediction model built based on plant operation data to obtain the predicted total carbon emissions at the current time, and obtaining the actual total carbon emissions at the current time from a carbon emission database;

[0059] S5, judging whether the difference between the predicted total carbon emissions and the actual total carbon emissions is greater than a preset difference threshold; if yes, sending an over-limit signal; if no, sending a normal signal.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] Based on the event trigger mechanism, the equipment monitoring module uses the window sliding method and other sophisticated methods to determine whether the sensor is abnormal by using historical data in the carbon emission database. Once the abnormal sensor event is detected, it can quickly and accurately locate the abnormal sensor and its corresponding fault time, send the equipment abnormality signal in time, and trigger the data filling module to work. This event-based instant response method effectively avoids the situation where data errors caused by sensor failures are not detected for a long time, thereby affecting subsequent carbon emission analysis and production and environmental protection decisions, ensuring the reliability of the data of the entire carbon emission monitoring system and improving the timeliness and accuracy of the system in responding to abnormal conditions;

[0062] When a sensor abnormality occurs, the data filling module predicts the normal value of the abnormal sensor based on the carbon emission data filling method based on multi-source data fusion, and uses the comprehensive time series analysis model to capture the time dimension law and the multivariate regression model to take advantage of the correlation between sensors. The weighted summation obtains a prediction result that is closer to the true value, and then updates the carbon emission database. The timely data filling mechanism ensures the integrity and consistency of the carbon emission database, and there will be no data gaps or subsequent analysis deviations caused by abnormal data. It provides a high-quality data foundation for long-term and stable carbon emission monitoring and management, which has significant advantages over existing technologies that are difficult to effectively deal with data missing or abnormal situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 A schematic diagram of a process flow of a carbon emission data anomaly detection system based on event triggering provided by the present invention;

[0065] Figure 2 A schematic diagram of a framework of a carbon emission data anomaly detection system based on event triggering provided by the present invention;

[0066] Figure 3 A flow chart showing the use of the sliding window method in the provided equipment detection module to determine whether a sensor is abnormal based on carbon emission sequence data. DETAILED DESCRIPTION

[0067] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] See also Figure 1-Figure 3 The first embodiment of the present invention provides a carbon emission data anomaly detection system based on event triggering, comprising:

[0069] Data collection module: used to collect data from several sensors and total carbon emissions using the monitoring center, and store the data from several sensors and total carbon emissions in a database according to sensor ID and time sequence to obtain a carbon emissions database;

[0070] Equipment monitoring module: used to determine whether the sensor is abnormal based on the historical data in the carbon emission database; if yes, obtain the abnormal sensor and abnormal sensor ID, and send an equipment abnormality signal; if no, start the carbon emission detection module;

[0071] Data filling module: used to predict the normal value of the abnormal sensor by using the carbon emission data filling method based on multi-source data fusion, obtain the predicted normal value sequence, and update the carbon emission database according to the abnormal sensor ID;

[0072] Carbon emission detection module: used to obtain the predicted total carbon emissions at the current time using the carbon emission prediction model built based on factory operation data, and obtain the actual total carbon emissions at the current time from the carbon emission database, and determine whether the difference between the predicted total carbon emissions and the actual total carbon emissions is greater than the preset difference threshold; if yes, send an over-standard signal; if not, send a normal signal.

[0073] It should be noted that the data collection module, the equipment monitoring module, the data filling module and the carbon emission detection module are in communication connection.

[0074] In the process of carbon emission monitoring, data loss, omission or incompleteness often occur due to equipment failure, human error or other unforeseen factors. For example, the sensor may stop collecting data due to long-term operation failure, or some data may be lost due to network problems during data transmission. These situations make it impossible for carbon emission data to fully and accurately reflect the actual carbon emission status, affecting the subsequent analysis and evaluation of carbon emissions and the formulation of relevant strategies.

[0075] The data collection module of the present invention can collect and store sensor data in an orderly manner. The equipment monitoring module can timely detect abnormal sensors and predict the normal values ​​of abnormal sensors through the data filling module using a multi-source data fusion method, thereby updating the carbon emission database, effectively solving the problem of data loss, omission or incompleteness, ensuring the integrity and accuracy of carbon emission data, and providing a reliable data basis for subsequent carbon emission research, corporate carbon accounting and supervision by environmental protection departments.

[0076] In this embodiment, in order to achieve effective integration and management of carbon emission data and provide a solid data foundation for subsequent equipment monitoring, data filling, carbon emission detection and other links, it is necessary to build a carbon emission database in the data acquisition module.

[0077] From the perspective of data integration, in actual carbon emission monitoring scenarios, there are often sensor data from many different sources and formats, which may include greenhouse gas concentration data, production process data related to carbon emissions, etc. Building a carbon emission database can uniformly store and organize various types of data collected based on sensor ID and time sequence, making it an orderly and easy-to-access data set. For example, in a large industrial plant, there may be multiple sensors for monitoring carbon emissions in different areas, as well as some sensors associated with the operating status of production equipment. These sensors continuously generate massive amounts of data. If there is no database for integration, the data will be in a chaotic and disordered state, making it difficult to effectively analyze and use.

[0078] In the data collection module, the monitoring center will be used to collect data from multiple sensors, and then the data from different sensors will be stored in the database according to the sensor ID and chronological order, and finally a carbon emissions database will be constructed.

[0079] Specifically, suppose that carbon emissions are being monitored in a medium-sized chemical plant. There are multiple different types of production workshops in this plant, and each workshop is equipped with various sensors related to carbon emissions to fully understand the carbon emissions situation.

[0080] First, greenhouse gas concentration sensors are installed in the factory's reaction workshop to monitor the concentration of greenhouse gases such as carbon dioxide (CO2) and methane (CH4) in the air in the workshop in real time. These sensors collect data every 5 minutes and are marked with the corresponding sensor ID (for example, the sensor ID is R1, R2, etc., representing greenhouse gas concentration sensors at different locations) and the collection time;

[0081] Energy consumption-related sensors are distributed in various workshops and the main power distribution room of the factory. For example, the power sensor will record the real-time power consumption of the production equipment in each workshop. Its sensor ID is set to E1, E2, etc., corresponding to the power consumption monitoring of different workshops or equipment groups; there are also steam flow sensors, which are used to monitor the steam flow that provides heat for chemical reactions. The sensor ID is S1, S2, etc. These energy-related sensors collect data at the same time interval and attach the corresponding ID and time information;

[0082] In the material transportation process, on-board sensors are installed on forklifts, trucks and other vehicles used to transport raw materials and finished products in the factory to record vehicle mileage, fuel consumption and other data. The sensor ID is set to V1, V2, etc., and the specific data collection timestamp is marked;

[0083] All these sensors of different locations and types transmit the collected data to the factory's monitoring center in real time through wired (such as network cables connected to the factory's local area network) or wireless (such as wireless communication protocols such as Zigbee and LoRa). The monitoring center is equipped with a special data receiving server, which runs the corresponding data acquisition software. It can identify and receive the data sent from each sensor and perform preliminary sorting according to the sensor ID and time sequence;

[0084] Taking the relational database MySQL as an example, the monitoring center will process and store the data of several sensors as follows:

[0085] First, create a master data table named "carbon_emission_data", which contains but is not limited to the following main fields:

[0086] (1) The "sensor_id" field is used to store the unique identifier of the sensor, such as R1, E1, V1, etc. The data type can be set to VARCHAR(20) to accommodate ID representations of different lengths;

[0087] (2) The "measurement_time" field is used to record the time when data was collected. The data type is DATETIME and is accurate to the specific year, month, day, hour, minute, and second to ensure that the time series characteristics of the data can be accurately reflected;

[0088] (3) The "measurement_value" field stores the value actually collected by the sensor. The data type can be flexibly set according to the data type measured by the sensor. For example, for the concentration value collected by the greenhouse gas concentration sensor, it can be set to DECIMAL (10, 4) to accurately represent the decimal value; for the power value collected by the power consumption sensor, it can be set to INT (indicating integer type, and the appropriate length can be reasonably set according to the actual power consumption range of the factory);

[0089] (4) The "sensor_type" field is used to describe the type of sensor, such as 'CO2_concentration' (carbon dioxide concentration sensor), 'electricity_consumption' (electricity consumption sensor), 'vehicle_fuel_consumption' (vehicle fuel consumption sensor), etc. The data type is set to VARCHAR(50);

[0090] Assume that at a certain moment, the data collected by the carbon dioxide concentration sensor R1 in the reaction workshop at 10:15 on November 15, 2024 is 0.0568% (concentration value), then a new record will be added to the "carbon_emission_data" table, the "sensor_id" field value is "R1", the "measurement_time" field value is "2024-11-1510:15:00", the "measurement_value" field value is "0.0568", and the "sensor_type" field value is "CO2_concentration";

[0091] For example, if the power consumption collected by the power sensor E1 in workshop 1 at the same time is 500 kWh, a new record will be added to the database according to the above structure, except that the "sensor_id" is "E1", "measurement_value" is "500", and "sensor_type" is "electricity_consumption", etc.

[0092] In this way, the various sensor data collected by the data acquisition module are stored in the carbon emission database in an orderly manner, providing a comprehensive and organized data foundation for the subsequent equipment monitoring, data filling, and carbon emission detection functions of the entire carbon emission monitoring system;

[0093] Of course, in addition to relational databases, non-relational databases (such as MongoDB in NoSQL databases) can also be used to store data according to actual needs and project characteristics. The storage structure and data organization method will be different, but the core purpose is to effectively manage carbon emission-related data according to sensor ID and time sequence.

[0094] In the equipment monitoring module, whether the sensor is abnormal will be judged based on the historical data in the carbon emission database. Once an abnormal sensor is found, an abnormal signal will be immediately sent to the staff, and the next data filling module will be started at the same time to compensate for the data errors caused by the sensor failure, and minimize the analysis bias caused by data errors and the risk of subsequent decision-making errors.

[0095] Specifically, first, the carbon emission sequence of a single sensor between the target time T1 and the current time T2 is extracted from the carbon emission database according to the sensor ID to obtain several carbon emission sequences Ci(t), which record the changes in carbon emission data during this time period.

[0096] Then, Ci(t) is divided into multiple windows according to the set window size w and sliding step size s. For a single window, the correlation coefficient ρ between sensor i and sensor j is calculated according to the formula ρij={∑[(Ci(t')-ACi)×(Cj(t')-ACj)]} / [√∑(Ci(t')-ACi)^2×∑(Cj(t')-ACj)^2], where ACi and ACj are the average values ​​of carbon emission data of sensors i and j in the current window, respectively, and the summation range is from the start time to the end time of the current window, that is, the value range of t' in the current window.

[0097] Next, an n×n similarity matrix S is constructed, whose elements Sij correspond to the calculated correlation coefficients ρij, and the number mi of elements in the i-th row of S that are less than the preset correlation threshold is obtained by statistics. After that, mi is compared with the preset threshold. If they are equal, sensor i is marked as an abnormal sensor and the start time of the current window is marked as the fault time T; if they are not equal, jump to the next step.

[0098] Finally, repeat the above operation process to determine whether all divided windows have been traversed. If the traversal is completed, sensor i is marked as a normal sensor. Otherwise, continue to analyze and judge the next window until all windows are detected and an accurate conclusion of the sensor status is obtained, providing a reliable basis for subsequent links.

[0099] The following is an example of how the equipment monitoring module uses the sliding window method to identify abnormal sensors:

[0100] Assume that there is a factory that has installed 5 carbon emission sensors (labeled as sensor 1, sensor 2, sensor 3, sensor 4, and sensor 5) in its production area to monitor carbon emissions in different production links. Set the target time T1 to the past 10 hours (i.e., 10 hours from the current time T2), with 1 hour as a data collection point, so that each sensor has a carbon emission sequence of 10 data points;

[0101] Extract the carbon emission data of each sensor in the past 10 hours (T1 to T2) from the carbon emission database, for example:

[0102] Sensor 1C1(t): [2.5, 2.6, 2.4, 2.7, 2.3, 2.8, 2.2, 2.9, 2.1, 3.0];

[0103] Sensor 2C2(t): [1.8, 1.9, 1.7, 2.0, 1.6, 2.1, 1.5, 2.2, 1.4, 2.3];

[0104] Sensor 3C3(t): [3.5, 3.6, 3.4, 3.7, 3.3, 3.8, 3.2, 3.9, 3.1, 4.0];

[0105] Sensor 4C4(t): [2.2, 2.3, 2.1, 2.4, 2.0, 2.5, 1.9, 2.6, 1.8, 2.7];

[0106] C5(t) of sensor 5: [4.5, 4.6, 4.4, 4.7, 4.3, 4.8, 4.2, 4.9, 4.1, 5.0];

[0107] Set the window size w=3 (i.e. every 3 data points as a window) and the sliding step size s=1 (moving 1 data point each time);

[0108] Taking sensor 1 as an example, its divided windows are as follows:

[0109] Window 1: [2.5, 2.6, 2.4]

[0110] Window 2: [2.6, 2.4, 2.7]

[0111] Window 3: [2.4, 2.7, 2.3] ...

[0113] Window 8: [2.2, 2.9, 2.1]

[0114] Next, for window 1, calculate the correlation coefficients between sensor 1 and other sensors: first calculate the average value of sensor 1 in window 1, AC1 = (2.5 + 2.6 + 2.4) / 3 = 2.5, and the average value of sensor 2 in window 1, AC2 = (1.8 + 1.9 + 1.7) / 3 = 1.8;

[0115] Calculate the correlation coefficient according to the formula:

[0116] ρ12={∑[(C1(t')-AC1)×(C2(t')-AC2)]} / [√∑(C1(t')-AC1)^2×∑(C2(t')-A2j)^2]= {[(2.5-2.5)×(1.8-1.8)]+[(2.6-2.5)×(1.9-1.8)]+[(2.4-2.5)×(1.7-1.8)]} / {√[ (2.5-2.5)^2+(2.6-2.5)^2+(2.4-2.5)^2]×[(1.8-1.8)^2+(1.9-1.8)^2+(1.7-1.8)^2]}=(0+0.01+0.01) / {√(0+0.01+0.01)×(0+0.01+0.01)}=0.02 / √(0.02×0.02)=1;

[0117] Similarly, the correlation coefficients ρ13, ρ14, ρ15 between sensor 1 and sensor 3, sensor 4, and sensor 5 in window 1 are calculated;

[0118] Construct a 5×5 similarity matrix S. For example, after the calculation of window 1 is completed, the first row elements of the matrix S are: S11=1, S12=ρ12, S13=ρ13, S14=ρ14, S15=ρ15. Then, the similarities of other sensors with the other four sensors in window 1 are calculated in turn to obtain the complete similarity matrix S.

[0119] After obtaining the complete similarity matrix S, we start to count the number of elements in each row of S that are less than the preset correlation threshold. Assuming the preset correlation threshold is 0.75 and the preset threshold is 3, for the first row of the matrix S, we check and count the number of elements less than 0.,75 element by element, denoted as m1. Then, we perform the same operation on the second, third, fourth, and fifth rows, respectively, to obtain m2, m3, m4, and m5.

[0120] Next, m1 to m5 are compared with the preset threshold 2. If the value of one of the mi is greater than or equal to the preset threshold, for example, m3 = 3, the corresponding sensor 3 is marked as an abnormal sensor, and the start time of the current window 1 is marked as the fault time T of sensor 3;

[0121] If mi is not equal to the preset threshold, the above steps are repeated until all windows are traversed, and the final state of each sensor is determined. If a sensor is not marked as abnormal during all window traversals, it is marked as a normal sensor.

[0122] If an abnormal sensor is found in the equipment monitoring module, the ID of the abnormal sensor and the fault time T are sent to the data filling module, and the normal value of the abnormal sensor is predicted using the carbon emission data filling method based on multi-source data fusion, and then the carbon emission database is updated according to the abnormal sensor ID;

[0123] Specifically, the carbon emission data filling method based on multi-source data fusion includes time series analysis model and multivariate regression model:

[0124] First, the pre-trained time series analysis model is used to predict the normal value of the abnormal sensor in [T, T2] to obtain the first predicted value sequence C1; then the multivariate regression model is used to predict the normal value of the abnormal sensor in [T, T2] to obtain the second predicted value sequence C2; ​​then the predicted normal value sequence C is obtained according to the weighted summation formula: C = λ1×C1+λ2×C2; wherein the weight coefficients λ1 and λ2 of the weighted summation are determined based on historical experience, and λ1+λ2=1; finally, the predicted normal value sequence is updated to the carbon emission database according to the abnormal sensor ID;

[0125] Carbon emission data is interactively affected by many factors in the factory environment, and its changing patterns are complex. It is difficult for a single data prediction method to fully and accurately grasp the data characteristics. For example, changes in order tasks will not only affect carbon emission data in the time series, but also have associated effects due to changes in sensor data corresponding to different production links. The time series model mainly focuses on the changing patterns of data in the time dimension, while the multivariate regression model focuses on utilizing the spatial correlation between sensors. By combining the prediction results of these two models (such as subsequent weighted summation), the normal values ​​of abnormal sensors can be predicted more comprehensively and accurately. This is crucial to ensuring the quality of carbon emission data, because accurate prediction values ​​can better fill in the missing data due to sensor anomalies and avoid carbon emission analysis errors caused by inaccurate data.

[0126] Among them, the construction methods of the pre-trained time series analysis model include:

[0127] B1-11, Stationarity test and processing: Extract the total carbon emission sequence Ct in the normal historical data from the carbon emission database, and use the stationarity test method (such as the commonly used unit root test method such as ADF test, etc.) to judge the stationarity of these carbon emission sequences Ct one by one. If a Ct is judged to be in a stationary state through the test, then the Ct is directly used as the stable Ct for subsequent use; if a Ct is judged to be non-stationary, perform a difference operation on it, such as first-order difference or multi-order difference, until it becomes a stationary sequence, so as to obtain the stable Ct;

[0128] B1-12, determine the model order and build the ARIMA model: for the obtained stable Ct, draw its autocorrelation function (ACF) and partial autocorrelation function (PreACF). By observing the characteristics of these two graphs, analyzing the tailing and truncation of the autocorrelation and partial autocorrelation coefficients, the information criterion method can also be used for further analysis to determine the order of the ARIMA model (p, d, q), where p represents the autoregressive order, q represents the moving average order, and d is the difference order determined by the previous difference operation. Construct the ARIMA model based on the determined order (p, d, q) to obtain the initial time series analysis model framework;

[0129] B1-13, model training and verification: the total carbon emission sequence Ct is divided into the first training set, the first verification set and the first test set according to the preset ratio (for example, the common 70% as the training set, 20% as the verification set, and 10% as the test set. The specific ratio can be adjusted according to the actual data volume and situation). The data of the first training set is input into the time series analysis model for training. During the training process, the model parameters (such as the coefficient of the autoregressive part, the coefficient of the moving average part, etc.) are continuously adjusted to gradually reduce the prediction error of the model on the training set (such as the mean square error, mean absolute error and other indicators). At the same time, in each round of training, the model is verified using the data of the first verification set to observe the performance of the model on the verification set (also measured by the above-mentioned similar error indicators). According to the verification results, it is determined whether overfitting or underfitting occurs, so as to adjust the training strategy, continuously iterate the training, and finally obtain the analysis model with the best verification result, that is, a model with good fitting effect on the training set and relatively strong generalization ability on the verification set;

[0130] B1-14, model testing: Use the data of the first test set to test the analysis model with the best verification result, compare the prediction results of the model with the actual carbon emission data in the test set, calculate the corresponding error index, and compare it with the pre-set first test threshold (for example, set the mean square error to be less than a certain specific value). If the performance of the model on the test set meets the requirements of the first test threshold, then it is determined that the pre-trained time series analysis model has been obtained, which can be used to predict the normal value of the abnormal sensor in the future; if the threshold requirements are not met, it is necessary to return to adjust the relevant parameters of the time series analysis model and repeat the above steps from B1-13 to B1-14 until a pre-trained model that meets the requirements is obtained.

[0131] After obtaining the pre-trained time series analysis model, we can use the model to predict the normal value of the abnormal sensor in [T, T2]:

[0132] First, the total carbon emission sequence C before the fault [T1, T] is extracted from the carbon emission database; then the obtained C(t) data is sequentially input into the pre-trained time series analysis model, and the total carbon emission prediction sequence Pre in the time interval [T, T2] is obtained through the calculation and prediction mechanism within the model. This sequence represents the total carbon emission in the future period predicted based on the historical time series law;

[0133] Then obtain the abnormal sensor carbon emission sequence C' in the same time range [T1, T] from the carbon emission database, and calculate the average proportion p of the abnormal sensor data to the total carbon emission data according to the formula p = [∑C'(a) / C(a)] / m'. Specifically, the abnormal sensor carbon emission value C'(a) and the total carbon emission value C(a) corresponding to the data points from time a from T1 to T are divided point by point and summed, and then divided by the number of total carbon emission data in this time period m', so as to calculate the proportion p, which can reflect the importance of the abnormal sensor relative to the overall carbon emissions in the past period of time;

[0134] Finally, the calculated average proportion p of abnormal sensor data in the total carbon emission data is multiplied one by one with each element in the total carbon emission prediction sequence Pre to obtain the first prediction value sequence C1. This sequence is the predicted normal value of the abnormal sensors in [T, T2] predicted based on the time series analysis model. It takes into account the historical proportion of abnormal sensors in the overall carbon emissions, making the prediction result more realistic.

[0135] The process of building a multiple regression model includes:

[0136] B2-11, obtain historical carbon emission sequence data: extract the carbon emission sequence data of several sensors in the normal historical data from the carbon emission database, and integrate them to form a historical carbon emission sequence. These data cover the carbon emission records of multiple normal sensors at different times, which is the basic data source for building a multivariate regression model and is conducive to mining the correlation between different sensors;

[0137] B2-12, construct a multiple linear regression model: The multiple linear regression model is constructed as follows: y = α0 + α1x1 + α2x2 + ... + αnxn; where y represents abnormal sensor data, x1, x2, ..., xn represent other normal sensor data, and α0, α1, ..., αn are regression coefficients;

[0138] B2-13, train the model to obtain the estimated value of the regression coefficient: use the appropriate regression parameter estimation method (such as the least square method and other common methods) to train the multivariate linear regression model for the acquired historical carbon emission sequence. By minimizing the objective function such as the sum of squared errors between the predicted value and the true value, the optimal estimated value of the regression coefficient is determined, so that the model can fit the relationship between abnormal sensor data and other normal sensor data as well as possible on the historical data, and obtain the estimated value of the regression coefficient;

[0139] B2-14, determine the multiple regression model: substitute the estimated values ​​of the regression coefficients obtained through training into the expression of the multiple linear regression model to determine the final multiple regression model;

[0140] Then, the multivariate regression model is used to predict the normal value of the abnormal sensor in [T, T2]:

[0141] First, based on the carbon emission database, the carbon emission sequences of several other normal sensors in the time interval [T, T2] are obtained to obtain the normal carbon emission sequence. These data represent the actual carbon emission of other normal sensors in the future time period when the abnormal sensor fails, which is the input data for the multivariate regression model to make predictions.

[0142] Then, the acquired normal carbon emission sequence data is input into the determined multiple regression model. Through the linear operation inside the model, according to the relationship between the regression coefficient obtained by previous training and each normal sensor data, the second prediction value sequence C2 is calculated. This sequence is another part of the predicted normal value of the abnormal sensor in [T, T2] predicted by the multiple regression model. It provides a reference for the prediction of abnormal sensors from the perspective of the correlation between different sensors.

[0143] Finally, according to the weighted summation formula, the predicted normal value sequence of the data filling module is obtained.

[0144] In the carbon emission detection module, the real-time total carbon emission in the carbon emission database is received, and then compared with the predicted total obtained by the carbon emission prediction model. If the difference between the two is greater than the preset difference threshold, an over-standard signal will be sent to reflect that the current carbon emission situation exceeds the expected range, alerting relevant personnel or systems to pay attention and take countermeasures in time; at the same time, the over-standard signal in this system can trigger the production adjustment mechanism within the factory, such as prompting the production department to reduce the operating power of high-energy-consuming equipment, optimize the production process to reduce unnecessary carbon emission links, or reallocate the energy supply system, give priority to clean energy or improve energy utilization efficiency, etc., so as to control the carbon emission level within a reasonable range as soon as possible;

[0145] Among them, the carbon emission prediction model is built based on deep learning technology, and its construction process is as follows:

[0146] C1, collect historical factory operation data from various relevant systems and monitoring equipment in the factory at pre-set time intervals, including: obtaining factory power consumption data from the factory's power metering system, collecting the working hours of factory machines through the operation time recording device installed on each machine, using the vehicle positioning and statistical system to count the total driving time of cars in the factory, extracting the number of workers from the personnel management system, relying on the production management system to record the quality of processed materials, and also obtaining data such as the carbon content coefficient of processed materials and the conversion efficiency of processed materials from raw material composition testing and production process documents, so as to accumulate a rich and multi-dimensional historical factory operation data set; and pre-processing the historical factory operation data to obtain pre-processed operation data;

[0147] C2, according to the total carbon emission information in the corresponding time interval, each set of pre-processed operation data is labeled to obtain labeled data, and each set of pre-processed operation data after labeling is combined with the corresponding labeled data one by one to form a model data set;

[0148] C3, build a neural network model based on the deep learning algorithm to obtain the initial network model; for example, if there is a more complex nonlinear relationship between the data and multi-dimensional features need to be comprehensively processed, then you can choose a classic feedforward neural network architecture such as a multi-layer perceptron (MLPre), which can effectively fit complex functional relationships; or consider using a long short-term memory network (LSTM). If the factory operation data has obvious time series characteristics (for example, the time-varying laws of factory electricity consumption and machine working hours in different time periods have an impact on carbon emissions), LSTM can well capture the long-term and short-term dependencies in the data, thereby better predicting carbon emissions;

[0149] C4, divide the model data set into the second training set, the second validation set and the second test set according to the preset ratio, and use the second training set and the second validation set to iteratively train and validate the initial network model. With the increase of training rounds, the degree of fit of the model on the training set will gradually improve. At the same time, the training strategy is continuously adjusted through the monitoring of the validation set, and finally the network model with the best verification result is found, that is, the model with the evaluation index reaching a better level on the validation set and no obvious overfitting phenomenon. This model has achieved a good balance between generalization ability and the ability to fit the training data.

[0150] C5, use the second test set to test the network model with the best verification result to determine whether the test result reaches the preset second test threshold; if yes, obtain the carbon emission prediction model; if not, adjust the parameters of the initial network model and repeat C3 to C5.

[0151] Unlike the common models that are only built based on historical carbon emission data, the carbon emission prediction model in this system starts from the factory's operating data and deeply explores the intrinsic connection between carbon emissions and various factory operating links. Through multi-dimensional operating data such as factory electricity consumption, machine working time, total vehicle driving time, number of workers, quality of processed materials, carbon content coefficient of processed materials, and conversion efficiency of processed materials, it is possible to capture the potential influencing factors of carbon emissions and their complex interactions more comprehensively and accurately. For example, fluctuations in factory electricity consumption reflect changes in the operating intensity of production equipment, which is closely related to carbon emissions; the quality, carbon content coefficient and conversion efficiency of processed materials are directly related to the generation of carbon emissions during the material processing process. This multi-factor comprehensive consideration makes the model prediction results more accurate and reliable, and effectively avoids the one-sidedness and limitations caused by a single factor or relying solely on historical carbon emission data.

[0152] A second embodiment of the present invention provides a method for detecting anomalies in carbon emission data based on event triggering, comprising:

[0153] S1, collect data from several sensors, and store the data from several sensors in a database according to sensor ID and time sequence to obtain a carbon emission database;

[0154] S2, judging whether the sensor is abnormal based on the historical data in the carbon emission database; if yes, obtaining the abnormal sensor and abnormal sensor ID, and sending a device abnormal signal; if no, jumping to S4;

[0155] S3, using the carbon emission data filling method based on multi-source data fusion to predict the normal value of the abnormal sensor, obtain the predicted normal value sequence, and update the carbon emission database according to the abnormal sensor ID;

[0156] S4, using a carbon emission prediction model built based on plant operation data to obtain the predicted total carbon emissions at the current time, and obtaining the actual total carbon emissions at the current time from a carbon emission database;

[0157] S5, judging whether the difference between the predicted total carbon emissions and the actual total carbon emissions is greater than a preset difference threshold; if yes, sending an over-limit signal; if no, sending a normal signal.

[0158] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0159] Working principle of the present invention:

[0160] Through the data collection module, the monitoring center collects data from several sensors and total carbon emissions, and then stores these data in the database according to the sensor ID and time sequence to build a carbon emissions database, providing a data basis for subsequent links;

[0161] Through the equipment detection module, the carbon emission sequence of a single sensor in a specific time period is extracted based on the sensor ID and the carbon emission database. The window sliding method is used to calculate the correlation coefficient between sensors based on the data in the window, and a similarity matrix is ​​constructed. The sensor is judged to be abnormal by comparing the statistical matrix elements with the preset correlation threshold.

[0162] In the data filling module, the pre-trained time series analysis model and multivariate regression model are used to predict the normal value of the abnormal sensor within a specific time. The predicted value sequences obtained by the two are weighted summed to obtain the predicted normal value sequence, and the carbon emission database is updated according to the predicted normal value sequence and the fault time.

[0163] In the carbon emission detection module, the carbon emission prediction model built based on the factory operation data obtains the predicted total carbon emissions at the current time, and at the same time obtains the actual total carbon emissions at the current time from the carbon emission database. The difference between the two is compared, and an over-standard signal or a normal signal is sent based on the comparison result between the difference and the preset difference threshold.

[0164] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A carbon emission data anomaly detection system based on event triggering, characterized in that: include: Data collection module: used to collect data from several sensors and total carbon emissions using the monitoring center, and store the data from several sensors and total carbon emissions in a database according to sensor ID and time sequence to obtain a carbon emissions database; Equipment monitoring module: used to determine whether the sensor is abnormal based on the historical data in the carbon emission database; If yes, the abnormal sensor and abnormal sensor ID are obtained, and the device abnormal signal is sent; If not, the carbon emission detection module is started; Data filling module: used to predict the normal value of the abnormal sensor by using the carbon emission data filling method based on multi-source data fusion, obtain the predicted normal value sequence, and update the carbon emission database according to the abnormal sensor ID; Carbon emission detection module: used to obtain the predicted total carbon emissions at the current time using the carbon emission prediction model, and obtain the actual total carbon emissions at the current time from the carbon emission database, and determine whether the difference between the predicted total carbon emissions and the actual total carbon emissions is greater than a preset difference threshold; if yes, an over-limit signal is sent; If not, a normal signal is sent.

2. The carbon emission data anomaly detection system based on event triggering according to claim 1 is characterized in that: Judging whether the sensor is abnormal according to the historical data in the carbon emission database includes: A1, based on the sensor ID and the carbon emission database, obtain the carbon emission sequence of a single sensor between the target time T1 and the current time T2, and obtain several carbon emission sequences Ci(t); where t represents time, and t∈[T1,T2], i=1,2,3,…,N, and N represents the number of sensors; A3, using the sliding window method to determine whether the sensor is abnormal based on Ci(t).

3. The carbon emission data anomaly detection system based on event triggering according to claim 2 is characterized in that: The method of using the sliding window method to judge whether the sensor is abnormal according to Ci(t) includes: A3-1, divide Ci(t) according to the window size w and the sliding step size s to obtain several windows; A3-2, for a single window, the correlation coefficient ρij of sensor i and sensor j is calculated according to the formula ρij={∑[(Ci(t')-ACi)×(Cj(t')-ACj)]} / [√∑(Ci(t')-ACi)^2×∑(Cj(t')-ACj)^2]; where Ci(t') and Cj(t') represent the carbon emission sequences of sensors i and j in the current window, respectively, and the summation range is from the start time to the end time of the current window, ACi and ACj represent the average values ​​of the carbon emission data of sensors i and j in the current window, respectively, and i,j=1,2,3,…,N; A3-3, construct an n×n similarity matrix S, with element Sij=ρij, and count the number mi of elements in the i-th row of S that are less than a preset correlation threshold; A3-4, determine whether mi is greater than or equal to a preset threshold; if yes, mark sensor i as an abnormal sensor, and mark the start time of the current window as the fault time T; if no, jump to A3-5; A3-5, determine whether the traversal of several windows has been completed; if yes, mark sensor i as a normal sensor; if no, repeat A3-2 to A3-5.

4. The carbon emission data anomaly detection system based on event triggering according to claim 1 is characterized in that: The method of using the carbon emission data filling method based on multi-source data fusion to predict the normal value of the abnormal sensor includes: B1, using the pre-trained time series analysis model to predict the normal value of the abnormal sensor in [T, T2], and obtain the first prediction value sequence C1; B2, using the multivariate regression model to predict the normal value of the abnormal sensor in [T, T2], and obtain the second prediction value sequence C2; B3, perform weighted summation of C1 and C2 to obtain the predicted normal value sequence C.

5. The carbon emission data anomaly detection system based on event triggering according to claim 4 is characterized in that: The construction method of the pre-trained time series analysis model includes: B1-11, based on the carbon emission database, obtain the total carbon emission sequence Ct in the normal historical data, and use the stationarity test method to determine whether several carbon emission sequences Ct are stationary; if yes, obtain the stationary Ct; if not, perform a differential operation on Ct to obtain the stationary Ct; B1-12, based on the autocorrelation function graph and partial autocorrelation function graph of the stabilized Ct, the order (p, d, q) is obtained, and the ARIMA model is constructed with (p, d, q) to obtain the time series analysis model; where p represents the autoregressive order, q represents the moving average order, and d represents the difference order; B1-13, dividing Ct into a first training set, a first validation set and a first test set according to a preset ratio, and iteratively training and validating the time series analysis model using the first training set and the first validation set to obtain an analysis model with the best validation result; B1-14, use the first test set to test and verify the analysis model with the best results, and compare whether the test results reach the first test threshold; if yes, obtain the pre-trained time series analysis model; if not, adjust the parameters of the time series analysis and repeat B2 to B4.

6. The carbon emission data anomaly detection system based on event triggering according to claim 4 is characterized in that: The method of using the pre-trained time series analysis model to predict the normal value of the abnormal sensor within [T, T2] includes: B1-21, obtain the total carbon emission sequence C of [T1,T] based on the carbon emission database; B1-22, input C(t) into the pre-trained time series analysis model to obtain the total carbon emission prediction sequence Pre of [T, T2]; B1-23, based on the carbon emission database, obtain the abnormal sensor carbon emission sequence C' of [T1, T], and calculate the average proportion p of the abnormal sensor data to the total carbon emission data according to the formula p = [∑C'(a) / C(a)] / m'; where the summation range is the data points from T1 to T in time a, and m' represents the number of total carbon emission data in [T1, T]; B1-24, multiply p by the elements in Pre to obtain the first prediction value sequence C1.

7. The carbon emission data anomaly detection system based on event triggering according to claim 4 is characterized in that: The construction method of the multivariate regression model includes: B2-11, based on the carbon emission database, obtain the carbon emission sequence data of several sensors in the normal historical data to obtain the historical carbon emission sequence; B2-12, construct a multivariate linear regression model: y = α0 + α1x1 + α2x2 + ... + αnxn; where y represents abnormal sensor data, x1, x2, ..., xn represent other normal sensor data, and α0, α1, ..., αn are regression coefficients; B2-13, use the regression parameter estimation method to train the multivariate linear regression model of the historical carbon emission series to obtain the estimated value of the regression coefficient; B2-14, substitute the estimated values ​​of the regression coefficients into the multiple linear regression model to obtain the multiple regression model.

8. The carbon emission data anomaly detection system based on event triggering according to claim 4 is characterized in that: The method of using a multivariate regression model to predict a normal value of an abnormal sensor within [T, T2] includes: B2-21, based on the carbon emission database, obtain the carbon emission sequences of several other normal sensors in [T, T2] to obtain the normal carbon emission sequence; B2-22, input the normal carbon emission sequence into the multivariate regression model to obtain the second prediction value sequence C2.

9. The carbon emission data anomaly detection system based on event triggering according to claim 1 is characterized in that: The process of constructing the carbon emission prediction model includes: C1, collecting historical factory operation data at a preset time interval, and preprocessing the historical factory operation data to obtain preprocessed operation data; wherein the historical factory operation data includes the factory power consumption, the working time of the factory machines, the total driving time of the cars in the factory, the number of workers, the quality of the processed materials, the carbon content coefficient of the processed materials, and the conversion efficiency of the processed materials within the preset time interval; C2, annotate the preprocessed operation data according to the total carbon emissions in the corresponding time interval to obtain annotated data, and combine the preprocessed operation data and the annotated data into a data set to obtain a model data set; C3, build a neural network model based on the deep learning algorithm to obtain the initial network model; C4, dividing the model data set into a second training set, a second validation set, and a second test set according to a preset ratio, and using the second training set and the second validation set to iteratively train and validate the initial network model to obtain a network model with the best validation result; C5, use the second test set to test the network model with the best verification result to determine whether the test result reaches the preset second test threshold; if yes, obtain the carbon emission prediction model; if not, adjust the parameters of the initial network model and repeat C3 to C5.

10. A method for detecting anomalies in carbon emission data based on event triggering, applied to a system for detecting anomalies in carbon emission data based on event triggering according to any one of claims 1 to 9, characterized in that: include: S1, collect data from several sensors, and store the data from several sensors in a database according to sensor ID and time sequence to obtain a carbon emission database; S2, judging whether the sensor is abnormal based on the historical data in the carbon emission database; If yes, the abnormal sensor and abnormal sensor ID are obtained, and a device abnormality signal is sent; if no, the process jumps to S4; S3, using the carbon emission data filling method based on multi-source data fusion to predict the normal value of the abnormal sensor, obtain the predicted normal value sequence, and update the carbon emission database according to the abnormal sensor ID; S4, using a carbon emission prediction model built based on plant operation data to obtain the predicted total carbon emissions at the current time, and obtaining the actual total carbon emissions at the current time from a carbon emission database; S5, judging whether the difference between the predicted total carbon emissions and the actual total carbon emissions is greater than a preset difference threshold; if yes, sending an over-limit signal; if no, sending a normal signal.

Images