Energy consumption data verification completion method and system for park carbon emission accounting and medium

Abstract

The application discloses a kind of energy consumption data verification completion methods, systems and media for park carbon emission accounting, belong to the cross field of green low carbon technology and intelligent data management.This scheme respectively establishes the energy consumption data physical topology model covering each energy type according to park energy supply network structure, and defines physical conservation constraint rules for different energy types;Statistical anomaly detection, topology anomaly detection and mode anomaly detection are carried out on energy consumption time series data;Root cause classification is carried out on abnormal data to obtain root cause label;According to the root cause label and the energy type to which the abnormal data belongs, select the corresponding intelligent completion strategy, repair the abnormal data, calculate the confidence score for each repaired data point, and include the data points with a confidence score higher than the preset score threshold in the carbon emission accounting data set.The scheme of the application supports multiple energy types, significantly improves the quality and reliability of carbon emission accounting data, and reduces manual dependence.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of green and low-carbon technology and intelligent data governance, and in particular relates to a method, system and medium for verifying and completing energy consumption data for carbon emission accounting in industrial parks. Background Technology

[0002] Industrial parks are crucial units for carbon emission management, and the accuracy of their carbon emission accounting highly depends on the quality of underlying energy consumption data. This data typically originates from timed reporting by smart meters, real-time aggregation by data acquisition systems, or manual data recording.

[0003] Currently, methods for quality management of energy consumption data typically employ threshold alarms, simple interpolation, or basic rule verification. Chinese patent CN120086213A discloses a layered cleaning method for building electricity consumption data. This method stratifies data according to equipment function type, identifies anomalies through clustering algorithms and inter-layer cross-verification, and fills in missing data using interpolation or support vector regression based on duration. This method has shown some effectiveness in improving the quality of building electricity consumption data. However, when the goal of data governance is focused on meeting the stringent requirements of high compliance and high auditability in carbon emission accounting, existing technologies, including the aforementioned Chinese patent, still have the following significant shortcomings: (1) Single energy type. Most existing solutions only target building electricity consumption data, while carbon emission accounting in the park needs to integrate multiple energy data such as electricity, water, gas, and heat. The metering characteristics, anomaly patterns, and physical constraints of different energy sources are very different, and a single electricity consumption processing model cannot be directly transferred.

[0004] (2) The topology is simple and lacks physical constraints. For example, the layering of Chinese patent CN120086213A is mainly based on the business logic of the device type, rather than the actual physical topology connection relationship. Its inter-layer mutual verification only checks whether the numerical difference is within the meter error range, without introducing a more universal and stricter physical conservation law, resulting in insufficient ability to detect hidden anomalies such as minor leaks and meter drift.

[0005] (3) The repair strategies are passive and lack specificity. For example, Chinese patent CN120086213A simply divides missing values ​​into long-term missing values ​​and local missing values, and uses SVR prediction and interpolation methods respectively. This classification ignores the root cause of the anomaly and will introduce a large error.

[0006] (4) The results are unreliable and untraceable. For example, the cleaning result of Chinese patent CN120086213A is a definite value, but no assessment of the reliability of the value is provided. In the scenario of carbon emission accounting, which has extremely high requirements for data quality, the direct participation of an unconfidence-assessed complete value in the calculation will introduce uncertainty into the final carbon emission result, which cannot meet the audit and compliance requirements.

[0007] Therefore, there is an urgent need for a comprehensive data verification and completion method that can overcome the above-mentioned shortcomings and deeply integrate physical topology, multidimensional anomaly diagnosis, root cause-driven repair, and confidence assessment. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a method, system and medium for verifying and completing energy consumption data for carbon emission accounting in industrial parks, in view of the above-mentioned defects of the prior art.

[0009] To achieve the above objectives, in a first aspect, the present invention provides a method for verifying and completing energy consumption data for carbon emission accounting in industrial parks, comprising: S1: Based on the energy supply network structure of the park, establish physical topology models of energy consumption data covering electricity, water, gas and heat energy types respectively. The topology model includes park-level, building-level, branch-level and equipment-level nodes; and define physical conservation constraint rules specific to the nodes in the physical topology model for different energy types. S2: Access the energy consumption time-series data of each node in the physical topology model, and perform data alignment and unit standardization; S3: Perform anomaly detection on the energy consumption time series data, including statistical anomaly detection based on historical data of the same period, topological anomaly detection based on the physical conservation constraint rules, and pattern anomaly detection based on time series pattern learning; S4: Based on the results of the anomaly detection, combined with the characteristics of the anomaly data, the energy type, and the physical conservation constraint rules, the anomaly data is classified into root causes to obtain root cause labels; S5: Based on the root cause label and the energy type to which the abnormal data belongs, select the corresponding intelligent completion strategy from the preset completion strategy set to repair the abnormal data. During the repair process, calculate the confidence score for each repaired data point. Only data points with confidence scores higher than the preset score threshold are included in the final dataset for carbon emission accounting.

[0010] In the energy consumption data verification and completion method for carbon emission accounting in industrial parks of the present invention, step S1, defining the physical conservation constraint rules specific to the nodes in the physical topology model for different energy types, includes: For the physical topology model of electricity, gas, and heat energy, the physical conservation constraints include: additivity constraints, monotonicity constraints, and dimensional consistency constraints. For the physical topology model of water energy, the physical conservation constraints include: additive constraints and monotonic constraints.

[0011] In the energy consumption data verification and completion method for carbon emission accounting in industrial parks of the present invention, step S3, the abnormal mode detection specifically involves: using a long short-term memory autoencoder model to perform unsupervised learning on historical normal energy consumption data to obtain a normal energy consumption mode model; inputting the time series data to be detected into the model, calculating its reconstruction error, and if the reconstruction error exceeds a preset error threshold, it is determined to be abnormal.

[0012] In the energy consumption data verification and completion method for carbon emission accounting in industrial parks of the present invention, in step S4, the input features of the root cause classification include missing patterns, deviation direction, abnormal duration, position in the physical topology model, and energy type identifier; the output root cause labels include communication interruption, meter drift, meter jamming, and topology configuration error; for water and gas energy, it also includes pipeline leakage; for thermal energy, it also includes temperature sensor drift and flow meter failure.

[0013] In the energy consumption data verification and completion method for carbon emission accounting in industrial parks of the present invention, step S5, selecting the corresponding intelligent completion strategy, includes: If the root cause label is communication interruption, a spatiotemporal collaborative prediction strategy is adopted, which uses complete data, meteorological data and calendar features of other nodes at the same or adjacent levels to predict missing data through a spatiotemporal transformer model. If the root cause label is meter drift, a physical topology reverse calculation strategy is adopted, which uses the sum of the energy consumption data of its child nodes and a preset physical loss rate to reverse calculate the energy consumption data of the node. If the root cause label is meter stagnation, a time-series mode repair strategy is adopted, which uses the normal energy consumption mode model to generate an alternative data sequence that conforms to the historical change pattern. If the root cause label is a pipeline leak, for water energy, the original reading is not completed, a leakage correction amount is generated and added to the parent node's master table, and a maintenance work order is triggered; for gas energy, automatic completion is not performed, only abnormal events are recorded and high-priority safety alarms are pushed. If the root cause label indicates temperature sensor drift and the flow data is normal, then the historical average supply and return water temperature difference at that metering point is used, combined with the measured flow rate, to reconstruct the heat. If the root cause label is a flow meter malfunction, the flow rate is inferred by using a correlation model between heat and pump frequency or valve opening, and then the heat is calculated.

[0014] In the energy consumption data verification and completion method for carbon emission accounting in industrial parks of the present invention, the intelligent completion strategy further includes a cross-energy collaborative completion strategy; when energy consumption data of a first energy type is determined to be missing or abnormal, and there is energy consumption data of a second energy type that is strongly correlated with the first energy-consuming device in physical or operational logic, the following steps are performed: Based on historical normal operation data, a correlation model between the consumption of the first energy type, the consumption of the second energy type, and external environmental parameters is trained. During periods when the first energy data is missing or abnormal, obtain complete consumption data for the second energy type and corresponding external environmental parameters. The consumption data of the second energy type and external environmental parameters are input into the correlation model to predict the complete data sequence of the first energy type.

[0015] In the energy consumption data verification and completion method for carbon emission accounting in the present invention, in step S5, the calculation of the confidence score integrates the prediction variance of the completion strategy itself, the matching degree between the completion result and the historical pattern, whether the completion result satisfies the physical conservation constraint rule in the physical topology model, and the inherent accuracy level of the metering instrument for this energy type.

[0016] The energy consumption data verification and completion method for carbon emission accounting in industrial parks of the present invention further includes a multi-energy data aggregation and cross-verification step: integrating all energy data after restoration, converting the consumption of different energy sources into a unified carbon dioxide equivalent based on the carbon emission factors of various energy sources, and using the macro trend of total energy consumption and total carbon emission in the industrial park to perform cross-verification of data rationality across energy sources.

[0017] Secondly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, it implements the energy consumption data verification and completion method for carbon emission accounting in industrial parks as described above.

[0018] Thirdly, the present invention provides an energy consumption data verification and completion system for carbon emission accounting in industrial parks, used to implement the energy consumption data verification and completion method for carbon emission accounting in industrial parks as described above, including: The multi-energy topology modeling and constraint module is used to build physical topology models for different energy sources and manage their unique physical conservation constraint rule library. The data access and preprocessing module is used to access multi-source and multi-energy data, align and standardize the units, and bind them to the corresponding topology nodes. The anomaly detection and root cause classification module is used to perform multi-dimensional anomaly detection and combine energy characteristics to perform root cause classification. The intelligent completion and confidence calculation module is used to dynamically select completion strategies for repair based on energy type and root cause, and to calculate confidence scores. The multi-energy fusion data output module is used to generate and output a carbon emission accounting ready dataset with confidence scores.

[0019] The present invention establishes physical topology models of energy consumption data covering various energy types based on the energy supply network structure of the park, and defines physical conservation constraints for different energy types; it performs statistical anomaly detection, topological anomaly detection, and pattern anomaly detection on energy consumption time-series data; it performs root cause classification on the abnormal data to obtain root cause labels; based on the root cause labels and the energy type to which the abnormal data belongs, it selects the corresponding intelligent completion strategy to repair the abnormal data, calculates a confidence score for each repaired data point, and includes data points with confidence scores higher than a preset score threshold into the carbon emission accounting dataset. Compared with the prior art, the present invention has the following beneficial effects: (1) Provide systemic solutions that support multiple energy sources and cross-level systems. By defining unique physical constraints for different energy sources and operating under a unified topology model and governance process, this solution addresses the pain points of scattered data sources, inconsistent standards, and difficulty in unified verification in multi-energy carbon emission accounting within industrial parks. By constructing a physical topology model of "equipment-branch road-building-park" and applying specific physical conservation constraint rules for each energy type for verification, this solution can identify and locate systemic, cross-level problems that traditional methods cannot address.

[0020] (2) Significantly improve the quality and credibility of carbon emission accounting data By integrating a triple mechanism of "statistical anomaly detection, physical topology constraint verification, and temporal pattern learning," this solution can identify hidden anomalies that traditional threshold methods or single algorithms cannot detect, increasing the anomaly detection rate to over 99.2%. Simultaneously, leveraging an intelligent data completion strategy based on root cause diagnosis, even in the event of partial meter failures or communication interruptions, it can still accurately repair data through methods such as topology backpropagation and spatiotemporal prediction, achieving a data integrity rate exceeding 99.5%. Third-party verification shows that applying this solution can reduce the overall error of annual carbon emission reports from ±8% to within ±3%, significantly improving the credibility of the reports and reducing compliance and audit risks caused by inaccurate data.

[0021] Furthermore, the dataset output by the scheme not only includes data values, but also data sources, processing methods, and confidence level labels. This complete data phylogenetic record fully complies with the stringent requirements of carbon accounting standards for data traceability, transparency, and uncertainty assessment.

[0022] (3) It possesses high intelligence and adaptability while reducing reliance on human intervention. Unlike existing technologies that select repair strategies solely based on the duration of missing data, this solution utilizes an automatic anomaly root cause classification model to diagnose the root cause of data problems and dynamically select the optimal completion strategy accordingly. This results in more accurate and reasonable data repair, avoiding systematic errors caused by a single method. Through automated anomaly detection, root cause analysis, and intelligent completion, the system can handle the vast majority of common data quality issues, significantly reducing the workload of manual data cleaning and entry. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram illustrating the steps of the energy consumption data verification and completion method for carbon emission accounting in industrial parks provided in an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the structure of an energy consumption data verification and completion system for carbon emission accounting in industrial parks, provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. It should be understood that the embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of the invention.

[0027] like Figure 1 As shown, this embodiment of the invention provides a method for verifying and completing energy consumption data for carbon emission accounting in industrial parks, including the following steps: S1: Based on the energy supply network structure of the park, establish physical topology models for energy consumption data covering electricity, water, gas, and heat energy types. The topology models include park-level, building-level, branch-level, and equipment-level nodes. For different energy types, define physical conservation constraint rules specific to the nodes in the physical topology models.

[0028] In this embodiment of the invention, a corresponding physical topology model is established based on the energy supply network structure of various energy sources within the park. For example, for electrical energy, the following four-layer physical topology is established: L1: Park's main incoming electricity meter; L2: Main meters for each building; L3: Distribution cabinet outgoing circuits and air conditioning main pipes; L4: Chiller units, charging piles, and lighting circuits. For thermal energy, the following three-layer physical topology is established: L1: Park's total heat meter for the cooling station; L2: Heat meters for each building's heat exchange station; L3: Flow meters and dual temperature sensors in the terminal VAV boxes. For water energy, the following topology is established: L1: Park's main municipal water inlet meter; L2: Building sub-meters; L3: Floor branch circuit meters, and pipe material attributes are associated to configure a leakage model.

[0029] In this embodiment of the invention, the physical conservation constraint rules defined for different energy types in the physical topology model include: for physical topology models of electricity, gas, and heat energy, the physical conservation constraint rules include: additive constraints, monotonicity constraints, and dimensional consistency constraints; for physical topology models of water energy, the physical conservation constraint rules include: additive constraints and monotonicity constraints.

[0030] Specifically, for electrical energy, the summation constraint means that the active power reading of the parent node should equal the sum of the active power readings of all its child nodes multiplied by (1 + preset line loss rate), where the loss rate can range from 0.5% to 2%. The summation constraint is a core means of detecting missing sub-meters, wiring errors, electricity theft, or continuous drift of a particular sub-meter. When the deviation between the total and the sub-sum continuously exceeds the reasonable loss range, a topology anomaly is triggered. The dimensional consistency constraint means that the voltage, current, and power of the same branch should satisfy P ≈ U × I × cosφ, where U is the line voltage, I is the line current, and cosφ is the power factor. The dimensional consistency constraint can effectively identify deep-seated faults that cannot be detected by power values ​​alone, such as voltage or current transformer faults, metering chip errors, or misaligned data upload channels. The monotonicity constraint means that the cumulative energy reading sequence is non-decreasing. The monotonicity constraint can quickly identify anomalies such as meter reset, communication message parsing errors causing numerical jumps, or meter storage failures.

[0031] For water energy, the additive constraint means that the cumulative flow reading of the parent node should be greater than or equal to the sum of the cumulative flow readings of all its child nodes plus a preset leakage rate, which is dynamically set according to the pipe type, pipe diameter, and service life. When the reading of the main inlet meter continuously exceeds the sum of the readings of all water-using sub-meters, and the deviation exceeds the normal leakage rate, the system can determine a suspected leak. The monotonicity constraint means that the cumulative water volume reading (m³) of the smart water meter should also be non-decreasing. The monotonicity constraint can identify water meter malfunctions or data transmission errors.

[0032] For gas energy, the additive constraint means that the standard cubic meter (Nm³) reading of the parent node should be equal to the sum of the readings of all its child nodes multiplied by (1 + a preset trace leakage rate), where the leakage rate can range from, for example, 0.2% to 0.8%. The additive constraint is used to diagnose gas pipeline leaks and metering system deviations. The monotonicity constraint means that the cumulative gas volume reading sequence is non-decreasing. The dimensional consistency constraint means that the standard volume value after temperature and pressure compensation is continuous in the time series without significant jumps, and the rate of change between adjacent time periods does not exceed, for example, 10% / h.

[0033] For thermal energy, the additivity constraint means that the cumulative heat reading of the parent node should be approximately equal to the sum of the heat readings of all its child nodes, allowing for heat exchange efficiency losses, with an efficiency factor ranging from, for example, 92% to 98%. The monotonicity constraint means that the cumulative heat reading sequence is non-decreasing. The dimensional consistency constraint means that... , For example, the value can be 5%Q. When heating, T_supply > T_return, and when cooling, T_supply < T_return. Where Q is the heat, m is the mass, c is the specific heat capacity, T_supply is the supply water temperature, and T_return is the return water temperature.

[0034] Physical topology models are established for pipeline structures of various energy sources, and specific physical constraints are formulated for the characteristics of various energy sources. This enables the system to perform in-depth verification and rationality analysis of data from the perspective of physical laws, and to discover deep-seated systemic data problems that cannot be identified by traditional threshold methods or statistical methods, thus significantly improving the scientific nature and effectiveness of data quality governance.

[0035] S2: Access the energy consumption time-series data of each node in the physical topology model, and perform data alignment and unit standardization.

[0036] In this embodiment of the invention, energy consumption time-series data is received from smart meters, data collectors, etc., and data alignment and unit standardization are performed. The processed data is then bound to the corresponding nodes in the physical topology model. The unit standardization includes: converting the gas flow rate under operating conditions to a standard state, such as the volumetric flow rate at 0°C and 101.325 kPa; and converting the heat data to a standard heat unit, such as gigajoules.

[0037] S3: Perform anomaly detection on the energy consumption time series data, including statistical anomaly detection based on historical data of the same period, topological anomaly detection based on the physical conservation constraint rules, and pattern anomaly detection based on time series pattern learning.

[0038] In this embodiment of the invention, the statistical anomaly detection specifically involves: calculating the Z-score based on historical data from the same period to identify abrupt change points. The topological anomaly detection refers to checking whether specified physical conservation constraints are violated. The pattern anomaly detection specifically involves: using a long short-term memory autoencoder model to perform unsupervised learning on historical normal energy consumption data to obtain a normal energy consumption pattern model; inputting the time-series data to be detected into the model, calculating its reconstruction error; if the reconstruction error exceeds a preset error threshold, it is determined to be an anomaly.

[0039] Long Short-Term Memory (LSTM) autoencoder models can autonomously learn the complex and dynamic normal energy consumption patterns of various energy-consuming devices, branches, or buildings under different seasons, workdays / holidays, and weather conditions. Compared to detection methods based on fixed thresholds or simple statistical rules, LSTM autoencoder models are more adaptable to unsteady and nonlinear load changes, accurately identifying hidden faults where numerical values ​​do not significantly exceed limits but curve shapes are abnormal. This greatly supplements the blind spots of statistical and topology detection, improving the comprehensiveness and intelligence of anomaly detection. By integrating statistical anomaly detection, topology anomaly detection, and pattern anomaly detection, the data anomaly detection rate is significantly improved.

[0040] S4: Based on the results of the anomaly detection, combined with the characteristics of the anomaly data, the energy type to which it belongs, and the physical conservation constraint rules, the anomaly data is classified into root causes to obtain root cause labels.

[0041] In this embodiment of the invention, the input features for root cause classification include missing patterns, deviation direction, duration of anomalies, location within the physical topology model, and energy type identifier; the output root cause labels include communication interruption, meter drift, meter jamming, and topology configuration error. For water and gas energy, pipe leaks are also included; for thermal energy, temperature sensor drift and flow meter failure are also included.

[0042] Root cause classification can be implemented using a trained machine learning classification model. Its core is to identify the specific fault information corresponding to the combination of input features. The meaning and function of the input features are as follows: Missing pattern indicates whether the data is completely missing, intermittently missing, or a constant value, which is key to distinguishing between communication interruptions and equipment malfunctions. Deviation direction indicates whether the abnormal data is consistently high, consistently low, or fluctuates irregularly, which helps distinguish between meter drift and temporary interference. Abnormal duration indicates whether the abnormality is instantaneous, lasts for hours, or is long-term; short-term problems may stem from temporary interference, while long-term problems point to equipment failure or configuration errors. The location in the physical topology model indicates whether the abnormality affects a single device, a branch, or the entire system, which can determine whether the problem is local or systemic. The energy type identifier directly limits the possible set of root causes; for example, only water and gas systems may have pipeline leaks. The following typical mapping relationships are provided in this embodiment of the invention: (1) Communication interruption Characteristic patterns: Missing pattern = "Completely missing data"; Abnormal duration = "Matches with the communication polling cycle"; Topology location = "May affect multiple devices under the same communication bus or gateway". Judgment logic: If the data flow completely stops and its impact range matches the network topology, the primary suspicion is a communication link problem.

[0043] (2) Meter drift Characteristic Pattern: Missing Pattern = "Data exists but is abnormal"; Deviation Direction = "Continuously unidirectionally higher or lower" (e.g., readings slowly increase or decrease); Abnormal Duration = "Long-term, gradual"; Topology Location = "Affects only a single meter node"; Topology Verification = "May cause a gradual deviation in the additive property between the data of this node and the data of its superior / subordinate nodes". Judgment Logic: If data exists but continuously and slowly deviates from the normal value or the expected value calculated based on the topology, and it is a problem with a single device, it is very likely that the meter's measurement accuracy has drifted.

[0044] (3) Meter jamming Characteristic pattern: Missing pattern = "Data remains constant" (non-zero or a fixed value); Deviation direction = "No change"; Abnormal duration = "Long-term"; Topological location = "Affects only a single meter node"; Simultaneously violates monotonicity constraints (for cumulative meters, readings should increase). Judgment logic: A reading that remains locked at a fixed value for a long time is a typical manifestation of mechanical or electronic component malfunction in the instrument.

[0045] (4) Topology configuration error Characteristic Patterns: Missing Pattern = "Data exists but logic is abnormal"; Deviation Direction = "A large deviation in a fixed direction exists between the sum of the parent node reading and the child node readings"; Abnormal Duration = "Persists for a long time after system commissioning or a certain point in time"; Topology Location = "Affects a group of nodes with parent-child relationships"; Topology Verification = "Severe and persistent violation of additivity constraints". Judgment Logic: This is the root cause directly detected through physical conservation constraints. When the system finds a stable and huge difference between the reading of a parent node and the sum of the readings of all its child nodes that cannot be explained by normal losses, it can be determined that there is an error in the physical wiring or system logical binding (i.e., "topology configuration").

[0046] (5) Pipeline leakage Characteristic patterns: Missing pattern = "Data exists but is abnormal"; Deviation direction = "Inlet main meter reading is consistently greater than the sum of all outlet sub-meter readings (positive deviation)"; Abnormal duration = "May persist"; Topological location = "Affects the inlet and outlet nodes of a certain section of the pipeline"; Topological verification = "Violation of additivity constraint". For water supply data, a pipeline leak is identified when there is a continuous low flow rate and stable flow rate during non-water use periods at night; for gas data, a pipeline leak is identified when there is a continuous low flow rate during non-cooking / heating periods.

[0047] (6) Temperature sensor drift Temperature sensor drift is a specific root cause of thermal energy issues. Characteristic patterns: Missing pattern = "Abnormal temperature readings (e.g., constant, abrupt changes, exceeding range)"; Deviation direction = "Abnormal temperature values ​​lead to abnormal calculated heat"; Topological location = "Affecting a single temperature measurement point"; Energy specificity verification = "Causing the constraint of the heat calculation formula to fail (e.g., supply and return water temperature difference is 0 or negative, but flow rate is normal)". Judgment logic: When the flow rate data required for heat calculation is normal, but the temperature reading is obviously unreasonable (e.g., supply and return water temperatures are the same), resulting in zero or abnormal calculated heat, temperature sensor drift can be inferred.

[0048] (7) Flowmeter malfunction The heat metering point must be equipped with a heat meter, a flow meter, and dual temperature sensors. The heat meter outputs the cumulative / instantaneous heat, and the flow meter outputs the mass flow rate. The dual temperature sensors measure the supply water temperature T_supply and the return water temperature T_return, respectively. A flow meter malfunction is only considered when the temperature difference is normal, the heat output is valid, but "flow rate × temperature difference" cannot explain the measured heat output.

[0049] By comprehensively considering the characteristics of missing patterns, deviation directions, durations, topological locations, and energy types, the system attributes anomalies to specific, actionable fault types such as communication interruptions, meter drift, and pipeline leaks. This provides a precise basis for subsequent targeted selection of the optimal repair strategy, avoiding secondary errors caused by a one-size-fits-all approach to repair.

[0050] S5: Based on the root cause label and the energy type to which the abnormal data belongs, select the corresponding intelligent completion strategy from the preset completion strategy set to repair the abnormal data. During the repair process, calculate the confidence score for each repaired data point. Only data points with confidence scores higher than the preset score threshold are included in the final dataset for carbon emission accounting.

[0051] In this embodiment of the invention, step S5, selecting the corresponding intelligent completion strategy, includes: if the root cause label is communication interruption, a spatiotemporal collaborative prediction strategy is adopted, using complete data, meteorological data, and calendar features of other nodes at the same or adjacent levels to predict missing data through a spatiotemporal transformer model. If the root cause label is meter drift, a physical topology back-calculation strategy is adopted, using the sum of energy consumption data of its child nodes, combined with a preset physical loss rate, to back-calculate the energy consumption data of that node. If the root cause label is meter jamming, a time-series pattern repair strategy is adopted, using the normal energy consumption pattern model to generate a replacement data sequence that conforms to historical change patterns. If the root cause label is pipeline leakage, for water energy, the original reading is not completed, a leakage correction amount is generated and superimposed on the parent node's master meter, and a maintenance work order is triggered; for gas energy, automatic completion is not performed, only abnormal events are recorded and high-priority safety alarms are pushed. For thermal energy, if the root cause label is temperature sensor drift and the flow data is normal, the historical average supply and return water temperature difference at that metering point is used to reconstruct the heat based on the measured flow rate; if the root cause label is flow meter failure, the flow rate is back-calculated using a correlation model between heat and pump frequency or valve opening, and then the heat is calculated.

[0052] In this embodiment of the invention, strategy selection based on root cause dynamics is more scientific and targeted than the traditional method of selecting based solely on the duration of missing data, thereby producing high-quality completion data that is closer to the real situation.

[0053] In some embodiments of the present invention, the intelligent completion strategy further includes a cross-energy collaborative completion strategy; when the energy consumption data of the first energy type is determined to be missing or abnormal, and there is energy consumption data of the second energy type that is strongly correlated with the first energy consuming device in terms of physical or operational logic, the following steps are performed: based on historical normal operation data, a correlation model is trained to obtain the correlation model between the consumption of the first energy type, the consumption of the second energy type, and external environmental parameters; during the period when the first energy data is missing or abnormal, complete consumption data of the second energy type and the corresponding external environmental parameters are obtained; the consumption data of the second energy type and the external environmental parameters are input into the correlation model to predict the completed data sequence of the first energy type.

[0054] The cross-energy collaborative completion strategy makes full use of the physical coupling and operational correlation between different energy flows in the park's integrated energy system. When single energy information is insufficient, this method provides a more reliable and physically based completion approach, improves the overall quality and consistency of multi-energy datasets, and enhances the robustness of the system under complex operating conditions.

[0055] In this embodiment of the invention, the calculation of the confidence score in step S5 integrates the prediction variance of the completion strategy itself, the matching degree between the completion result and the historical pattern, whether the completion result satisfies the physical conservation constraint rule in the physical topology model, and the inherent accuracy level of the metering instrument for this energy type.

[0056] The confidence score provides a quantifiable and interpretable quality label for each completed data point. By comprehensively considering the uncertainty of the prediction itself, its consistency with historical patterns, its adherence to physical laws, and the inherent accuracy of the instrument, the obtained confidence score can objectively and comprehensively reflect the credibility of the completed data. This enables downstream carbon emission accounting applications to clearly distinguish data with different levels of reliability and provides a scientific basis for setting data adoption thresholds. In some embodiments of this invention, only data points with a confidence score of not less than 0.7 are included in the final dataset for carbon emission accounting; data with a confidence score < 0.7 are marked as "requiring manual review" and do not participate in formal carbon emission accounting.

[0057] In this embodiment of the invention, the energy consumption data verification and completion method further includes a multi-energy data aggregation and cross-verification step: integrating all energy data after repair, converting the consumption of different energy sources into a unified carbon dioxide equivalent based on the carbon emission factors of various energy sources, and using the macro trend of total energy consumption and total carbon emissions in the park to perform cross-verification of data rationality across energy sources.

[0058] By converting treated data from multiple energy sources, including electricity, water, gas, and heat, into carbon dioxide equivalents based on carbon emission factors, a unified dataset for carbon emission accounting is directly generated, greatly simplifying the accounting process. By analyzing the macro-trend correlation between total energy consumption and total carbon emissions in the park, the final data rationality can be verified. For example, if total energy consumption is stable but the converted total carbon emissions show abnormal fluctuations, an alarm can be triggered to check for errors in the correction of certain energy data or inappropriate application of carbon emission factors.

[0059] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the energy consumption data verification and completion method for carbon emission accounting in industrial parks as described above.

[0060] like Figure 2 As shown in the figure, this embodiment of the invention also provides an energy consumption data verification and completion system for carbon emission accounting in industrial parks, including a multi-energy topology modeling and constraint module, a data access and preprocessing module, an anomaly detection and root cause classification module, and an intelligent completion and confidence calculation module. The multi-energy topology modeling and constraint module is used to establish physical topology models for different energy sources and manage their unique physical conservation constraint rule bases. The data access and preprocessing module is used to access multi-source, multi-energy data, perform alignment, unit standardization, and bind it to corresponding topology nodes. The anomaly detection and root cause classification module is used to perform multi-dimensional anomaly detection and perform root cause classification based on energy characteristics. The intelligent completion and confidence calculation module is used to dynamically select completion strategies for repair based on energy type and root cause, and calculate a confidence score. The energy consumption data verification and completion system for carbon emission accounting in industrial parks also includes a multi-energy fusion data output module, used to generate and output a carbon emission accounting-ready dataset with a confidence score.

[0061] For the physical topology models of electricity, gas, and heat energy, the physical conservation constraints include: additive constraints, monotonicity constraints, and dimensional consistency constraints; for the physical topology models of water energy, the physical conservation constraints include: additive constraints and monotonicity constraints.

[0062] Multi-dimensional anomaly detection includes statistical anomaly detection, topological anomaly detection, and pattern anomaly detection. Statistical anomaly detection specifically involves calculating the Z-score based on historical data from the same period to identify abrupt changes. Topological anomaly detection refers to checking for violations of specified physical conservation constraints. Pattern anomaly detection specifically involves using a Long Short-Term Memory (LSTM) autoencoder model to perform unsupervised learning on historical normal energy consumption data to obtain a normal energy consumption pattern model; inputting the time-series data to be detected into the model, calculating its reconstruction error, and determining an anomaly if the reconstruction error exceeds a preset error threshold.

[0063] The input features of the root cause classification module include missing patterns, deviation direction, duration of anomalies, location within the physical topology model, and energy type identifier; the output root cause labels include communication interruption, meter drift, meter jamming, and topology configuration errors. For water and gas energy, it also includes pipeline leaks; for thermal energy, it also includes temperature sensor drift and flow meter failure.

[0064] The dynamic selection and completion strategy includes: if the root cause label is communication interruption, a spatiotemporal collaborative prediction strategy is adopted, using complete data, meteorological data, and calendar features from other nodes at the same or adjacent levels to predict missing data through a spatiotemporal transformer model. If the root cause label is meter drift, a physical topology back-calculation strategy is adopted, using the sum of energy consumption data of its child nodes and a preset physical loss rate to back-calculate the energy consumption data of that node. If the root cause label is meter jamming, a time-series pattern repair strategy is adopted, using the normal energy consumption pattern model to generate a replacement data sequence that conforms to historical change patterns. If the root cause label is pipeline leakage, for water energy, the original reading is not completed, a leakage correction amount is generated and superimposed on the parent node's master meter, and a maintenance work order is triggered; for gas energy, automatic completion is not performed, only abnormal events are recorded and high-priority safety alarms are pushed. For thermal energy, if the root cause label is temperature sensor drift and the flow data is normal, the historical average supply and return water temperature difference at that metering point is used to reconstruct the heat based on the measured flow rate; if the root cause label is flow meter failure, the flow rate is back-calculated using a correlation model between heat and pump frequency or valve opening, and then the heat is calculated.

[0065] Example 1 This embodiment takes the power data management of an office building (Building A) in a smart park on a certain day as an example to explain in detail the implementation process of the energy consumption data verification and completion method for carbon emission accounting in the park according to the present invention. The overall process includes the following steps: S101: Physical Topology Modeling. Based on the power distribution system diagram of the park, establish a four-level physical topology model of its power consumption and define physical conservation constraints.

[0066] Park Level (L1): Data of the main incoming electricity meter in the park.

[0067] Building Level (L2): Data of the main incoming power meter in the power distribution room on the first basement floor of Building A.

[0068] Branch Level (L3): Includes the electricity meter data of low-voltage outgoing circuits such as lighting branch circuits, air conditioning branch circuits, and socket branch circuits on each floor of Building A.

[0069] Equipment level (L4): Includes smart meter or sensor data (if any) from terminal devices such as “3F East Side Conference Room Lighting” and “3F Fresh Air Unit”.

[0070] In this model, the core additive constraint is defined as: L2_A Building Summary Table ≈ ∑ (Lighting Branch Circuit + Air Conditioning Branch Circuit + Socket Branch Circuit) (1 + γ), where γ is the preset average loss rate of the line segment (e.g., 0.5%).

[0071] S102: Data Access and Binding. Obtain the electricity consumption (kWh) data of all the above nodes at 15-minute intervals for the entire day of February 1, 2026, from the park's energy management system. Bind the data to the nodes in the topology model one by one according to the meter address.

[0072] S103: Multi-dimensional anomaly detection. The system automatically scanned the data from the previous day (February 1st) in the early morning of February 2nd. Statistical detection, topology detection, and pattern detection were performed, details of which are as follows: Statistical analysis revealed that 16 data points for the 3F lighting branch between 10:00 and 14:00 were all null values ​​(NULL), and the Z-score was infinite, which was marked as a statistical anomaly.

[0073] Topology analysis: Calculating the sum of meter readings for each L3 branch during the period from 10:00 to 14:00 revealed a relatively stable gap of approximately 15% between this sum and the main meter reading for Building A (L2_A). Historical data for the same period shows that lighting branch loads account for approximately 12-18% of the total load in Building A during this time. Therefore, the system determined a topology anomaly: the main meter readings cannot be explained by the known air conditioning and socket branches, suggesting that the lighting branches should have loads but data is missing.

[0074] Pattern detection: The sequence of the entire day on February 1st for the 3F lighting branch (with missing segments filled with 0s) was input into a pre-trained long short-term memory autoencoder model. The model's reconstruction error of the daytime load pattern of this branch on weekdays increased sharply in the missing segments, far exceeding the threshold, and was marked as a pattern anomaly.

[0075] S104: Automatic Classification of Anomaly Root Causes. The root cause classifier receives the following anomaly characteristics: complete absence for 4 consecutive hours, topology check indicating expected data, and no transient fluctuations. Combined with a query of the device communication logs, it is found that the branch meter had a "communication timeout" alarm during the corresponding period. After comprehensive judgment, the classifier outputs the root cause label: COMM_LOSS (communication interruption), with a confidence level of 98%.

[0076] S105: Intelligent completion and confidence calculation. Based on the root cause label COMM_LOSS, a spatiotemporal co-prediction strategy is automatically selected. A pre-trained spatiotemporal transformer model is invoked. Input features include: (1) Spatiotemporal characteristics: The complete load sequence (spatial correlation) of “3F air conditioning branch” and “3F socket branch” within the missing time period.

[0077] (2) Contextual characteristics: Weather data for the day: sunny, outdoor temperature 12℃; Calendar characteristics: Tuesday, weekday. Other branches in the same building: air conditioning load is stable, socket load is periodic (meeting mode).

[0078] (3) Historical pattern: The average load curve of this lighting branch during Tuesdays from 10:00 to 14:00 over the past 4 weeks. The average lighting load during the same period over the past 4 weeks was 8.2 kW, with a standard deviation of 0.5 kW.

[0079] The model outputs a sequence of load forecasts for the 16 missing time points (10:00, 10:15, ..., 13:45, 14:00), for example: [8.1, 8.3, 8.0, ..., 7.9] kW.

[0080] Then, calculate the overall confidence score Score_c for each completion point. Prediction certainty score (S1): The ST-Transformer model provides variance σ² for this sequence prediction. A smaller variance indicates higher prediction certainty; S1 = 0.8. Historical pattern matching score (S2): Calculate the cosine similarity between the completed sequence and historical concurrent pattern sequences; the similarity is 0.94, so S2 = 0.94. Topological consistency score (S3): Substitute the completed "3F lighting branch" sequence into the additive constraint formula for verification. The calculation shows that the error between the completed lighting load + air conditioning load + socket load and the A building's total meter is <1%. The constraint is satisfied, so S3 = 1.0. Overall confidence score Score_c = w1 S1 + w2 S2 + w3 S3, assuming weights w1=0.3, w2=0.3, w3=0.4, the average overall confidence level of the 16 completion points is calculated to be 0.92.

[0081] S106: Generate a carbon emission-ready dataset. The system integrates the original correct data with these 16 completed data points labeled with the completion method, root cause, and confidence level (0.92) to form a complete electricity consumption dataset for Building A as of February 1, 2026.

[0082] The park's carbon emissions calculation engine is configured to only include data with a confidence level ≥ 0.7 in the daily carbon emissions calculation. Therefore, this supplementary data was automatically adopted.

[0083] Effect evaluation: Calculations show that if the missing data for these 4 hours is ignored, the daily electricity consumption calculation error for Building A would be >15%. After using the solution of this invention to complete the calculation, the daily total electricity consumption calculation error is reduced to within 2%, effectively supporting the accurate calculation of carbon emissions.

[0084] Example 2 This embodiment provides a detailed explanation of the data characteristics, physical constraints, and specific processing strategies for four typical energy sources: electricity, water, gas, and heat, to demonstrate the universality and refinement of this solution.

[0085] Park Overview and Measurement System The Luyuan Campus comprises six office buildings (A–F), one data center (DC), two staff canteens, and one regional energy station. The deployed smart metering devices are shown in Table 1, and all devices support data uploads to the campus energy management platform every 15 minutes.

[0086] Table 1. Measurement System of Example 2

[0087] S201: Physical Topology Modeling. Based on the energy supply network diagram of the park, establish physical topology models for various energy types and define physical conservation constraints.

[0088] The system parses the park's BIM model and automatically generates four independent topology trees: Example of power topology: Campus master table (L1) → Data center master table (L2) → IT room PDU branch (L3) → Server rack #01 (L4).

[0089] Water supply topology example: Municipal main meter (L1) → Building B sub-meter (L2) → 3F men's restroom flushing branch (L3) → The system automatically associates the pipe material of this branch with "PE pipe" and sets the leakage rate to 0.1% / day.

[0090] Gas topology example: Gas main meter (L1) → Canteen No. 1 stove (L2).

[0091] Example of thermal topology: Total heat meter of energy station (L1) → Heat meter of heat exchange station in building C (L2) → VAV#C305 (L3: Flow meter + T_supply + T_return).

[0092] The system loads exclusive constraint rules for each type of energy, as shown in Table 2. Monotonicity constraints are omitted for each energy type.

[0093] Table 2 Summary of Physical Conservation Constraints for Each Energy Type

[0094] S203 and S204 Anomaly Detection and Root Cause Classification. Example scenarios are as follows: Scenario 1: Data Center Power Meter Drift Phenomenon: The daily electricity consumption of L2 "DC master meter" has been 3.2% higher than normal for 5 consecutive days, while L3 / L4 are normal.

[0095] Detection: Statistical analysis: Z-score = 3.1 → Abnormal; Topology check: L2 > Σ(L3) × 1.02 → Violation of additivity; Pattern detection: Error exceeds the limit.

[0096] Root cause: Meter drift.

[0097] Scenario 2: Minor water leak in Building B at night Phenomenon: The male bathroom on the 3rd floor of Building B in L3 had a continuous flow rate of 0.08 m³ / h from 23:00 to 5:00 for 7 consecutive days.

[0098] Detection: Statistical analysis: Deviation from historical nighttime baseline (≈0) >99%; Topology detection: L2 reading > Σ(L3) + leakage amount → abnormal; Root cause analysis: Pipe leak.

[0099] Scenario 3: Minor gas leak in the cafeteria Phenomenon: The flow rate of L2 "Canteen 2#" was 0.15 Nm³ / h from 2:00 to 5:00 in the morning.

[0100] Detection: Statistical analysis: Abnormalities were observed during non-cooking periods; Pattern detection: No similar patterns found in the past; Root cause: Pipe leak.

[0101] Scenario 4: Drift of the thermal energy temperature sensor in Building C Phenomenon: The temperature of L3 "VAV#C305" T_ return is 11.8℃ (should be ≈12℃), resulting in ΔT=4.8℃ (normal is 5℃).

[0102] Detection: Dimensional consistency: Q_calc - Q_meter = 6.2% > 5% → Violation; Topology detection: L2 heat level is low; Root cause: Temperature sensor drift.

[0103] S205, Energy-Specific Intelligent Completion. A summary of completion strategies for various scenarios is shown in Table 3.

[0104] Table 3 Summary of Completion Strategies in Example 2

[0105] The confidence levels and carbon emission accounting results for each scenario are shown in Table 4.

[0106] Table 4. Confidence level and carbon emission accounting results of Example 2

[0107] Final output: All data with a confidence level ≥ 0.7 are tagged "Carbon-Ready" and pushed to the carbon management platform.

[0108] After implementing the solution of this invention, the detection rate of abnormal energy consumption data in the park has increased to 99.2%, accurately identifying hidden anomalies missed by traditional methods; the data integrity rate is >99.5%, and even if some meters malfunction, they can still be completed through topology and mode; carbon emission accounting error has been reduced by more than 60%. Third-party verification shows that the annual carbon emission report deviation has decreased from ±8% to within ±3%; audit traceability is supported: each data point includes its source, method, and confidence level, meeting the requirements of standards such as ISO 14064; manual intervention has been reduced by 70%: only 5% of low-confidence data requires manual processing.

[0109] Example 3 This embodiment provides an example of a multi-energy collaborative processing scenario, using a heating system in a park that uses a "gas boiler + electric water pump" as an example to illustrate multi-energy collaborative processing.

[0110] Topology modeling: In the thermal topology, the boiler is a heat-generating device (equipment level); in the gas topology, it is a gas-consuming device (equipment level); in the power topology, the water pump is an electrical-consuming device (equipment level). The system establishes a cross-energy "device-energy" mapping relationship through device ID association.

[0111] Abnormal scenario: Gas meter communication interruption resulted in a 2-hour loss of gas consumption data. During the period of missing gas data, water pump power and outdoor temperature data were complete.

[0112] Strategy Selection: Since the boiler operates in conjunction with the electric water pump, the system selects a prediction strategy based on correlated power data. Historical data is analyzed to establish a regression model of boiler gas consumption (G) with water pump power (P_pump) and outdoor temperature (T_out): G = f(P_pump, T_out). The water pump power and outdoor temperature data for the missing gas data periods are input into the trained regression model to predict the missing 2-hour gas consumption sequence.

[0113] Confidence Calculation: The overall confidence level of the gas filler value is obtained by combining the R² score of the regression model and the confidence level of the input electrical power data.

[0114] Example 4 This embodiment further illustrates the handling strategy for topology configuration errors.

[0115] Scenario: A new charging station is installed, but its data is incorrectly associated with the air conditioning branch instead of being created as an independent node in the topology model. Long-term verification reveals that the load pattern of the air conditioning branch is abnormal, and its value is far higher than the historical average for the same period.

[0116] Root cause classification: Long-term, stable violations of expectations based on historical patterns are classified as topology configuration errors.

[0117] Handling strategy: Do not perform autocomplete. The system generates a "Suspected Topology Error" work order, pushes it to the operation and maintenance platform, marks the data confidence of the node as 0, and prompts the carbon emission accounting engine to use it with caution or wait for manual confirmation.

[0118] The above are merely specific embodiments of the present invention and should not be construed as limiting the scope of the present invention. Equivalent variations made by those skilled in the art based on this invention, as well as changes well-known to those skilled in the art, should still fall within the scope of the present invention.

Claims

1. A method for verifying and completing energy consumption data for carbon emission accounting in industrial parks, characterized in that, include: S1: Based on the energy supply network structure of the park, establish physical topology models of energy consumption data covering electricity, water, gas and heat energy types respectively. The topology model includes park-level, building-level, branch-level and equipment-level nodes; and define physical conservation constraint rules specific to the nodes in the physical topology model for different energy types. S2: Access the energy consumption time-series data of each node in the physical topology model, and perform data alignment and unit standardization; S3: Perform anomaly detection on the energy consumption time series data, including statistical anomaly detection based on historical data of the same period, topological anomaly detection based on the physical conservation constraint rules, and pattern anomaly detection based on time series pattern learning; S4: Based on the results of the anomaly detection, combined with the characteristics of the anomaly data, the energy type, and the physical conservation constraint rules, the anomaly data is classified into root causes to obtain root cause labels; S5: Based on the root cause label and the energy type to which the abnormal data belongs, select the corresponding intelligent completion strategy from the preset completion strategy set to repair the abnormal data. During the repair process, calculate the confidence score for each repaired data point. Only data points with confidence scores higher than the preset score threshold are included in the final dataset for carbon emission accounting.

2. The energy consumption data verification and completion method for carbon emission accounting in industrial parks according to claim 1, characterized in that, In step S1, defining the physical conservation constraint rules specific to the nodes in the physical topology model for different energy types includes: For the physical topology model of electricity, gas, and heat energy, the physical conservation constraints include: additivity constraints, monotonicity constraints, and dimensional consistency constraints. For the physical topology model of water energy, the physical conservation constraints include: additive constraints and monotonic constraints.

3. The energy consumption data verification and completion method for carbon emission accounting in industrial parks according to claim 1, characterized in that, In step S3, the anomaly detection specifically involves: using a long short-term memory autoencoder model to perform unsupervised learning on historical normal energy consumption data to obtain a normal energy consumption pattern model; inputting the time series data to be detected into the model, calculating its reconstruction error, and determining it as an anomaly if the reconstruction error exceeds a preset error threshold.

4. The energy consumption data verification and completion method for carbon emission accounting in industrial parks according to claim 1, characterized in that, In step S4, the input features for root cause classification include missing patterns, deviation direction, duration of anomalies, location in the physical topology model, and energy type identifier; the output root cause labels include communication interruption, meter drift, meter jamming, and topology configuration error; for water and gas energy, it also includes pipeline leakage; for thermal energy, it also includes temperature sensor drift and flow meter failure.

5. The energy consumption data verification and completion method for carbon emission accounting in industrial parks according to claim 4, characterized in that, In step S5, selecting the corresponding intelligent completion strategy includes: If the root cause label is communication interruption, a spatiotemporal collaborative prediction strategy is adopted, which uses complete data, meteorological data and calendar features of other nodes at the same or adjacent levels to predict missing data through a spatiotemporal transformer model. If the root cause label is meter drift, a physical topology reverse calculation strategy is adopted, which uses the sum of the energy consumption data of its child nodes and a preset physical loss rate to reverse calculate the energy consumption data of the node. If the root cause label is meter stagnation, a time-series mode repair strategy is adopted, which uses the normal energy consumption mode model to generate an alternative data sequence that conforms to the historical change pattern. If the root cause label is a pipeline leak, for water energy, the original reading is not completed, a leakage correction amount is generated and added to the parent node's master table, and a maintenance work order is triggered; for gas energy, automatic completion is not performed, only abnormal events are recorded and high-priority safety alarms are pushed. If the root cause label indicates temperature sensor drift and the flow data is normal, then the heat is reconstructed by using the historical average supply and return water temperature difference at that metering point and combining it with the measured flow rate. If the root cause label is a flow meter malfunction, the flow rate is inferred by using a correlation model between heat and pump frequency or valve opening, and then the heat is calculated.

6. The energy consumption data verification and completion method for carbon emission accounting in industrial parks according to claim 1, characterized in that, The intelligent completion strategy also includes a cross-energy collaborative completion strategy; when energy consumption data of the first energy type is determined to be missing or abnormal, and there is energy consumption data of the second energy type that is strongly correlated with the first energy consuming device in terms of physical or operational logic, the following steps are executed: Based on historical normal operation data, a correlation model between the consumption of the first energy type, the consumption of the second energy type, and external environmental parameters is trained. During periods when the first energy data is missing or abnormal, obtain complete consumption data for the second energy type and corresponding external environmental parameters. The consumption data of the second energy type and external environmental parameters are input into the correlation model to predict the complete data sequence of the first energy type.

7. The energy consumption data verification and completion method for carbon emission accounting in industrial parks according to claim 1, characterized in that, In step S5, the calculation of the confidence score takes into account the prediction variance of the completion strategy itself, the matching degree between the completion result and the historical pattern, whether the completion result satisfies the physical conservation constraint rules in the physical topology model, and the inherent accuracy level of the metering instrument for this energy type.

8. The energy consumption data verification and completion method for carbon emission accounting in industrial parks according to claim 1, characterized in that, The energy consumption data verification and completion method also includes multi-energy data aggregation and cross-verification steps: integrating all energy data after restoration, converting the consumption of different energy sources into a unified carbon dioxide equivalent based on the carbon emission factors of various energy sources, and using the macro trend of total energy consumption and total carbon emissions in the park to perform cross-verification of data rationality across energy sources.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the energy consumption data verification and completion method for carbon emission accounting in industrial parks as described in any one of claims 1 to 8.

10. A system for verifying and completing energy consumption data for carbon emission accounting in industrial parks, used to implement the method for verifying and completing energy consumption data for carbon emission accounting in industrial parks as described in any one of claims 1-8, characterized in that, include: The multi-energy topology modeling and constraint module is used to build physical topology models for different energy sources and manage their unique physical conservation constraint rule library. The data access and preprocessing module is used to access multi-source and multi-energy data, align and standardize the units, and bind them to the corresponding topology nodes. The anomaly detection and root cause classification module is used to perform multi-dimensional anomaly detection and combine energy characteristics to perform root cause classification. The intelligent completion and confidence calculation module is used to dynamically select completion strategies for repair based on energy type and root cause, and to calculate confidence scores. The multi-energy fusion data output module is used to generate and output a carbon emission accounting ready dataset with confidence scores.

Images