Public institution carbon emission metering method and system based on edge computing

Abstract

The application provides a kind of public institution carbon emission metering method and system based on edge computing, belongs to carbon emission metering technical field.The method comprises the following steps: collecting multi-source energy consumption, emission, industry benchmark and meteorological data, correcting energy consumption baseline relying on meteorological data, generating hourly carbon emission sequence by using emission factor method;identify outlier data, screen deviation items against benchmark, extract identification field to build hierarchical traceability link, generate traceability atlas based on metering unit, verify two types of abnormal time series coupling, determine compliance according to accounting standards, and output abnormal attribution using ensemble learning model.The application cleanses data on edge, reduces transmission load and filters dirty data;optimizes accounting benchmark relying on meteorological and personnel dimensions;builds traceability atlas by identification matching to locate emission sources hierarchically;uses multi-feature time series to jointly determine to reduce false negatives, combines automatic attribution to replace manual verification, and opens up the whole process of carbon emission automation control.

Description

Technical Field

[0001] This invention belongs to the field of carbon emission measurement technology, specifically a carbon emission measurement method and system for public institutions based on edge computing. Background Technology

[0002] Quantitative control of carbon emissions is the core of energy conservation management in public institutions. Various metering devices are commonly used in the industry to collect monitoring information on water, electricity, gas, and the environment. Data calculation and emission statistics are then performed centrally on remote servers to understand the overall building usage and emissions.

[0003] Existing general management and control models have gradually revealed their shortcomings in terms of universality during long-term application. Traditional centralized data processing architectures rely on the continuous transmission of massive amounts of raw monitoring data over the network. The front end lacks on-site data cleaning and verification capabilities, and invalid or distorted data generated during the collection process all enter the back-end accounting process, directly affecting the reliability of statistical results. At the same time, mainstream emission statistics methods only perform simple summaries based on basic energy consumption values, making it difficult for statistical values ​​to reflect the actual emission status of buildings. Moreover, they can only output macro totals and cannot achieve hierarchical tracing of emission sources. Data anomaly detection methods are simplistic, relying solely on fixed numerical thresholds for comparison, which easily leads to misidentification and omission. Text alarm information generated by equipment operation and maintenance is difficult to transform into usable analytical evidence and lacks standardized compliance verification logic. The causes of data anomalies rely entirely on manual item-by-item analysis. The data collection, accounting, tracing, and risk verification stages are independent of each other, resulting in significant limitations in overall automation and refined management and control capabilities. Summary of the Invention

[0004] This invention provides a carbon emission measurement method and system for public institutions based on edge computing, in order to address the deficiencies in the existing technology.

[0005] On the one hand, this invention provides a method for measuring carbon emissions of public institutions based on edge computing, including: Collect multi-source energy consumption data and emission monitoring data from public institutions, and obtain industry benchmark carbon emission data and meteorological correction data.

[0006] By combining meteorological correction data to adjust the energy consumption baseline for temperature, and by performing component carbon emission accounting based on the emission factor method, an hourly carbon emission measurement sequence is generated.

[0007] Identify outlier data from in-situ collection in multi-source energy consumption data, and compare the hourly carbon emission measurement sequence with industry benchmark carbon emission data to screen for benchmark deviations.

[0008] Extract the feature identification fields from multi-source energy consumption data and emission monitoring data, and construct a hierarchical traceability link between energy-consuming equipment, emission sources, and functional areas.

[0009] Metering unit objects are extracted from multi-source energy consumption data and emission monitoring data. A topological connection relationship between each metering unit object based on the corresponding link of hierarchical traceability is constructed to generate a carbon emission traceability map of public institutions.

[0010] The time series segments of indicators to which outlier data collected in situ belong are coupled and verified with the benchmark deviation items to identify abnormal emission measurement data.

[0011] The compliance attributes of abnormal emission measurement data are determined according to carbon emission accounting standards, and an abnormal attribution identification model based on ensemble learning is constructed. The abnormal emission measurement data is input and the abnormal emission attribution category is output.

[0012] According to the edge computing-based carbon emission metering method for public institutions provided by this invention, multi-source energy consumption data includes sub-item energy consumption metering data for electricity, natural gas, tap water, centralized heating, and central air conditioning systems within the public institution, as well as operating parameters of energy-consuming equipment and personnel density data for building functional areas. Emission monitoring data includes real-time monitoring data of indoor CO2 concentration and monitoring data of exhaust gas emission outlet concentration and flow. Industry benchmark carbon emission data includes benchmark values ​​for carbon emission intensity per unit building area of ​​public institutions, benchmark values ​​for per capita carbon emissions, and benchmark values ​​for carbon emission factors of various energy sources. Meteorological correction data includes real-time data of outdoor dry-bulb temperature, relative humidity, and atmospheric pressure.

[0013] According to the edge computing-based carbon emission measurement method for public institutions provided by the present invention, the process of generating hourly carbon emission measurement sequences includes: The temperature correction factor is calculated based on the day-to-day method, and the energy consumption baseline is meteorologically corrected.

[0014] The emission factor method was used to calculate the individual carbon emissions and then summed them up to obtain the total carbon emissions.

[0015] Based on the accounting results, an hourly carbon emission measurement sequence is generated, and the carbon emission intensity per unit building area and per capita carbon emission intensity are calculated.

[0016] According to the edge computing-based carbon emission measurement method for public institutions provided by this invention, the process of identifying outlier data collected in situ includes: Calculate the median of each indicator data series in the multi-source energy consumption data.

[0017] Calculate the median of the absolute deviation of the data sequence.

[0018] Calculate the corrected robust standard deviation estimate.

[0019] By setting a deviation threshold coefficient, outlier data collected in situ can be identified based on the median and robust standard deviation estimates.

[0020] According to the edge computing-based carbon emission measurement method for public institutions provided by this invention, the process of screening benchmark deviation items includes: Obtain the benchmark value of carbon emission intensity per unit building area corresponding to public institution scenarios from industry benchmark carbon emission data.

[0021] Calculate the deviation rate between the real-time carbon emission intensity per unit building area of ​​public institutions and the benchmark value.

[0022] A deviation rate threshold is set. When the deviation rate is greater than the deviation rate threshold, the emission monitoring indicator is considered to be a benchmark deviation item.

[0023] According to the edge computing-based carbon emission measurement method for public institutions provided by this invention, the process of extracting feature identification fields and constructing hierarchical source tracing links includes: Preprocess the multi-source energy consumption data and emission monitoring data to remove missing frames, jump values ​​and duplicate sampling records.

[0024] An edge semantic parsing model is used to extract features from unstructured monitoring logs and identify data attributes and physical meanings.

[0025] Based on the data attributes, fields that can uniquely identify metering records and coexist in multi-source energy consumption data and emission monitoring data are selected as feature identification fields. Feature identification fields include equipment unique identifier code, spatial area code, energy consumption type label and collection timestamp.

[0026] Multi-source data are aligned in time sequence and sampled with uniform granularity according to the timestamp field. Feature identifier fields are matched group by group according to the hierarchical relationship of functional area-energy-consuming equipment-emission source. If the fields match completely, it is determined that there is a source-tracing connection between the corresponding units.

[0027] Establish a hierarchical source tracing and matching table to record the corresponding connections between functional areas, energy-consuming equipment, and emission sources.

[0028] According to the edge computing-based carbon emission measurement method for public institutions provided by this invention, the process of generating a carbon emission source map for public institutions includes: Energy-consuming equipment, emission outlets, and functional areas within public institutions are defined as metering unit objects, and the subordinate relationship of the equipment to its area and the transmission relationship of energy consumption to emission conversion are defined as topological connecting edges.

[0029] Nodes represent measurement unit objects, edges represent topological connections, and a graph topology storage structure is constructed based on a hierarchical tracing and matching lookup table.

[0030] For each topological connection edge, the connection contribution weight is calculated based on the carbon emission ratio and equipment runtime ratio of the corresponding path, and is used as the attribute value of the edge.

[0031] According to the edge computing-based carbon emission measurement method for public institutions provided by this invention, the process of identifying abnormal emission measurement data includes: Causal analysis was performed on the energy consumption and emission concepts involved in the in-situ collected outlier data and the benchmark deviation items to construct a causal logic network for carbon emissions.

[0032] The semantic nodes corresponding to the in-situ collected outlier data and the benchmark deviation terms in the causal logic network are determined, and the feature sequences corresponding to the nodes are subjected to mean-dimensionless processing.

[0033] The causal relationship between two nodes is calculated using the grey relational analysis method.

[0034] A logical coupling threshold is set. When the overall gray correlation degree is less than the logical coupling threshold, it is judged as abnormal emission measurement data.

[0035] The time series segments of outlier data collected in situ and the indicators belonging to the benchmark deviation terms are extracted, and after z-score standardization, the dynamic time warping distance is used to measure the morphological matching degree of the two time series.

[0036] A time-series coupling threshold is set. When the dynamic time warp distance is greater than the time-series coupling threshold, it is determined to be abnormal emission measurement data.

[0037] According to the edge computing-based carbon emission measurement method for public institutions provided by this invention, the process of determining compliance attributes and performing attribution identification includes: Construct a knowledge ontology of carbon emission accounting rules.

[0038] Abnormal emission measurement data is mapped to a rule knowledge ontology to obtain a set of instances. Descriptive logic is used for consistent reasoning. If there is a contradiction in the reasoning, the data is determined to be non-compliant; otherwise, it is determined to be compliant.

[0039] Historical abnormal emission data and corresponding attribution samples are collected to construct a training set. An ensemble learning abnormal attribution recognition model based on gradient boosting tree is trained. The multi-dimensional features of abnormal emission metering data are input, and the abnormal emission attribution categories are output. The abnormal emission attribution categories include metering equipment failure, abnormal energy consumption loss, and inaccurate emission data.

[0040] On the other hand, the present invention also provides a carbon emission metering system for public institutions based on edge computing, comprising: The data acquisition module is used to collect multi-source energy consumption data and emission monitoring data from public institutions, and to obtain industry benchmark carbon emission data and meteorological correction data.

[0041] The emissions accounting module is used to perform temperature correction on the energy consumption baseline by combining meteorological correction data, perform component carbon emissions accounting based on the emission factor method, and generate hourly carbon emissions measurement sequences.

[0042] The anomaly screening module is used to identify outlier data collected in situ from multi-source energy consumption data, and to compare the hourly carbon emission measurement sequence with industry benchmark carbon emission data to screen for benchmark deviation items.

[0043] The source tracing and mapping module is used to extract feature identification fields from multi-source energy consumption data and emission monitoring data, construct hierarchical source tracing links between energy-consuming equipment, emission sources and functional areas, extract metering unit objects from multi-source energy consumption data and emission monitoring data, construct the topological connection relationship of each metering unit object based on the hierarchical source tracing links, and generate a carbon emission source tracing map of public institutions.

[0044] The coupling verification module is used to perform coupling verification of the time series segments of indicators to which the deviation items of outlier data collected in situ belong with the benchmark, and to identify abnormal emission measurement data.

[0045] The anomaly attribution module is used to determine the compliance attributes of abnormal emission measurement data according to carbon emission accounting standards, and to build an anomaly attribution identification model based on ensemble learning. It takes abnormal emission measurement data as input and outputs the anomaly emission attribution category.

[0046] The carbon emission measurement method and system for public institutions based on edge computing provided by this invention completes full-dimensional data collection, cleaning and verification locally through edge clusters, reducing the pressure of long-distance transmission of raw data, filtering out distorted and invalid data from the source, and solving the defects of centralized architecture such as dirty data interfering with accounting and excessive network load.

[0047] By correcting environmental conditions and supplementing the relevant emission accounting dimensions for personnel, the basic emission calculation benchmark is optimized, thus addressing the shortcomings of simply adding energy consumption values ​​and statistical results that deviate from the actual emission levels.

[0048] By constructing a hierarchical source tracing relationship and emission topology through unified data identification and matching, the emission source can be located layer by layer, which solves the problem that only macro total amount can be obtained and the emission source of a zone or individual device cannot be traced.

[0049] By using a multi-feature time-series joint discrimination mechanism to identify data anomalies, replacing the single threshold comparison method, it solves the defects of frequent misjudgments and omissions in traditional discrimination modes, realizes compliance judgment and automatic classification of anomalies, and solves the defects of fragmented process, reliance on manual verification of anomalies, and low processing efficiency. Attached Figure Description

[0050] The invention will now be further described with reference to the accompanying drawings.

[0051] Figure 1 This is a flowchart illustrating the carbon emission measurement method for public institutions based on edge computing in this invention. Figure 2 This is a flowchart illustrating the screening of benchmark deviation items in this invention; Figure 3 This is a schematic diagram of the process of extracting feature identification fields and constructing hierarchical tracing links in this invention; Figure 4 This is a schematic diagram of the process for generating carbon emission source maps of public institutions in this invention; Figure 5 This is a schematic diagram of the process for identifying abnormal emission measurement data in this invention; Figure 6 This is a schematic diagram of the structure of the carbon emission metering system for public institutions based on edge computing in this invention. Detailed Implementation

[0052] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0053] like Figures 1 to 6 As shown in the embodiments of the present invention, the carbon emission measurement method and system for public institutions based on edge computing can be implemented by an edge computing-based carbon emission measurement method for public institutions, the method including: Deploy edge computing node clusters to collect multi-source energy consumption data and emission monitoring data from public institutions, and obtain industry benchmark carbon emission data and meteorological correction data.

[0054] Edge computing nodes are deployed using a regional sharding architecture, located in the power distribution room, air conditioning room, low-voltage room on each floor, and near exhaust outlets of public institutions. Each node is responsible for data collection and local preprocessing of terminal devices within its corresponding physical area, and the nodes synchronize baseline parameters through a local area network.

[0055] The edge nodes use low-power edge gateways based on the ARM architecture as hardware carriers to connect with terminal metering devices such as smart meters, gas meters, water meters, heat meters, indoor CO2 concentration sensors, indoor temperature and humidity sensors, and flue gas flow detectors on site.

[0056] All terminal metering devices and edge nodes synchronize their clocks through a local area network NTP server, ensuring a unified time base for multi-source data.

[0057] Industry benchmark carbon emission data are obtained synchronously from publicly available industry benchmark databases. The data values ​​comply with the technical specifications for carbon emission accounting of general public institutions, and the corresponding benchmark parameters are matched according to building type.

[0058] Meteorological correction data is obtained through the public data interface of local meteorological observation stations. The data items include three parameters: hourly outdoor dry-bulb temperature, relative humidity, and atmospheric pressure. Among them, outdoor dry-bulb temperature and atmospheric pressure are used for standard condition conversion of natural gas volume and baseline correction of energy consumption during the heating season. Relative humidity is combined with dry-bulb temperature to calculate outdoor wet-bulb temperature, which is used for baseline correction of energy consumption of central air conditioning system during the cooling season.

[0059] If data synchronization fails, it will automatically retry 3 times with a 5-minute interval; if it fails 3 times in a row, a synchronization error log will be recorded, and the data after the breakpoint will be retransmitted after the network is restored to ensure the consistency of edge-cloud data.

[0060] Multi-source energy consumption data includes sub-item energy consumption metering data for electricity, natural gas, tap water, centralized heating, and central air conditioning systems within public institutions, as well as operating parameters of energy-consuming equipment and personnel density data for building functional areas.

[0061] Natural gas combustion falls under the category of direct emissions from the site, while electricity, tap water, and centralized heating fall under the category of indirect emissions from purchased energy sources.

[0062] The emission monitoring data includes real-time indoor CO2 concentration monitoring data and exhaust gas outlet concentration and flow monitoring data. Among them, exhaust gas outlet monitoring data corresponds to direct emission scenarios such as natural gas combustion, indoor CO2 concentration data is used for personnel density verification and respiratory emission calculation, and indoor temperature and humidity data are collected at the same location as indoor CO2 concentration data for water vapor correction of CO2 concentration.

[0063] Personnel density data is collected through building access control systems and infrared personnel counting devices. The number of people on duty is counted hourly by functional area to form an hourly personnel density sequence.

[0064] The load rate of energy-consuming equipment is calculated from the collected operating parameters. The calculation method is the ratio of the real-time active power of the equipment to the rated active power, with a value range of 0 to 1, reflecting the actual load level of the equipment.

[0065] Industry benchmark carbon emission data include benchmark values ​​for carbon emission intensity per unit building area of ​​public institutions, benchmark values ​​for per capita carbon emissions, and benchmark values ​​for carbon emission factors of various energy sources.

[0066] The meteorological correction data includes real-time data on outdoor dry-bulb temperature, relative humidity, and atmospheric pressure.

[0067] Data preprocessing is performed locally on the edge computing node, temperature correction is applied to the energy consumption baseline by combining meteorological correction data, and carbon emission accounting is performed based on the emission factor method to generate an hourly carbon emission measurement sequence.

[0068] The data preprocessing stage sequentially completes missing value completion, jump value removal, and duplicate record deduplication. For data segments with no more than 3 consecutive missing sampling points, linear interpolation is used to complete the missing values. For segments with more than 3 consecutive missing sampling points, they are directly marked as invalid data and removed.

[0069] The jump value determination uses the adjacent point difference method to calculate the numerical difference between the current sampling point and the previous sampling point. If the absolute value of the difference exceeds 50% of the average of the previous 10 sampling points, it is determined to be a jump value and is removed. When the average of the previous 10 sampling points is lower than 5% of the equipment's rated value (i.e., in a low-load state), the jump value determination threshold is switched to 2% of the equipment's rated value to avoid false judgments near zero. When there are fewer than 10 valid sampling points during the initial system startup, jump value removal is not performed temporarily; the determination is initiated after a total of 10 valid sampling points are accumulated.

[0070] Duplicate records are deduplicated using both the device's unique code and the collection timestamp, retaining only the valid record from the first collection.

[0071] For natural gas volume data, a standard condition conversion is performed by combining real-time atmospheric pressure and temperature data to convert the operating condition volume to the volume under standard conditions. The conversion follows the ideal gas law, with the standard condition reference being 20℃ and 101.325kPa. The conversion formula is expressed as follows: In the formula, This represents the volume of natural gas under standard conditions. This represents the measured volume of natural gas under operating conditions. This represents the measured atmospheric pressure under operating conditions. This is the atmospheric pressure under standard conditions, which is 101.325 kPa. The thermodynamic temperature under standard conditions is 293.15 K. The measured thermodynamic temperature under operating conditions.

[0072] The temperature correction factor is calculated based on the day-to-day method, and meteorological correction is applied to the energy consumption baseline. The calculation results are truncated to upper and lower limits, with the value range limited to between 0.8 and 1.2. The formula is as follows: In the formula, For temperature correction coefficients of a single type of energy, For the temperature influence coefficient of the corresponding energy, To correspond to the measured outdoor temperature under the corrected baseline, This is the reference temperature threshold for the corresponding energy source.

[0073] Seasonal division is determined by the five-day moving average temperature. When the five-day daily average temperature is below 10℃, the heating season begins; when the five-day daily average temperature is above 22℃, the cooling season begins; and the remaining periods are transitional seasons.

[0074] In the heating season scenario, the outdoor dry-bulb temperature is used as the calculation benchmark, with a benchmark temperature threshold of 18℃. In the cooling season scenario, the outdoor wet-bulb temperature is used as the calculation benchmark. The wet-bulb temperature is obtained iteratively from the outdoor dry-bulb temperature and relative humidity during the same period using the Magnus empirical formula. The formula for calculating saturated vapor pressure is: In the formula, This is the saturated vapor pressure, measured in kilopascals (kPa). The outdoor dry-bulb temperature is expressed in degrees Celsius. The actual vapor pressure is obtained by calculating the saturated vapor pressure and then combining it with relative humidity. The wet-bulb temperature is then iteratively calculated, with a reference temperature threshold of 19°C. The transitional season temperature correction factor is set to 1, and no temperature correction is applied.

[0075] The temperature influence coefficient for electric cooling is taken as 0.025 degrees Celsius, and the temperature influence coefficient for natural gas heating is taken as 0.032 degrees Celsius.

[0076] The total carbon emissions are calculated by summing the individual carbon emissions using the emission factor method, as expressed by the formula: In the formula, The total carbon emissions of public institutions during the statistical period are expressed in kilograms of carbon dioxide. Total number of energy types; For the first Activity level data for various energy types, with electricity activity level in kilowatt-hours, natural gas activity level in standard cubic meters, tap water activity level in cubic meters, and district heating activity level in gigajoules; For the first The carbon emission factors corresponding to each type of energy are as follows: the emission factor for electricity is the annual power supply emission factor published by the power grid in the region, which is uniformly converted to kilograms of carbon dioxide per kilowatt-hour; the emission factor for natural gas is 2.1622 kilograms of carbon dioxide per standard cubic meter; the emission factor for tap water is 0.91 kilograms of carbon dioxide per cubic meter; and the emission factor for centralized heating is 110 kilograms of carbon dioxide per gigajoule. For the first The temperature correction coefficients for different energy types are as follows: only electricity and natural gas are included in the temperature correction, while the correction coefficients for tap water and district heating are set to 1.

[0077] Based on the accounting results, an hourly carbon emission measurement sequence is generated, and the carbon emission intensity per unit building area and per capita carbon emission intensity are calculated.

[0078] Carbon emission intensity per unit building area is the ratio of total carbon emissions to total building area within the statistical period, while carbon emission intensity per capita is the ratio of total carbon emissions to the average number of employees on duty in the building at any given time within the statistical period. Both indicators are calculated on an hourly basis to form a continuous hourly measurement sequence.

[0079] At the same time, indoor CO2 concentration monitoring data is used to calculate the carbon emissions from human respiration, which serves as a supplementary item to the total emissions.

[0080] Indoor CO2 concentration monitoring data needs to be corrected for moisture content by combining indoor temperature and humidity parameters from the same period, and converted to volume concentration under standard dry air conditions. The correction formula is as follows: In the formula, This represents the volume concentration of CO2 in dry air. This represents the measured CO2 volume concentration. This represents the mole fraction of water vapor in the humid air, calculated from indoor temperature and relative humidity.

[0081] The calculation adopts the per capita emission factor method, with the per capita hourly respiratory carbon emission factor set at 0.035 kg of carbon dioxide per person per hour. The total respiratory emissions are calculated by combining hourly population density data.

[0082] Simultaneously, the indoor-outdoor concentration difference method was used for cross-validation. Respiratory emissions were calculated based on the building ventilation volume and the indoor-outdoor CO2 concentration difference. The building ventilation volume was read from the operating parameters of the fresh air handling unit in the building automation system, in cubic meters per hour. The outdoor CO2 background concentration was taken as the general atmospheric background value of 420 ppm. The formula is as follows: In the formula, Carbon emissions per hour of breathing, expressed in kilograms; This refers to the hourly ventilation volume, expressed in cubic meters per hour. The volume concentration of CO2 under dry indoor air conditions is expressed in ppm. Outdoor CO2 background volume concentration, in ppm; The density of CO2 under standard conditions is 1.977 kg per cubic meter. This is the conversion factor for converting ppm units to volume fractions.

[0083] If the deviation rate of hourly respiratory emissions obtained by the two methods is greater than 20%, it is marked as data to be verified. This type of data retains the mark, is not included in the official statistics of total emissions, and participates in the coupled verification together with the outlier data collected in situ.

[0084] Identify outlier data from in-situ collection in multi-source energy consumption data, and compare the hourly carbon emission measurement sequence with industry benchmark carbon emission data to screen for benchmark deviations.

[0085] The process of identifying outlier data collected in situ includes: The median of each indicator data sequence in the multi-source energy consumption data is calculated. The data sequence is constructed using a sliding time window. The window length is the historical sampling data of the past 168 hours, or 7 days. The window step size is the same as the sampling step size, which is 15 minutes. It is updated on a rolling basis as new data is added.

[0086] When the system is in its initial stage and there is less than 7 days of historical data, the statistical window is filled with the average value of the same period of the industry's general energy consumption benchmark series for the same building type and scale. After the accumulated valid data reaches 7 days, it is switched to a pure historical data sliding window mode.

[0087] The median of a data series is calculated. The median is used to characterize the central tendency of time-series data, and the formula is expressed as: The formula for calculating the median absolute deviation of a data series is as follows: In the formula, This indicates the number of data samples, corresponding to the total number of valid sampling points within the sliding window. Indicates the first One data sample, This represents the median of a data sequence. This represents the median of the absolute deviation, and the MAD correction factor is taken as the standard value of 1.4826.

[0088] The corrected robust standard deviation estimate is calculated using the following formula: In the formula, This represents the robust standard deviation estimate. This represents the MAD correction factor.

[0089] Set deviation threshold coefficient When data points satisfy If the deviation threshold is 3, then the data point is considered to be an outlier collected in situ.

[0090] The process of screening for benchmark deviations includes: The annual benchmark value of carbon emission intensity per unit building area corresponding to public institution scenarios is obtained from industry benchmark carbon emission data, and the benchmarking and verification are carried out on a monthly cumulative basis with the natural month as the statistical period.

[0091] A dynamic hourly benchmark is established, and statistics are classified by weekdays and holidays. Weekdays are Monday to Friday, and holidays are statutory holidays and Saturdays and Sundays. Adjusted days off are matched according to the actual workday attributes. The average of historical data of the same week type and time period without abnormal markers in the past 3 months of the building is taken as the hourly benchmark. The dynamic benchmark is updated once a month.

[0092] When the system is in its initial stage and there is less than 3 months of historical data, the industry benchmark value of the same type of building in the same period is used as a dynamic benchmark. After accumulating 3 months of valid historical data, it will switch to the historical data statistics mode.

[0093] The deviation rate between the real-time carbon emission intensity per unit building area of ​​public institutions and the benchmark value is calculated using the following formula: In the formula, Indicates the carbon emission intensity deviation rate. This indicates the carbon emission intensity per unit building area of ​​public institutions within a statistical period. This represents the benchmark value of carbon emission intensity per unit building area for the corresponding period.

[0094] The deviation rate threshold is set at 20%. When the deviation rate is greater than the deviation rate threshold, the emission monitoring indicator is considered to be a benchmark deviation item and marked as an abnormal candidate data to be verified.

[0095] Extracting feature identification fields from multi-source energy consumption data and emission monitoring data, and constructing a hierarchical source tracing link between energy-consuming equipment, emission sources, and functional areas, the process includes: An edge semantic parsing model is used to extract features from unstructured monitoring logs and identify data attributes and physical meanings.

[0096] This model is a named entity recognition model obtained by distillation based on the BiLSTM-CRF architecture. The model input is the standard format of the equipment operation log and alarm text, and the output is three types of structured entities: equipment number, fault type, and operating status, which are added to the attribute fields of the metering data.

[0097] Based on the data attributes, fields that can uniquely identify metering records and coexist in multi-source energy consumption data and emission monitoring data are selected as feature identification fields. Feature identification fields include equipment unique identifier code, spatial area code, energy consumption type label and collection timestamp.

[0098] The unique identification code of the equipment adopts a seven-level coding rule. The first two digits are the building code, the third digit is the building zone code, the fourth to sixth digits are the floor code, the seventh to tenth digits are the equipment serial number, and the spatial area code corresponds to the functional zone of the building, such as office area, meeting room, computer room, public corridor, etc.

[0099] Energy consumption type labels are divided into five categories: electricity, natural gas, tap water, centralized heating, and central air conditioning, with each category corresponding to an independent type code.

[0100] The multi-source data is aligned in time sequence and the sampling granularity is unified according to the timestamp field. The time alignment benchmark is the whole 15-minute time and the statistical rule of left-closed and right-open time window is adopted.

[0101] A resampling method is used to convert data from different collection frequencies into a unified 15-minute granularity. For high-frequency emission monitoring data, the arithmetic mean within the 15-minute window is taken as the value at that granularity. For low-frequency energy consumption data, a forward-filling method is used to complete the granularity matching, so that all data correspond one-to-one in the time dimension.

[0102] The feature identifier fields are matched group by group according to the hierarchical relationship of functional area-energy-consuming equipment-emission source. If the fields match completely, it is determined that there is a source tracing connection between the corresponding units.

[0103] The first layer of matching establishes a subordinate association between functional areas and all energy-consuming equipment based on spatial area coding. The second layer of matching establishes a transmission association between energy-consuming equipment and corresponding on-site emission sources for direct emission energy sources such as natural gas, and between energy-consuming equipment and upstream purchased energy emission nodes for indirect emission energy sources such as electricity and tap water.

[0104] For multiple similar devices in the same area, the emission weights of the corresponding emission sources are allocated according to the proportion of the energy consumption of a single device to the total energy consumption of the area.

[0105] Establish a hierarchical source tracing and matching table to record the corresponding relationship between functional areas, energy-consuming equipment, and emission sources. The table is updated daily.

[0106] The process involves extracting metering unit objects from multi-source energy consumption data and emission monitoring data, constructing a topological connection relationship between each metering unit object based on the corresponding hierarchical traceability link, and generating a carbon emission traceability map for public institutions. The process includes: Energy-consuming equipment, emission outlets, functional areas, and purchased energy nodes within public institutions are defined as metering unit objects. The subordinate relationship of equipment to its region, the transmission relationship of energy consumption to emission conversion, and the correspondence of indirect emissions from purchased energy are defined as topological connecting edges.

[0107] Graph topology storage uses an attribute graph model. Each node has four basic attributes: cell name, unique code, cell type, and rated parameters. Each edge has three attributes: connection type, contribution weight, and statistical period.

[0108] Edge nodes use an adjacency list structure to store the subgraph of the area they belong to, while the cloud uses a graph database to store the complete graph of the entire building, supporting subgraph query and traversal by region and by device type.

[0109] Nodes represent measurement unit objects, and edges represent topological connections. A graph topology storage structure is constructed based on a hierarchical traceability matching table. The overall graph adopts an edge-cloud collaborative storage method. Edge nodes store sub-graphs of their respective jurisdictions, while the cloud stores the complete graph of all buildings. The graph is updated daily, and the newly added node and edge attribute data are synchronized at midnight every day.

[0110] For each topological connection edge, the connection contribution weight is calculated based on the carbon emission proportion and equipment runtime proportion of the corresponding path, and is used as the attribute value of the edge. The formula is expressed as follows: In the formula, Represents a unit of measurement object With measurement unit object The contribution weight of the connection between them This indicates the carbon emissions corresponding to the connection path. This represents the total carbon emissions within the statistical area. This indicates the cumulative runtime of the corresponding device. This indicates the total duration within the statistical period. , These are the weighting coefficients for carbon emission share and runtime share, respectively.

[0111] Weighting coefficient Take 0.6, The value is set to 0.4, and the weight range is from 0 to 1.

[0112] The process of coupling and verifying the time series segments of indicators to which outlier data collected in situ belongs with the benchmark deviation term to identify abnormal emission measurement data includes: Causal analysis was performed on the energy consumption and emission concepts involved in the in-situ collected outlier data and the benchmark deviation items to construct a causal logic network for carbon emissions. ,in, This represents a set of concepts related to carbon emissions. Represents the set of edges with causal relationships.

[0113] The concept set includes seven categories of concepts: energy load, equipment load level, ambient temperature, population density, emission intensity, emission concentration, and equipment operating time. The causal relationship edge set includes five categories of causal relationships: energy load positively drives emission intensity, equipment load level positively drives emission intensity, ambient temperature positively drives energy load, population density positively drives energy load, and emission concentration positively correlates with emission intensity.

[0114] The mapping rules between outlier data and semantic nodes are as follows: If the outlier data collected in situ belongs to the energy load category, the corresponding semantic nodes are energy load, equipment load level, ambient temperature, and personnel density, and the corresponding feature sequences are energy power sequence, equipment load rate sequence, outdoor temperature sequence, and personnel density sequence, respectively; if the benchmark deviation item belongs to the emission intensity category, the corresponding semantic nodes are emission intensity, energy load, and equipment load level, and the corresponding feature sequences are carbon emission intensity sequence, energy power sequence, and equipment load rate sequence, respectively; if emission concentration monitoring data is involved, the corresponding semantic nodes are emission concentration, emission intensity, and energy load, and the corresponding feature sequences are emission concentration sequence, carbon emission intensity sequence, and energy power sequence, respectively.

[0115] The semantic nodes corresponding to the in-situ collected outlier data and the benchmark deviation term in the causal logic network are determined. The feature sequences corresponding to the nodes are averaged and dimensionless. The statistical window takes 24-hour time segments corresponding to 48 sampling points before and after the outlier. The formula is expressed as follows: In the formula, , The two types of nodes after dimensionless transformation are respectively the first... 3D eigenvalues , These are the original feature sequence values, , These are the mean values ​​of the corresponding feature sequences.

[0116] The causal relationship between two nodes was calculated using the grey relational analysis method. The total number of feature dimensions was 5, namely the sequence mean, sequence variance, peak coefficient, hourly rate of change, and percentage of working hours.

[0117] The calculation formulas for each feature dimension are as follows: The maxima coefficient is used to characterize the severity of peak fluctuations in time series data, and the formula is expressed as: In the formula, Peak coefficient, The maximum value in the sequence. This represents the sequence mean. When the sequence mean is less than 1% of the corresponding indicator's rated value, the peak coefficient is set to 1.

[0118] The hourly rate of change is used to quantify the overall fluctuation range of time-series data hourly. Before calculation, the sampling data at the 15-minute granularity is first averaged by natural hours to obtain an hourly series. Then, the hourly rate of change is calculated based on the hourly series. The formula is as follows: In the formula, The rate of change is per hour. This represents the total number of hours within the statistical period. For the first Hourly series mean.

[0119] The percentage of working hours is used to characterize the distribution of data anomalies within the effective working hours, and the formula is expressed as: In the formula, The percentage of working hours. This is the set of sampling points corresponding to the period from 8:00 to 18:00 each day. This represents the total number of sampling points within the statistical period.

[0120] The grey relational coefficient is used to calculate the degree of correlation and fit between two sets of time series data under a single feature dimension. The formula is expressed as: In the formula, For the first Grey correlation coefficients under each feature dimension The resolution coefficient is set to the universally optimal value of 0.5.

[0121] The overall grey relational degree is used to comprehensively characterize the overall causal relationship between two sets of time-series data under multi-dimensional features. It is calculated from the average correlation coefficient of each dimension, and the formula is expressed as: In the formula, Indicates the overall gray relational degree. This indicates the total number of feature dimensions.

[0122] The logical coupling threshold is set to 0.6. When the overall gray correlation degree is less than the logical coupling threshold, it is judged as abnormal emission measurement data.

[0123] Extract time series segments of outlier data collected in situ and the indicators to which the benchmark deviation item belongs. The segment length is 48 sampling points before and after the outlier, for a total of 24 hours of time series data.

[0124] After z-score normalization, the dynamic time warp distance is used to measure the morphological matching degree between two time series. The formula for z-score normalization is as follows: In the formula, , These are the standardized time series values. , These are the original timing values, , These are the mean values ​​for the corresponding time series. , These are the standard deviations of the corresponding time series.

[0125] The dynamic time warping distance is used to measure the morphological similarity between two time series curves. A Sakoe-Chiba band constraint is used to limit the warping path offset range, expressed by the following formula: In the formula, This represents the dynamic time-warped distance between two time series. This represents the optimal normalized path between two sequences. This represents the Euclidean distance between points in the sequence.

[0126] The window value is derived based on a 15-minute sampling granularity. The five sampling points correspond to a 75-minute time span, covering the conventional transmission lag time from changes in energy load of public institutions to emission response. This ensures the physical rationality of timing matching while avoiding excessive path distortion that could lead to matching distortion.

[0127] The temporal coupling threshold is obtained through the statistics of historical normal samples. Historical normal samples are selected from time series data of the same period without abnormal markers in the past 30 days. The effective sample size is no less than 30 groups. The arithmetic mean of the DTW distance of all normal sample pairs is calculated. The threshold is 1.5 times the mean. The threshold is updated once a quarter.

[0128] When the system is first deployed without historical data, the initial value of the time-series coupling threshold is 1.2. This value is set for 24-hour and 15-minute granular energy consumption time-series data after z-score standardization. It is a common initial reference value for similar scenarios. After the accumulated valid samples meet the requirements, it will automatically switch to the local statistical value.

[0129] When the dynamic time warp distance is greater than the time coupling threshold, it is determined to be abnormal emission measurement data.

[0130] If either of the two types of verification is deemed mismatched, it is marked as abnormal emission measurement data; if both are deemed mismatched, it is marked as high-priority abnormal.

[0131] High-priority anomalies are pushed to the top of the alarm queue on the cloud management platform and displayed first.

[0132] The compliance attributes of abnormal emission measurement data are determined according to the general technical specifications for carbon emission accounting, and an anomaly attribution identification model based on ensemble learning is constructed. The model takes abnormal emission measurement data as input and outputs the anomaly emission attribution category. The process includes: Constructing a knowledge ontology of carbon emission accounting rules ,in, Represents the set of accounting concepts. Represents a set of constraint relationships between concepts. Represents a collection of attributes.

[0133] The accounting concept set includes six categories of concepts: accounting boundary, direct emissions, indirect emissions, emission factors, emission source classification, and accounting cycle. The constraint relationship set includes four categories of constraints: inclusion relationship, conversion relationship, limit constraint relationship, and value constraint relationship. The attribute set includes five attributes: emission threshold, factor value range, accounting cycle duration, reporting time limit, and boundary range.

[0134] The ontology is constructed in accordance with the OWL2DL specification, and the rule axioms are set according to the general public agency carbon emission accounting technical specifications. The executable reasoning rules include emission factor value constraint rules, monthly emission intensity limit constraint rules, accounting boundary determination rules, and reporting cycle verification rules.

[0135] The emission factor value constraint rules are defined as follows: if the value of an emission factor exceeds the general value range for the corresponding energy type, the factor is deemed to have a value violation. The general value ranges for emission factors of various energy types are: electricity 0.5 to 0.9 kg CO2 per kilowatt-hour, natural gas 2.0 to 2.3 kg CO2 per standard cubic meter, tap water 0.8 to 1.0 kg per cubic meter, and centralized heating 90 to 130 kg per gigajoules. The monthly emission intensity limit constraint rules are defined as follows: if the monthly carbon emission intensity per unit building area of ​​a building exceeds 1.2 times the annual benchmark value for the corresponding type of building, the building is deemed to have an emission intensity violation. The accounting boundary judgment rule is defined as follows: if the emission source corresponding to the accounting data exceeds the statutory accounting boundary range of public institutions, the accounting boundary violation is deemed to exist. The reporting cycle verification rule is defined as follows: if the data statistical cycle exceeds the prescribed time limit for accounting reporting, the reporting cycle violation is deemed to exist.

[0136] Abnormal emission measurement data is mapped to a rule knowledge ontology to obtain a set of instances. Descriptive logic is used for consistent reasoning. If there is a contradiction in the reasoning, the data is determined to be non-compliant; otherwise, it is determined to be compliant.

[0137] The reasoning is performed using the Pellet inference engine. The attributes of the abnormal data, such as emissions, emission factors, accounting boundaries, and time periods, are used as assertion inputs and verified one by one with the axioms and rules in the ontology. If logical contradictions occur, such as emissions exceeding the limit requirements, emission factor values ​​exceeding the general range, accounting boundaries not conforming to the specifications, or accounting periods exceeding the expiration date, the emission behavior corresponding to the abnormal data is determined to be non-compliant.

[0138] Both compliance reasoning and attribution model reasoning are executed in the cloud, while edge nodes only complete the initial screening of anomalies and local data preprocessing.

[0139] A training set was constructed by collecting historical abnormal emission data and corresponding attribution samples. The sample annotation was completed by professionals with experience in carbon emission accounting, and the annotation was based on on-site verification results and equipment calibration records.

[0140] Abnormal emissions are classified into three categories: metering equipment failure, abnormal energy consumption loss, and inaccurate emission data. Metering equipment failure corresponds to scenarios such as unqualified metering device calibration and data jumps without corresponding changes in operating conditions. Abnormal energy consumption loss corresponds to scenarios such as decreased equipment energy efficiency, pipeline leakage, and increased ineffective energy consumption. Inaccurate emission data corresponds to scenarios such as conflicting data from multiple sources and data tampering.

[0141] The ensemble learning anomaly attribution recognition model based on gradient boosting trees was trained. The input features consisted of 12 dimensions, namely, degree of anomaly deviation, duration of anomaly, time period of occurrence, number of associated devices, meteorological matching degree, frequency of historical anomalies, proportion of sequence variance, peak deviation multiple, proportion of working time period, regional correlation, device type, and data source type.

[0142] The quantification methods for each feature are as follows: Abnormal deviation degree is the deviation rate between the abnormal value and the benchmark value; Abnormal duration is the total duration corresponding to consecutive abnormal sampling points, in minutes; Occurrence time is numerically coded into four categories: weekday daytime, weekday nighttime, holiday daytime, and holiday nighttime; Number of associated devices is the total number of terminal metering devices associated with the abnormality; Meteorological matching degree is the grey relational degree between energy consumption data and temperature data; Historical abnormality frequency is the number of times the same type of abnormality occurred in the same device within the past 30 days; Sequence variance ratio is the ratio of the variance of the data sequence during the abnormal period to the variance of the benchmark sequence during the same period; Peak deviation multiple is the ratio of the abnormal peak value to the mean of the benchmark sequence during the same period; Working period ratio is the proportion of the working period duration to the total abnormal duration during the abnormal duration; Regional correlation is the proportion of the number of devices associated with the abnormal region to the total number of devices in the region; Equipment type is numerically coded into four categories: electricity, natural gas, tap water, and centralized heating; Data source type is numerically coded into two categories: terminal metering devices and system-calculated data.

[0143] The model was trained using five-fold cross-validation, with the core hyperparameters set as follows: maximum decision tree depth of 6, learning rate of 0.05, number of iterations of 200, and minimum number of samples per leaf node of 10.

[0144] The model outputs the category along with the corresponding confidence level. The confidence level threshold is 0.7. Abnormal data with a confidence level lower than this threshold are automatically marked as requiring manual review.

[0145] The system integrates data from multiple sources to generate a full dataset for carbon emission measurement, and after anomaly information is marked, it is synchronized to the cloud-based management and control platform.

[0146] The full dataset consists of three data layers: the raw data collection layer, the accounting result layer, and the anomaly labeling layer. The raw data collection layer stores all preprocessed raw energy consumption and emission data. The accounting result layer stores hourly carbon emission measurement sequences and various derived indicators. The anomaly labeling layer stores information such as the location, type, compliance attributes, attribution category, and confidence level of abnormal data.

[0147] Carbon emission accounting is completed based on all valid data. Identified abnormal data retains its original value and adds anomaly markers, and does not participate in dynamic benchmark calculation and model training.

[0148] Each data record is accompanied by a complete timestamp, region code, device code, value, and unit field. Data synchronization adopts an incremental synchronization mechanism, which synchronizes the newly added metering data of the edge nodes to the cloud platform every hour.

[0149] The cloud-based management platform has functions such as data query, statistical analysis, anomaly warning, and automatic report generation, which can provide data support for public institutions' carbon emission management, energy-saving renovation, and carbon emission accounting and reporting.

[0150] In summary, this embodiment provides a carbon emission measurement method for public institutions based on edge computing. It completes full-dimensional data collection, cleaning, and verification locally through edge clusters, reducing the pressure of long-distance transmission of raw data, filtering out distorted and invalid data from the source, and solving the defects of centralized architecture such as dirty data interfering with accounting and excessive network load.

[0151] By correcting environmental conditions and supplementing the relevant emission accounting dimensions for personnel, the basic emission calculation benchmark is optimized, thus addressing the shortcomings of simply adding energy consumption values ​​and statistical results that deviate from the actual emission levels.

[0152] By constructing a hierarchical source tracing relationship and emission topology through unified data identification and matching, the emission source can be located layer by layer, which solves the problem that only macro total amount can be obtained and the emission source of a zone or individual device cannot be traced.

[0153] By using a multi-feature time-series joint discrimination mechanism to identify data anomalies, replacing the single threshold comparison method, it solves the defects of frequent misjudgments and omissions in traditional discrimination modes, realizes compliance judgment and automatic classification of anomalies, and solves the defects of fragmented process, reliance on manual verification of anomalies, and low processing efficiency.

[0154] Based on the same general inventive concept, this invention also protects a carbon emission measurement system for public institutions based on edge computing. The carbon emission measurement system for public institutions based on edge computing provided by this invention will be described below. The carbon emission measurement system for public institutions based on edge computing described below can be referred to in correspondence with the carbon emission measurement method and system for public institutions based on edge computing described above.

[0155] The carbon emission measurement system for public institutions based on edge computing includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The processor includes a data acquisition module, an emission accounting module, an anomaly screening module, a source tracing and mapping module, a coupling verification module, and an anomaly attribution module.

[0156] The data acquisition module is used to collect multi-source energy consumption data and emission monitoring data from public institutions, and to obtain industry benchmark carbon emission data and meteorological correction data.

[0157] The emissions accounting module is used to perform temperature correction on the energy consumption baseline by combining meteorological correction data, perform component carbon emissions accounting based on the emission factor method, and generate hourly carbon emissions measurement sequences.

[0158] The anomaly screening module is used to identify outlier data collected in situ from multi-source energy consumption data, and to compare the hourly carbon emission measurement sequence with industry benchmark carbon emission data to screen for benchmark deviation items.

[0159] The source tracing and mapping module is used to extract the feature identification fields of multi-source energy consumption data and emission monitoring data, construct hierarchical source tracing links between energy-consuming equipment, emission sources and functional areas, extract metering unit objects from multi-source energy consumption data and emission monitoring data, construct the topological connection relationship of each metering unit object based on the hierarchical source tracing links, and generate a carbon emission source tracing map of public institutions.

[0160] The coupling verification module is used to perform coupling verification of the time series segments of indicators to which the deviation items of the in-situ collected outlier data and the benchmark belong, and to identify abnormal emission measurement data.

[0161] The anomaly attribution module is used to determine the compliance attributes of abnormal emission measurement data according to carbon emission accounting standards, and to build an anomaly attribution identification model based on ensemble learning. It takes abnormal emission measurement data as input and outputs the anomaly emission attribution category.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A carbon emission measurement method for public institutions based on edge computing, characterized in that, include: Collect multi-source energy consumption data and emission monitoring data from public institutions, and obtain industry benchmark carbon emission data and meteorological correction data; By combining meteorological correction data to perform temperature correction on the energy consumption baseline, and by performing component carbon emission accounting based on the emission factor method, an hourly carbon emission measurement sequence is generated. Identify outlier data from in-situ collection in the multi-source energy consumption data, and compare the hourly carbon emission measurement sequence with industry benchmark carbon emission data to screen for benchmark deviation items; Extract the feature identification fields from the multi-source energy consumption data and emission monitoring data, and construct a hierarchical traceability link between energy-consuming equipment, emission sources and functional areas; Extract metering unit objects from the multi-source energy consumption data and emission monitoring data, construct the topological connection relationship of each metering unit object based on the corresponding link of the hierarchical source tracing, and generate a carbon emission source tracing map of public institutions; The in-situ collected outlier data and the time series segment of the index to which the deviation item belongs are coupled and verified to identify abnormal emission measurement data; The compliance attributes of abnormal emission measurement data are determined according to carbon emission accounting standards, and an abnormal attribution identification model based on ensemble learning is constructed. The abnormal emission measurement data is input and the abnormal emission attribution category is output.

2. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The multi-source energy consumption data includes sub-item energy consumption metering data for electricity, natural gas, tap water, centralized heating, and central air conditioning systems within public institutions, as well as operating parameters of energy-consuming equipment and personnel density data for building functional areas; the emission monitoring data includes real-time monitoring data of indoor CO2 concentration and monitoring data of exhaust gas emission outlet concentration and flow rate; the industry benchmark carbon emission data includes benchmark values ​​for carbon emission intensity per unit building area of ​​public institutions, benchmark values ​​for per capita carbon emissions, and benchmark values ​​for carbon emission factors of various energy sources; the meteorological correction data includes real-time data of outdoor dry-bulb temperature, relative humidity, and atmospheric pressure.

3. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The process of generating hourly carbon emission measurement sequences includes: The energy consumption baseline is meteorologically corrected based on the temperature correction factor calculated using the day-per-day method. The emission factor method was used to calculate the individual carbon emissions and then summed to obtain the total carbon emissions. Based on the accounting results, an hourly carbon emission measurement sequence is generated, and the carbon emission intensity per unit building area and per capita carbon emission intensity are calculated.

4. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The process of identifying outlier data collected in situ includes: Calculate the median of each indicator data sequence in the multi-source energy consumption data; Calculate the median of the absolute deviation of the data series; Calculate the corrected robust standard deviation estimate; A deviation threshold coefficient is set, and outlier data collected in situ are identified based on the median and robust standard deviation estimates.

5. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The process of screening for benchmark deviations includes: Obtain the benchmark value of carbon emission intensity per unit building area corresponding to the public institution scenario from the aforementioned industry benchmark carbon emission data; Calculate the deviation rate between the real-time carbon emission intensity per unit building area of ​​public institutions and the benchmark value; A deviation rate threshold is set. When the deviation rate is greater than the deviation rate threshold, the emission monitoring indicator is considered to be a benchmark deviation item.

6. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The process of extracting feature identifier fields and constructing hierarchical tracing links includes: The multi-source energy consumption data and the emission monitoring data are preprocessed to remove missing frames, jump values ​​and duplicate sampling records; An edge semantic parsing model is used to extract features from unstructured monitoring logs and identify data attributes and physical meanings. Based on the data attributes, a field that can uniquely identify metering records and coexists in multi-source energy consumption data and emission monitoring data is selected as a feature identification field. The feature identification field includes a unique equipment identifier code, a spatial area code, an energy consumption type label, and a collection timestamp. Multi-source data are aligned in time sequence and sampled in a uniform granularity according to the timestamp field. Feature identifier fields are matched group by group according to the hierarchical relationship of functional area-energy-consuming equipment-emission source. If the fields match completely, it is determined that there is a source tracing connection between the corresponding units. Establish a hierarchical source tracing and matching table to record the corresponding connections between functional areas, energy-consuming equipment, and emission sources.

7. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The process of generating a carbon emission source map for public institutions includes: Energy-consuming equipment, emission outlets, and functional areas within public institutions are defined as metering unit objects, and the subordinate relationship of equipment to its area and the transmission relationship of energy consumption to emission conversion are defined as topological connection edges. Nodes represent measurement unit objects, edges represent topological connections, and a graph topology storage structure is constructed based on a hierarchical tracing and matching lookup table. For each topological connection edge, the connection contribution weight is calculated based on the carbon emission ratio and equipment runtime ratio of the corresponding path, and is used as the attribute value of the edge.

8. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The process of identifying abnormal emission measurement data includes: The in-situ collected outlier data and the benchmark deviation items are used to perform causal analysis on the energy consumption and emission concepts involved in the indicators, and a carbon emission causal logic network is constructed. Determine the semantic nodes in the causal logic network corresponding to the in-situ collected outlier data and the benchmark deviation term, and perform mean-scaled dimensionless processing on the feature sequences corresponding to the nodes. The causal relationship between two nodes is calculated using the grey relational analysis method. A logical coupling threshold is set. When the overall gray correlation degree is less than the logical coupling threshold, it is judged as abnormal emission measurement data. The time series segments of outlier data collected in situ and the indicators belonging to the benchmark deviation terms are extracted, and after z-score standardization, the dynamic time warping distance is used to measure the morphological matching degree of the two time series. A time-series coupling threshold is set. When the dynamic time warp distance is greater than the time-series coupling threshold, it is determined to be abnormal emission measurement data.

9. The carbon emission measurement method for public institutions based on edge computing according to claim 1, characterized in that, The process of determining compliance attributes and identifying causes includes: Construct a knowledge ontology of carbon emission accounting rules; Abnormal emission measurement data is mapped to a rule knowledge ontology to obtain an instance set. Consistent reasoning is performed using descriptive logic. If there is a contradiction in the reasoning, the data is determined to be non-compliant; otherwise, it is determined to be compliant. A training set is constructed by collecting historical abnormal emission data and corresponding attribution samples. An ensemble learning abnormal attribution recognition model based on gradient boosting tree is trained. The multi-dimensional features of abnormal emission metering data are input, and the abnormal emission attribution categories are output. The abnormal emission attribution categories include metering equipment failure, abnormal energy consumption loss, and inaccurate emission data.

10. A carbon emission metering system for public institutions based on edge computing, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the edge computing-based carbon emission measurement method for public institutions as described in any one of claims 1 to 9.

Images