A data intelligent analysis method and system applied to carbon emission monitoring

Abstract

The application provides a data intelligent analysis method and system applied to carbon emission monitoring, and comprises the following method: a dynamic non-negative row vector is obtained by executing an NMF load decoupling algorithm, the dynamic non-negative row vector comprises a resident coefficient and an industrial coefficient, carbon emission intensities of residents and industries corresponding to a current period are calculated according to the non-negative row vector and a space-time database, and are updated to the space-time database; future carbon emission prediction values corresponding to future different periods are calculated and generated according to the space-time database; and abnormality detection calculation is performed according to the carbon emission intensity and the future carbon emission prediction values corresponding to the current period. By using the above method, the carbon emission intensity situation of the whole region can be indirectly reflected intuitively, in real time and in a non-falsifiable manner.

Description

Technical Field

[0001] This invention relates to carbon emission monitoring, specifically to a data intelligent analysis method and system for carbon emission monitoring. Background Technology

[0002] Accurate carbon emission monitoring is a crucial prerequisite for achieving emission reduction targets. Currently, mainstream monitoring technologies mainly rely on two methods: direct monitoring based on sensor deployment and algorithmic methods based on activity data and emission factors. However, both methods have significant limitations and are insufficient to meet the needs of regional-level dynamic carbon emission tracing and anomaly identification.

[0003] For the first type, direct monitoring based on sensor deployment directly measures atmospheric carbon emission concentrations by installing carbon dioxide concentration sensors in the monitored area. Although sensor technology is relatively mature, its practical application faces two major bottlenecks: First, carbon emission diffusion paths are severely affected by meteorological conditions; factors such as wind speed, wind direction, and temperature cause dynamic changes in concentration distribution, making it difficult for sensor data to accurately reflect the true emission sources. Second, the massive number of sensors required leads to high deployment costs and difficulties in optimizing their locations. For areas with fugitive emissions or complex terrain, ideal deployment conditions are difficult to meet, resulting in insufficient data representativeness.

[0004] For the second type, the indirect estimation method based on material accounting relies on the material consumption, production process data, and emission factors reported by enterprises to calculate carbon emissions using formulas. While this method is simple to operate, its core problem lies in the susceptibility of data manipulation: under carbon market quota constraints, enterprises may conceal energy consumption data or falsify reports to reduce compliance costs, resulting in calculated carbon emissions far exceeding actual emissions. Furthermore, the self-monitored factors by enterprises lack a verification mechanism.

[0005] Neither of the existing two technological approaches has solved the practical problem of carbon emission monitoring, especially in areas with mixed residential and industrial populations, where carbon emission monitoring is even more challenging. Sensor-based methods are limited by environmental diffusion and cost, making it difficult to achieve full-area coverage; while kernel-based methods, due to human factors and data lag, cannot meet the needs for real-time monitoring and anomaly early warning.

[0006] Therefore, there is an urgent need for a data intelligence analysis method and system that can effectively monitor areas where residents and industries mix, and achieve more accurate and rapid early warning without installing a large number of sensors. Summary of the Invention

[0007] To overcome the existing technical problems, this invention provides a data intelligent analysis method and system for carbon emission monitoring.

[0008] The present invention adopts the following technical solution.

[0009] A data intelligent analysis method for carbon emission monitoring includes the following methods:

[0010] Acquire total load data, meteorological data, calendar attribute data, location data, region type data, and carbon emission activity data for each region, and integrate them into the spatiotemporal database of each region based on the corresponding timestamps;

[0011] The industrial activity weights for each region are initialized based on the regional type data in each spatiotemporal database. The initialization basis matrix for the region is constructed by combining the preset typical basic feature fingerprint database. The NMF load decoupling algorithm is executed to obtain dynamic non-negative row vectors, which include resident coefficients and industrial coefficients. The carbon emission intensity of residents and industries for the current time period is calculated based on the non-negative row vectors and the spatiotemporal database, and then updated to the spatiotemporal database.

[0012] Based on the spatiotemporal database, the predicted future carbon emissions for different future time periods are calculated and generated.

[0013] Anomaly detection calculations are performed based on carbon emission intensity and the predicted future carbon emissions for the current period to determine whether anomalies exist. If anomalies are found, root cause analysis is performed to generate an anomaly detection analysis dataset for the corresponding region and period.

[0014] As a further improvement of the present invention, while generating the predicted future carbon emissions for different future time periods, the predicted future carbon emissions are synchronized to the spatiotemporal database.

[0015] Following the anomaly detection and analysis dataset, the steps include: calculating the monitoring value score for each region based on the spatiotemporal database; generating corresponding investigation actions and anomaly risk coefficients based on the anomaly detection and analysis dataset; and iteratively calculating the optimal solution to obtain the investigation scheduling instruction based on the monitoring value score, the location data of each region, the anomaly risk coefficient, the investigation actions, and the maximum limit threshold corresponding to each anomaly investigation team.

[0016] As a further improvement of the present invention, the typical basic feature fingerprint database includes resident fingerprints and industrial fingerprints. The resident fingerprints are fingerprint matrices of pure resident daily load curves, and the industrial fingerprints are fingerprint matrices of pure industrial daily load curves.

[0017] The specific steps for constructing the initial basis matrix of the region by combining the preset typical basic feature fingerprint database include: calculating the resident activity weight based on the industrial activity weight, multiplying the industrial activity weight by the industrial fingerprint to obtain the industrial basic form of the region, multiplying the resident activity weight by the resident fingerprint to obtain the resident basic form of the region, and summing the industrial basic form and the resident basic form to obtain the initial basis matrix of the region.

[0018] As a further improvement of the present invention, the specific steps of executing the NMF load decoupling algorithm to obtain the dynamic non-negative row vector include: initializing a dynamic non-negative row vector of size one by two; generating a dynamic non-negative matrix based on the initialization basis matrix; constructing an objective function by multiplying the dynamic non-negative matrix and the dynamic non-negative row vector; iterating the dynamic non-negative matrix, the dynamic non-negative row vector, and the total load data using the NMF standard algorithm to minimize the Euclidean distance between the value obtained by multiplying the dynamic non-negative matrix and the dynamic non-negative row vector and the value of the total load data; and obtaining the residential load curve, the industrial load curve, and the dynamic non-negative row vector corresponding to the region. The dynamic non-negative row vector includes the residential coefficient corresponding to the residential load curve and the industrial coefficient corresponding to the industrial load curve.

[0019] The specific steps for calculating the carbon emission intensity of residents and industries for the current period include: calculating the total electricity consumption for the current period based on the total load data;

[0020] The carbon intensity of residents is calculated by multiplying the resident coefficient, carbon emission activity data, and total electricity consumption.

[0021] The carbon intensity of industry is obtained by calculating the product of the industrial coefficient, carbon emission activity data, and total electricity consumption.

[0022] As a further improvement of the present invention, the meteorological data includes temperature data;

[0023] The specific steps for calculating and generating future carbon emission forecasts for different future time periods based on the spatiotemporal database include: adjusting the dynamic reference period range based on calendar attribute data; selecting temperature data, residential carbon emission intensity, industrial carbon emission intensity, residential load curve, and industrial load curve from the spatiotemporal database that are the same as the current time period within the dynamic reference period range, and recording them as reference temperature data, reference residential carbon emission intensity, reference industrial carbon emission intensity, reference residential load curve, and reference industrial load curve; calculating the difference between the current time period's temperature data and the reference temperature data to obtain the temperature difference; if the temperature difference is greater than the dynamic temperature change threshold, then removing all reference data corresponding to the reference temperature data.

[0024] Calculate the variance between the current period's residential carbon emission intensity and the reference residential carbon emission intensity to obtain the residential carbon emission intensity similarity value; calculate the variance between the current period's industrial carbon emission intensity and the reference industrial carbon emission intensity to obtain the industrial carbon emission intensity similarity value; calculate the sum of the residential carbon emission intensity similarity value and the industrial carbon emission intensity similarity value to obtain the basic similarity value.

[0025] The average residential slope is obtained by calculating the residential slope for multiple sub-periods based on the residential load curve and averaging the industrial slope for multiple sub-periods based on the industrial load curve. The average reference residential slope is obtained by calculating the reference residential slope for multiple sub-periods based on the reference residential load curve and averaging the reference industrial slope for multiple sub-periods based on the reference industrial load curve. The variance between the average residential slope and the reference average residential slope is calculated to obtain the similarity value of the residential slope. The variance between the average industrial slope and the reference average industrial slope is calculated to obtain the similarity value of the industrial slope.

[0026] The basic similarity value, resident slope similarity value, and industrial slope similarity value are summed according to their weights to obtain multiple historical reference value scores. Each historical reference value score is normalized to obtain multiple historical data learning weights. The resident slope similarity value and industrial slope similarity value located in the same time period on the same day are grouped together and normalized separately to obtain multiple resident learning weights and multiple industrial learning weights that correspond one-to-one with the resident learning weights.

[0027] By fitting historical carbon emission intensity for different future periods based on historical data learning weights, resident learning weights, and industrial learning weights, we can obtain predicted future carbon emission values ​​for different future periods.

[0028] As a further improvement of the present invention, the specific steps for determining whether there is an anomaly include performing anomaly detection calculation based on carbon emission intensity and the predicted future carbon emission value for the corresponding current period: calculating the carbon emission prediction range based on the predicted future carbon emission value for the corresponding current period and a preset prediction acceptance ratio; if the carbon emission intensity falls outside the carbon emission prediction range for multiple consecutive periods and the deviation direction is consistent, then an anomaly is considered to exist.

[0029] The anomaly detection analysis dataset includes at least one fault diagnosis cause and one actual fault cause;

[0030] The specific steps for root cause analysis and generating anomaly detection analysis datasets for the corresponding region and time period include: identifying and capturing the timestamps of anomalies in the residential load curve and industrial load curve for the current time period, recording them as anomaly timestamps; obtaining the residential and industrial anomaly slopes for the sub-time periods before and after the anomaly timestamps; selecting residential and industrial slopes from the spatiotemporal database that correspond to the residential and / or industrial anomaly slopes and whose differences are less than a preset fault similarity threshold, and calculating the fault similarity; retrieving the anomaly detection analysis datasets corresponding to the residential and / or industrial slopes; extracting the true fault causes from each anomaly detection analysis dataset; calculating at least one fault diagnosis cause according to the fault similarity; and at least one analysis accuracy rate corresponding one-to-one with the fault diagnosis cause.

[0031] If there is no resident slope or industrial slope that corresponds to the abnormal resident slope or industrial slope and the difference is less than the preset fault threshold, then the fault diagnosis cause will be set to blank.

[0032] By integrating all fault diagnosis causes and corresponding analysis accuracy, an anomaly detection and analysis dataset is obtained.

[0033] This invention also proposes a data intelligent analysis system for carbon emission monitoring, which employs a data intelligent analysis method for carbon emission monitoring, including:

[0034] The data input unit is used to input the region type data and carbon emission activity data for each region.

[0035] The data acquisition unit is signal-connected to the data input unit and is used to acquire total load data, meteorological data, calendar attribute data, location data, regional type data and carbon emission activity data for each region, and integrate them into the spatiotemporal database of each region according to the corresponding timestamp.

[0036] The spatiotemporal database is connected to the data acquisition unit via signals.

[0037] The computing unit, connected to the spatiotemporal database, initializes the industrial activity weights for each region based on the regional type data in each spatiotemporal database, and constructs the initialization basis matrix for the region by combining it with a preset typical basic feature fingerprint database. It executes the NMF load decoupling algorithm to obtain dynamic non-negative row vectors, which include resident coefficients and industrial coefficients. Based on the non-negative row vectors and the spatiotemporal database, it calculates the carbon emission intensity of residents and industries for the current time period and updates it to the spatiotemporal database. It calculates and generates future carbon emission prediction values ​​for different future time periods. Based on the carbon emission intensity and the future carbon emission prediction values ​​for the current time period, it performs anomaly detection calculations to determine whether there are any anomalies. If there are, it performs root cause analysis to generate an anomaly detection analysis dataset for the region for that time period and updates it to the spatiotemporal database.

[0038] The beneficial effects of this invention are as follows: Due to privacy constraints, household-level electricity consumption data from power supply bureaus cannot be transmitted or provided arbitrarily. Therefore, we can only analyze anonymized aggregated data, such as total load data for distribution areas. By analyzing the electricity consumption habits of industry and residents, and using the NMF algorithm to decouple the residential and industrial coefficients, this scheme provides a crucial prerequisite for accurately calculating the combined carbon emissions contribution of residents and industry. Simultaneously, calculations based on total load data can intuitively, in real-time, and tamper-proofly reflect the carbon emission intensity of the entire region indirectly. Furthermore, mobile carbon emission instruments are used for sampling monitoring to ensure there is no "dual use" of electricity. Subsequent root cause analysis can improve the real-time nature of monitoring and early warning efficiency. Preliminary early warning judgments through root cause analysis significantly reduce monitoring costs and provide a scalable technical path for refined management of regional carbon emissions. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0041] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product.

[0042] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings. The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0043] Reference Figure 1 A data intelligent analysis method for carbon emission monitoring includes the following methods:

[0044] Acquire total load data, meteorological data, calendar attribute data, location data, region type data, and carbon emission activity data for each region, and integrate them into the spatiotemporal database of each region based on the corresponding timestamps;

[0045] Because the power supply bureau's household-level electricity consumption data is subject to privacy constraints and cannot be freely transmitted, we can only analyze it using anonymized aggregated data, such as the total load data of the distribution area. Compared to existing technologies that rely on installing numerous carbon emission monitoring sensors, this invention only requires acquiring the total load value of each area, meteorological data such as temperature, humidity, and weather conditions (sunny, cloudy, rainy), calendar attribute data such as weekdays and holidays, and location data. These are all easily obtainable data that require almost no additional sensor installation. Furthermore, to effectively monitor carbon emissions in areas with mixed residential and industrial populations and achieve more accurate predictions, this invention adds area type data and carbon emission activity data to adjust the forecast. Area type data essentially refers to the ratio of residential to industrial scale within the area; specifically, it's the industrial scale ratio, ranging from 0 to 1. A value of 1 indicates that the area is purely industrial load with no residents. Therefore, the industrial activity weight can be directly obtained from the area type data. This weight is determined based on the number of user types and electricity consumption per meter. For example, if a large chemical plant exists in the area, although it belongs to the industrial category and has a small number of electricity meters, its electricity consumption is extremely high. Therefore, the area's category data needs to be adjusted upwards. It's important to note that electricity consumption by shops can be directly equated to residential scale, but if the shop's operations generate significant carbon emissions, it should be considered part of the industrial scale. For carbon emission activity data, a comprehensive assessment can be made based on multiple raw materials, motors, processes, and recycling systems used in the industry. For instance, if the main raw material is fabric, and the factory only uses it for sewing and garment manufacturing, the carbon emission activity data would be low and can be simply considered the same as residential carbon emission activity data. Although carbon emission activity data still requires manual assessment, intermittent manual offline or drone spot checks can be used to determine if the factory's production objects and tools have changed (and this is unaffected by companies concealing information). If there are no changes, the previous data can be used. Compared to existing technologies that require real-time acquisition of all factory purchase information and specific purchase quantities to calculate carbon emissions, this method is much more accurate and unaffected by factories concealing information to meet carbon emission targets. When the factory's production targets remain unchanged, its electricity consumption is almost directly linked to the product volume, which in turn indirectly reflects carbon emissions. Of course, the carbon emission activity data differs between summer and winter. In summer, due to the increased number of cooling devices, a lower carbon emission activity data value is required.

[0046] In addition, the industry average carbon emission factor can be used as the initial value for carbon emission activity data in the initial stage of the system. During operation, the carbon emission activity data can be calibrated and optimized by combining a very small amount of randomly sampled mobile carbon emission monitoring data.

[0047] It's important to note that because this invention indirectly derives carbon emission intensity through electricity consumption, in extreme cases where a factory secretly transports high-carbon-emission materials and uses similarly loaded machinery for production, this invention cannot detect this through electricity consumption monitoring ("dual use of electricity"). When a certain time period is triggered, a drone will be deployed to conduct multiple inspections at multiple locations within the factory's operating hours, carrying carbon emission monitoring equipment. This invention is generally applicable to factories with carbon emission quotas (or factories subject to certain carbon emission requirements), where monitoring electricity consumption can reveal whether operations exceed those quotas.

[0048] It is important to note here that the accuracy of the regional type data is only intended to provide an initial value for subsequent decoupling iterations, which significantly reduces the number of iterations. Therefore, it will not affect the subsequent calculation of carbon emission intensity.

[0049] The industrial activity weights for each region are initialized based on the regional type data in each spatiotemporal database. The initialization basis matrix for the region is constructed by combining the preset typical basic feature fingerprint database. The NMF load decoupling algorithm is executed to obtain dynamic non-negative row vectors, which include resident coefficients and industrial coefficients. The carbon emission intensity of residents and industries for the current time period is calculated based on the non-negative row vectors and the spatiotemporal database, and then updated to the spatiotemporal database.

[0050] By analyzing the electricity consumption habits of industry and residents respectively, and using the NMF algorithm to decouple the residential and industrial coefficients, this scheme provides a key premise for accurately calculating the contribution of mixed carbon emissions from residents and industry. At the same time, the calculation based on total load data can intuitively, in real time and tamper-proofly reflect the carbon emission intensity of the entire region indirectly.

[0051] As a further improvement of the present invention, the typical basic feature fingerprint database includes resident fingerprints and industrial fingerprints. The resident fingerprints are fingerprint matrices of pure resident daily load curves, and the industrial fingerprints are fingerprint matrices of pure industrial daily load curves.

[0052] The specific steps for constructing the initial basis matrix of the region by combining the preset typical basic feature fingerprint database include: calculating the resident activity weight based on the industrial activity weight, multiplying the industrial activity weight by the industrial fingerprint to obtain the industrial basic form of the region, multiplying the resident activity weight by the resident fingerprint to obtain the resident basic form of the region, and summing the industrial basic form and the resident basic form to obtain the initial basis matrix of the region.

[0053] Specifically, as mentioned above, the industrial activity weight for each region is initialized based on the region type data in each spatiotemporal database. Essentially, the industrial activity weight is directly equal to the region type data value. Therefore, the resident activity weight is calculated by subtracting the industrial activity weight from one.

[0054] For obtaining residential and industrial fingerprints, electricity consumption data from purely residential or purely industrial areas with similar annual temperature data can be used. For example, many places have industrial zones, and a single transformer substation may contain only factories and enterprises; the same principle applies to residential fingerprints. It's important to note that in one specific embodiment of this invention, for each mixed residential and industrial area, the residential fingerprint can be obtained during the few days of the Spring Festival holiday, and then combined with the total load curve to obtain the industrial fingerprint. If fingerprints from other purely residential or purely industrial areas are used, after summing to obtain the initial basis matrix of that area, it needs to be multiplied by the ratio of the total load of the two areas (this is also a specific embodiment for generating a dynamic non-negative matrix based on the initial basis matrix; if the fingerprint of this area is used, no multiplication by any factor is required).

[0055] It should be noted again that this initialization basis matrix is ​​only used as the initial value for subsequent iterative algorithms. Therefore, if there are errors in obtaining this matrix, it will not affect the accuracy of the subsequent carbon emission intensity.

[0056] Furthermore, because the load curves of both residential and industrial sectors are highly time-dependent, the value of the industrial activity weight is not static. The ratio of residential to industrial scale mentioned above is only an initial value calculation (therefore, this value can be determined without obtaining data such as electricity meter readings; the initial value of the industrial activity weight can be obtained directly from a single carbon emission intensity calculation). After carbon emission monitoring in the region has been implemented for a period of time, the industrial coefficient corresponding to the same historical period can be directly borrowed to represent the industrial activity weight. When borrowing, calendar attribute data should also be considered, such as whether it is a weekday, a holiday, or a weekend (if it is determined that the region does not rest on Saturdays and Sundays, it can be considered a weekday).

[0057] As a further improvement of the present invention, the specific steps of executing the NMF load decoupling algorithm to obtain the dynamic non-negative row vector include: initializing a dynamic non-negative row vector of size one by two; generating a dynamic non-negative matrix based on the initialization basis matrix; constructing an objective function by multiplying the dynamic non-negative matrix and the dynamic non-negative row vector; iterating the dynamic non-negative matrix, the dynamic non-negative row vector, and the total load data using the NMF standard algorithm to minimize the Euclidean distance between the value obtained by multiplying the dynamic non-negative matrix and the dynamic non-negative row vector and the value of the total load data; and obtaining the residential load curve, the industrial load curve, and the dynamic non-negative row vector corresponding to the region. The dynamic non-negative row vector includes the residential coefficient corresponding to the residential load curve and the industrial coefficient corresponding to the industrial load curve.

[0058] The specific steps for calculating the carbon emission intensity of residents and industries for the current period include: calculating the total electricity consumption for the current period based on the total load data;

[0059] The carbon intensity of residents is calculated by multiplying the resident coefficient, carbon emission activity data, and total electricity consumption.

[0060] The carbon intensity of industry is obtained by calculating the product of the industrial coefficient, carbon emission activity data, and total electricity consumption.

[0061] Since the Non-negative Matrix Factorization (NMF) algorithm is a conventional iterative calculation method, it will not be elaborated upon further in this paper. Its essence is to continuously adjust the values ​​of various points on the residential load curve, the industrial load curve, and the industrial coefficient plus the residential coefficient (which is equal to one). The process pauses after a certain number of iterations or when the change after iteration is less than a preset threshold. This technical solution effectively separates residential and industrial load conditions. Without installing any load sensors, knowing the total load of the distribution area is sufficient to determine the approximate distribution of residential and industrial loads.

[0062] For residential carbon emission activity data, it can be directly approximated to 0. For carbon emission activity data in carbon emission intensity calculations, as mentioned above, it can be determined based on multiple raw materials mainly used in industry. For example, if the main raw material is fabric, and the factory only uses it for sewing and garment making, then the carbon emission activity data will be low and can be simply considered the same as the residential carbon emission activity data. Although carbon emission activity data still requires manual judgment, it is sufficient to intermittently check whether the factory's production targets have changed; if not, the previous data is used. Compared to existing technologies that require real-time acquisition of all factory purchase information, including specific quantities and types of goods, to calculate carbon emissions, this method is much more accurate. Furthermore, it is not affected by factories underreporting their carbon emission targets. Because when the factory's production targets remain unchanged, its electricity consumption is almost directly linked to its product volume, which indirectly reflects carbon emissions. Of course, the carbon emission activity data differs between summer and winter; in summer, due to the increased use of cooling equipment, a lower carbon emission activity data value is required. Converting electricity consumption into carbon emissions through carbon emission activity data will inevitably introduce errors, as each factory's electricity consumption varies. However, this invention aims to remind you that the purpose of this technical solution is to monitor whether there are abnormal carbon emissions in areas where residents and industries mix, not to accurately calculate the actual carbon emission intensity of the area. Therefore, only a general range is needed.

[0063] As a specific embodiment of the present invention, if it is desired to locate anomalies later, multiple sensors monitoring the load can be installed using a binary search method. It should be noted that no sensors are needed for monitoring carbon emissions; multiple sensors monitoring the load are only required for anomaly location.

[0064] Based on the spatiotemporal database, the predicted future carbon emissions for different future time periods are calculated and generated.

[0065] As a further improvement of the present invention, the meteorological data includes temperature data;

[0066] The specific steps for calculating and generating future carbon emission forecasts for different future time periods based on the spatiotemporal database include: adjusting the dynamic reference period range based on calendar attribute data; selecting temperature data, residential carbon emission intensity, industrial carbon emission intensity, residential load curve, and industrial load curve from the spatiotemporal database that are the same as the current time period within the dynamic reference period range, and recording them as reference temperature data, reference residential carbon emission intensity, reference industrial carbon emission intensity, reference residential load curve, and reference industrial load curve; calculating the difference between the current time period's temperature data and the reference temperature data to obtain the temperature difference; if the temperature difference is greater than the dynamic temperature change threshold, then removing all reference data corresponding to the reference temperature data.

[0067] The dynamic reference period range is determined as described above, selecting weekdays, holidays, and rest days based on the calendar attribute data of the day (if the total load curve of the total load data is significantly different from the pure residential daily load curve, it is considered a weekday). At least ten reference days are used as the dynamic reference period range; however, if any historical day number is abnormal, that day is disregarded. Calculating the temperature difference is to prevent areas with large temperature fluctuations from having their accuracy severely affected by temperature.

[0068] Calculate the variance between the current period's residential carbon emission intensity and the reference residential carbon emission intensity to obtain the residential carbon emission intensity similarity value; calculate the variance between the current period's industrial carbon emission intensity and the reference industrial carbon emission intensity to obtain the industrial carbon emission intensity similarity value; calculate the sum of the residential carbon emission intensity similarity value and the industrial carbon emission intensity similarity value to obtain the basic similarity value.

[0069] Since different factories have different working hours and rest periods, calculating basic similarity values ​​helps to improve the learning weight of historical data that most closely resembles the working conditions of the day.

[0070] The average residential slope is obtained by calculating the residential slope for multiple sub-periods based on the residential load curve and averaging the industrial slope for multiple sub-periods based on the industrial load curve. The average reference residential slope is obtained by calculating the reference residential slope for multiple sub-periods based on the reference residential load curve and averaging the reference industrial slope for multiple sub-periods based on the reference industrial load curve. The variance between the average residential slope and the reference average residential slope is calculated to obtain the similarity value of the residential slope. The variance between the average industrial slope and the reference average industrial slope is calculated to obtain the similarity value of the industrial slope.

[0071] Calculating the slope similarity value for residents and the slope similarity value for industries allows for a more detailed assessment of the degree of trend change between historical data and the current situation, thereby increasing the learning weight for residents or industries with similar trend changes.

[0072] The basic similarity value, resident slope similarity value, and industrial slope similarity value are summed according to their weights to obtain multiple historical reference value scores. Each historical reference value score is normalized to obtain multiple historical data learning weights. The resident slope similarity value and industrial slope similarity value located in the same time period on the same day are grouped together and normalized separately to obtain multiple resident learning weights and multiple industrial learning weights that correspond one-to-one with the resident learning weights.

[0073] Normalizing historical reference value scores specifically means dividing the historical reference value by the sum of all historical reference values. Normalizing resident slope similarity values ​​and industrial slope similarity values ​​involves dividing the resident slope similarity value for the same day and time period by the sum of the resident slope similarity value and the industrial slope similarity value. Then, the sum of the resident learning weights for different days and the sum of the industrial learning weights must equal 1.

[0074] By fitting historical carbon emission intensity for different future periods based on historical data learning weights, resident learning weights, and industrial learning weights, we can obtain predicted future carbon emission values ​​for different future periods.

[0075] To facilitate understanding, this invention provides an example. For instance, monitoring is conducted at 1:58 PM on November 10, 2025. To predict the carbon emission intensity for the period from 2:00 PM to 2:30 PM, where each period is half an hour and each sub-period is five minutes, three similar historical days can be used for reference: day a, day b, and day c. Assuming their historical data learning weights are 0.3, 0.3, and 0.4 respectively, the residential learning weights are 0.2, 0.3, and 0.5 respectively, and the industrial learning weights are 0.4, 0.35, and 0.25 respectively, then the historical residential carbon emission intensity for the period from 2:00 PM to 2:30 PM on day a is multiplied by 0.2, and the historical industrial carbon emission intensity is multiplied by 0.8. Similarly, the carbon emission intensity for days b and c is calculated. Finally, the historical residential carbon emission intensity for these three days is multiplied by 0.3, 0.3, and 0.4 respectively, and the sum is used to obtain the predicted future residential carbon emission value. Similarly, the predicted future industrial carbon emission value is calculated, and the sum of the two yields the predicted future carbon emission value.

[0076] It should be noted that the future carbon emission forecast can predict two to three time periods. After each forecast, the old forecast data is overwritten, and the latest future carbon emission forecast is used (this value is also synchronized to the spatiotemporal dataset library). If the system malfunctions and cannot calculate, the latest version of the future carbon emission forecast will be used.

[0077] Anomaly detection calculations are performed based on carbon emission intensity and the predicted future carbon emissions for the current period to determine whether anomalies exist. If anomalies are found, root cause analysis is performed to generate an anomaly detection analysis dataset for the corresponding region and period.

[0078] Root cause analysis can improve the real-time performance of monitoring and the efficiency of early warning. Preliminary early warning judgments through root cause analysis significantly reduce monitoring costs and provide a scalable technical path for the refined management of regional carbon emissions.

[0079] As a further improvement of the present invention, the specific steps for determining whether there is an anomaly include performing anomaly detection calculation based on carbon emission intensity and the predicted future carbon emission value for the corresponding current period: calculating the carbon emission prediction range based on the predicted future carbon emission value for the corresponding current period and a preset prediction acceptance ratio; if the carbon emission intensity falls outside the carbon emission prediction range for multiple consecutive periods and the deviation direction is consistent, then an anomaly is considered to exist.

[0080] Specifically, the carbon emission prediction range can be set at 5%, depending on the total load of the entire region. The larger the total load, the smaller the carbon emission prediction range.

[0081] It's important to note here that because this is a future carbon emission forecast fitted from historical data, if historical data shows a high-load situation (e.g., production equipment wasn't running overtime in the past, resulting in a lower future carbon emission forecast at that time, leading to a high-load alarm in the historical data when production equipment was running overtime), it won't cause the current period's future carbon emission forecast to increase accordingly. This is because the historical data showed a high-load alarm at that time and was therefore outside the dynamic reference period. Naturally, when fitting the current period's future carbon emission forecast, it won't refer to the historical data for that day; the core principle is to use normal conditions as the benchmark. "Consistent deviation" means consistently exceeding or falling below the carbon emission forecast range.

[0082] In addition, it should be noted that employees' non-production overtime work will not cause significant load fluctuations and therefore will not trigger high load alarms, because the load generated by production equipment is generally much greater than that of equipment such as computers and lighting.

[0083] The anomaly detection analysis dataset includes at least one fault diagnosis cause and one actual fault cause;

[0084] The specific steps for root cause analysis and generating anomaly detection analysis datasets for the corresponding region and time period include: identifying and capturing the timestamps of anomalies in the residential load curve and industrial load curve for the current time period, recording them as anomaly timestamps; obtaining the residential and industrial anomaly slopes for the sub-time periods before and after the anomaly timestamps; selecting residential and industrial slopes from the spatiotemporal database that correspond to the residential and / or industrial anomaly slopes and whose differences are less than a preset fault similarity threshold, and calculating the fault similarity; retrieving the anomaly detection analysis datasets corresponding to the residential and / or industrial slopes; extracting the true fault causes from each anomaly detection analysis dataset; calculating at least one fault diagnosis cause according to the fault similarity; and at least one analysis accuracy rate corresponding one-to-one with the fault diagnosis cause.

[0085] If there is no resident slope or industrial slope that corresponds to the abnormal resident slope or industrial slope and the difference is less than the preset fault threshold, then the fault diagnosis cause will be set to blank.

[0086] By integrating all fault diagnosis causes and corresponding analysis accuracy, an anomaly detection and analysis dataset is obtained.

[0087] Root cause analysis essentially involves referencing historical fault diagnosis and combining it with residential and industrial load curves to quickly identify the causes of recurring, high-frequency faults, such as short circuits or damage to production equipment, or expansion of production scale. This avoids wasting the anomaly investigation team's time and effort in subsequent troubleshooting. If a load alarm occurs on a given day due to normal business operations such as expanded production scale, that day can be set as normal and its data can be used for future reference.

[0088] As a further improvement of the present invention, while generating the predicted future carbon emissions for different future time periods, the predicted future carbon emissions are synchronized to the spatiotemporal database.

[0089] Following the anomaly detection and analysis dataset, the steps include: calculating the monitoring value score for each region based on the spatiotemporal database; generating corresponding investigation actions and anomaly risk coefficients based on the anomaly detection and analysis dataset; and iteratively calculating the optimal solution to obtain the investigation scheduling instruction based on the monitoring value score, the location data of each region, the anomaly risk coefficient, the investigation actions, and the maximum limit threshold corresponding to each anomaly investigation team.

[0090] Subsequent root cause analysis can improve the real-time nature of monitoring and the efficiency of early warning. Preliminary early warning judgments through root cause analysis significantly reduce monitoring costs and provide a scalable technical path for refined management of regional carbon emissions. Specifically, investigation actions include on-site investigations, drone investigations, and "nothing" (meaning no investigation action). "Nothing" investigation actions could be due to reasons such as short circuits or damage to production equipment, or expansion of production scale. When an investigation action is "nothing," it is directly excluded from the investigation scheduling instructions, except for those that have not been investigated within the preset investigation time.

[0091] The expression for the monitoring value score is as follows:

[0092] ,

[0093] in, It is a monitoring value score. It is the carbon emission risk weighting coefficient. It is the sum of the carbon emission intensity of residents and industry. This is a projected value for future carbon emissions. It is a time decay function. It is the time interval since the last investigation, in months or weeks. It is an adjustment factor for the degree of influence. It is a time adjustment factor.

[0094] It should be noted here that, as a specific embodiment of the present invention, if a factory has not been inspected within a preset inspection time, it will directly enter the inspection queue. For example, the carbon emissions of the factory can be monitored multiple times through the inspection actions of drones.

[0095] For ease of understanding, as an example of this invention, the carbon emission risk weighting coefficient is set to 2 (determined according to industry type), the total carbon emission intensity is 200, the future carbon emission prediction value is 150, there has been no investigation for three months since the last time, the impact adjustment factor is 0.8, and the time adjustment factor is 0.2. The value of the time decay function is approximately 1.361, and the value calculated from the monitoring value score is approximately 3.63. For ease of understanding, numerical optimization calculations have been performed on the specific example.

[0096] The objective function expression in the iterative optimal solution calculation is as follows:

[0097] ,

[0098] ,

[0099] ,

[0100] in, It is the total value of traveling to region a in the nth iteration. It is the total cost of traveling to region a in the nth iteration. It is a monitoring value score. It is the abnormal risk coefficient. It is a value weight adjustment factor. It represents the path time of the t-th anomaly investigation team from starting point o to region a in the nth iteration. This is the average investigation time for this investigation action. It is a time cost weighting adjustment factor. It is a factor that amplifies energy loss. It represents the remaining workload of the t-th anomaly investigation team in the nth iteration.

[0101] Since the iterative optimal solution calculation method is a common technique in conventional calculations, it will not be elaborated upon further in this invention. The principle and logic of constructing the objective function are as follows: taking each anomaly investigation team as a line, each area with load alarms is screened, and the optimal solution is iteratively selected to determine the areas and paths that can be investigated with one day's effort. Subsequently, the selected areas are excluded, and the iterative optimal solution calculation is performed for another anomaly investigation team. The selection of anomaly investigation teams can be determined by the interval between the last investigation action. The value of the anomaly risk coefficient is determined based on the cause of the fault diagnosis. The more serious the fault diagnosis cause, the greater the anomaly risk coefficient. The value weight adjustment factor and the time cost weight adjustment factor are common adjustment factors used to adjust the numerical relationship between the total value and the total cost. The effort attenuation amplification factor is determined based on the investigation action. For example, drone investigation requires higher effort requirements, so this value should be greater than that of on-site investigation.

[0102] As a further improvement of the present invention, the present invention also proposes a data intelligent analysis system for carbon emission monitoring, which employs a data intelligent analysis method for carbon emission monitoring, including:

[0103] The data input unit is used to input the region type data and carbon emission activity data for each region.

[0104] The data acquisition unit is signal-connected to the data input unit and is used to acquire total load data, meteorological data, calendar attribute data, location data, regional type data and carbon emission activity data for each region, and integrate them into the spatiotemporal database of each region according to the corresponding timestamp.

[0105] The spatiotemporal database is connected to the data acquisition unit via signals.

[0106] The computing unit, connected to the spatiotemporal database, initializes the industrial activity weights for each region based on the regional type data in each spatiotemporal database, and constructs the initialization basis matrix for the region by combining it with a preset typical basic feature fingerprint database. It executes the NMF load decoupling algorithm to obtain dynamic non-negative row vectors, which include resident coefficients and industrial coefficients. Based on the non-negative row vectors and the spatiotemporal database, it calculates the carbon emission intensity of residents and industries for the current time period and updates it to the spatiotemporal database. It calculates and generates future carbon emission prediction values ​​for different future time periods. Based on the carbon emission intensity and the future carbon emission prediction values ​​for the current time period, it performs anomaly detection calculations to determine whether there are any anomalies. If there are, it performs root cause analysis to generate an anomaly detection analysis dataset for the region for that time period and updates it to the spatiotemporal database.

[0107] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A data intelligent analysis method applied to carbon emission monitoring, characterized in that, The method comprises the following steps: Obtain total load data, meteorological data, calendar attribute data, location data, region type data, and carbon emission activity data for each region, and integrate them into a spatio-temporal database for each region according to the corresponding timestamps; The region type data is the proportion of the scale of residents and the scale of industries in the region; Initialize the industrial activity weight for each region according to the region type data in each spatio-temporal database, construct an initialization basis matrix for the region in combination with a preset typical basic feature fingerprint library, and perform an NMF load decoupling algorithm to obtain a dynamic non-negative row vector, which comprises a resident coefficient and an industrial coefficient, calculate the carbon emission intensity of residents and industries in the current period according to the non-negative row vector and the spatio-temporal database, and update them to the spatio-temporal database; Calculate future carbon emission prediction values corresponding to future different time periods according to the spatio-temporal database; Perform anomaly detection calculation according to the carbon emission intensity and the future carbon emission prediction values corresponding to the current period to determine whether there is an anomaly, and if so, perform root cause analysis to generate an anomaly detection analysis data set corresponding to the region in the period; The typical basic feature fingerprint library comprises resident fingerprints and industrial fingerprints, and the resident fingerprints are a fingerprint matrix of pure resident daily load curves, and the industrial fingerprints are a fingerprint matrix of pure industrial daily load curves; The specific steps of constructing the initialization basis matrix for the region in combination with the preset typical basic feature fingerprint library comprise the following steps: calculating a resident activity weight according to the industrial activity weight, multiplying the industrial activity weight by the industrial fingerprints to obtain an industrial basic form of the region, multiplying the resident activity weight by the resident fingerprints to obtain a resident basic form of the region, and summing the industrial basic form and the resident basic form to obtain the initialization basis matrix of the region.

2. The data intelligent analysis method applied to carbon emission monitoring according to claim 1, characterized in that, While generating the future carbon emission prediction values corresponding to the future different time periods, the future carbon emission prediction values are also synchronized to the spatio-temporal database; The anomaly detection analysis data set further comprises the following steps: calculating a monitoring value score of each region according to the spatio-temporal database, generating corresponding investigation actions and anomaly risk coefficients according to the anomaly detection analysis data set, and performing iterative optimal solution calculation according to the monitoring value score, the location data of each region, the anomaly risk coefficients, the investigation actions, and a maximum limit threshold corresponding to each anomaly investigation team to obtain an investigation scheduling instruction.

3. The method for intelligent analysis of data applied to carbon emission monitoring according to claim 1, characterized in that, The specific steps of performing the NMF load decoupling algorithm to obtain the dynamic non-negative row vector comprise the following steps: initializing a dynamic non-negative row vector with a size of one by two, generating a dynamic non-negative matrix according to the initialization basis matrix, constructing a target function according to the multiplication of the dynamic non-negative matrix and the dynamic non-negative row vector, and performing NMF standard algorithm iteration on the dynamic non-negative matrix, the dynamic non-negative row vector, and the total load data to minimize the Euclidean distance between the value obtained by multiplying the dynamic non-negative matrix and the dynamic non-negative row vector and the value of the total load data, thereby obtaining a resident load curve, an industrial load curve, and a dynamic non-negative row vector corresponding to the region, wherein the dynamic non-negative row vector comprises a resident coefficient corresponding to the resident load curve and an industrial coefficient corresponding to the industrial load curve; The specific steps of calculating the carbon emission intensity of residents and industries in the current period comprise the following steps: calculating the total power consumption corresponding to the current period according to the total load data; The carbon intensity of residents is calculated by multiplying the resident coefficient, carbon emission activity data, and total electricity consumption. The carbon intensity of industry is obtained by calculating the product of the industrial coefficient, carbon emission activity data, and total electricity consumption.

4. The method for intelligent analysis of data applied to carbon emission monitoring according to claim 3, characterized in that, Meteorological data includes temperature data; The specific steps for calculating and generating future carbon emission forecasts for different future time periods based on the spatiotemporal database include: adjusting the dynamic reference period range based on calendar attribute data; selecting temperature data, residential carbon emission intensity, industrial carbon emission intensity, residential load curve, and industrial load curve from the spatiotemporal database that are the same as the current time period within the dynamic reference period range, and recording them as reference temperature data, reference residential carbon emission intensity, reference industrial carbon emission intensity, reference residential load curve, and reference industrial load curve; calculating the difference between the current time period's temperature data and the reference temperature data to obtain the temperature difference; if the temperature difference is greater than the dynamic temperature change threshold, then removing all reference data corresponding to the reference temperature data. Calculate the variance between the current period's residential carbon emission intensity and the reference residential carbon emission intensity to obtain the residential carbon emission intensity similarity value; calculate the variance between the current period's industrial carbon emission intensity and the reference industrial carbon emission intensity to obtain the industrial carbon emission intensity similarity value; calculate the sum of the residential carbon emission intensity similarity value and the industrial carbon emission intensity similarity value to obtain the basic similarity value. The average residential slope is obtained by calculating the residential slope for multiple sub-periods based on the residential load curve and averaging the industrial slope for multiple sub-periods based on the industrial load curve. The average reference residential slope is obtained by calculating the reference residential slope for multiple sub-periods based on the reference residential load curve and averaging the reference industrial slope for multiple sub-periods based on the reference industrial load curve. The variance between the average residential slope and the reference average residential slope is calculated to obtain the similarity value of the residential slope. The variance between the average industrial slope and the reference average industrial slope is calculated to obtain the similarity value of the industrial slope. The basic similarity value, resident slope similarity value, and industrial slope similarity value are summed according to their weights to obtain multiple historical reference value scores. Each historical reference value score is normalized to obtain multiple historical data learning weights. The resident slope similarity value and industrial slope similarity value located in the same time period on the same day are grouped together and normalized separately to obtain multiple resident learning weights and multiple industrial learning weights that correspond one-to-one with the resident learning weights. By fitting historical carbon emission intensity for different future periods based on historical data learning weights, resident learning weights, and industrial learning weights, we can obtain predicted future carbon emission values ​​for different future periods.

5. The method for intelligent analysis of data applied to carbon emission monitoring according to claim 1, characterized in that, The specific steps for determining whether there is an anomaly include: calculating the carbon emission prediction range based on the carbon emission intensity and the corresponding future carbon emission prediction value for the current period; if the carbon emission intensity falls outside the carbon emission prediction range for multiple consecutive periods and the deviation direction is consistent, then an anomaly is considered to exist. The anomaly detection analysis dataset includes at least one fault diagnosis cause and one actual fault cause; The specific steps for root cause analysis and generating anomaly detection analysis datasets for the corresponding region and time period include: identifying and capturing the timestamps of anomalies in the residential load curve and industrial load curve for the current time period, recording them as anomaly timestamps; obtaining the residential and industrial anomaly slopes for the sub-time periods before and after the anomaly timestamps; selecting residential and industrial slopes from the spatiotemporal database that correspond to the residential and / or industrial anomaly slopes and whose differences are less than a preset fault similarity threshold, and calculating the fault similarity; retrieving the anomaly detection analysis datasets corresponding to the residential and / or industrial slopes; extracting the true fault causes from each anomaly detection analysis dataset; calculating at least one fault diagnosis cause according to the fault similarity; and at least one analysis accuracy rate corresponding one-to-one with the fault diagnosis cause. If there is no resident slope or industrial slope that corresponds to the abnormal resident slope or industrial slope and the difference is less than the preset fault threshold, then the fault diagnosis cause will be set to blank. By integrating all fault diagnosis causes and corresponding analysis accuracy, an anomaly detection and analysis dataset is obtained.

6. A data intelligent analysis system applied to carbon emission monitoring, characterized in that, The data intelligent analysis method for carbon emission monitoring according to any one of claims 1-5 includes: The data input unit is used to input the region type data and carbon emission activity data for each region. The data acquisition unit is signal-connected to the data input unit and is used to acquire total load data, meteorological data, calendar attribute data, location data, regional type data and carbon emission activity data for each region, and integrate them into the spatiotemporal database of each region according to the corresponding timestamp. The spatiotemporal database is connected to the data acquisition unit via signals. The computing unit, connected to the spatiotemporal database, initializes the industrial activity weights for each region based on the regional type data in each spatiotemporal database, and constructs the initialization basis matrix for the region by combining it with a preset typical basic feature fingerprint database. It executes the NMF load decoupling algorithm to obtain dynamic non-negative row vectors, which include resident coefficients and industrial coefficients. Based on the non-negative row vectors and the spatiotemporal database, it calculates the carbon emission intensity of residents and industries for the current time period and updates it to the spatiotemporal database. It calculates and generates future carbon emission prediction values ​​for different future time periods. Based on the carbon emission intensity and the future carbon emission prediction values ​​for the current time period, it performs anomaly detection calculations to determine whether there are any anomalies. If there are, it performs root cause analysis to generate an anomaly detection analysis dataset for the region for that time period and updates it to the spatiotemporal database.

Images