AI energy efficiency and carbon footprint integrated management method and system for zero-carbon park

Abstract

The application discloses an AI energy efficiency and carbon footprint integrated management method and system for a zero-carbon park, and particularly relates to the technical field of zero-carbon park energy efficiency management and carbon achievement verification, which comprises obtaining park real-time data corresponding to a target action, the park real-time data comprising device states before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data, and performing synchronous mapping on the park real-time data in a digital twin according to a unified time scale and a unified object identifier, and outputting a state before the action; by synchronously constructing an entity state sequence containing the target action and a mirror state sequence without the target action in the digital twin, and performing difference stripping, attribution fragment extraction and result merging under the unified time scale, the unified object identifier and the same external disturbance and operating boundary conditions, an attributable emission reduction result formed by the target action itself is identified and confirmed.

Description

Technical Field

[0001] This invention relates to the field of energy efficiency management and carbon footprint verification technology for zero-carbon parks. More specifically, this invention relates to an AI-based integrated management method and system for energy efficiency and carbon footprint of zero-carbon parks. Background Technology

[0002] In the process of energy efficiency management and carbon achievement verification in zero-carbon parks, the relevant technologies have long focused on whether the energy-saving and emission-reduction effects can be truly confirmed after the implementation of equipment modification, control logic adjustment or operation and maintenance measures. In engineering, the focus is on collecting energy consumption metering data, operation status data and carbon emission calculation results before and after the modification, and combining them with meteorological conditions, load levels, historical operation records or baseline models to calculate the differences and give the corresponding conclusions on energy saving and emission reduction. Taking a comprehensive park that includes office buildings, commercial areas and production units as an example, after completing the adjustment of the cold station group control parameters, the switching of air conditioning terminal control or the change of energy storage dispatch mode, on the one hand, it is not possible to affect the normal energy use of tenants, the maintenance of indoor environment and continuous production operation; on the other hand, it is necessary to make a clear determination of the emission reduction effect brought about by the adjustment when changes in occupancy rate, shift adjustment, temporary work stoppage, external electricity price fluctuation, changes in the proportion of green electricity access and equipment performance degradation exist at the same time. In this application scenario, a common phenomenon occurs: although energy consumption and carbon emissions can be observed on paper, it is difficult to explain whether the decrease was directly caused by the renovation or adjustment, or by a combination of factors such as load contraction, temperature changes, operating period changes, or energy structure changes during subsequent review, distribution of benefits, or reporting of results. This leads to repeated discrepancies in the recognition of renovation benefits, service settlement, and confirmation of carbon results. The root cause is that the existing processing methods mainly rely on comparisons between different time periods or historical baseline estimations, which cannot provide corresponding reference results for when the target action was not implemented at the same time, under the same external disturbance, and under the same operating boundary. The technical problem this application aims to solve is: how to accurately identify and confirm the attributable emission reduction results generated by the target actions themselves during the process of equipment transformation, control strategy adjustment, or operation and maintenance measures in zero-carbon industrial parks, under conditions of complex external disturbances and continuously changing operating boundaries. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an AI-based integrated management method and system for energy efficiency and carbon footprint in zero-carbon parks. By simultaneously constructing an entity state sequence containing the target action and a mirror state sequence without the target action in a digital twin, and performing differential stripping, attribution fragment extraction, and result merging under unified time scale, unified object identification, and the same external disturbances and operating boundary conditions, the attributable emission reduction results formed by the target action itself can be identified and confirmed, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks, comprising: S1. Obtain the real-time data of the park corresponding to the target action. The real-time data of the park includes the equipment status before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data. Perform synchronous mapping on the real-time data of the park in the digital twin according to a unified time scale and a unified object identifier, and output the status before the action. S2. Using the state before the action as input, retain the state rewriting process corresponding to the target action in the digital twin to form an entity state sequence, and remove the state rewriting process corresponding to the target action while keeping the meteorological data, load data and boundary data unchanged to form a mirror state sequence. S3. Read the corresponding state items in the entity state sequence and mirror state sequence at the same time position, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and remove the same external influences based on the same meteorological data, load data and boundary data in the corresponding state items, and output the difference sequence. S4. Compare the energy flow difference direction and carbon flow difference direction of adjacent difference terms in the difference sequence. When both directions are consistent, perform continuous continuation of the corresponding difference terms. When either direction is reversed, cut off the continuation relationship and output the attribution segment.

[0005] In a preferred embodiment, it further includes: S5. Accumulate the energy flow difference and carbon flow difference in each attribution segment to obtain the energy efficiency change and carbon footprint change corresponding to each attribution segment. Then merge the energy efficiency change and carbon footprint change corresponding to each attribution segment in chronological order and output the attributable emission reduction results corresponding to the target action.

[0006] In a preferred embodiment, S1 includes: S1-1. Obtain the original time values ​​and object identifiers corresponding to the equipment status, energy flow data, carbon flow data, meteorological data, load data and boundary data before the action. Perform a base time scale conversion on each original time value and rearrange them according to a unified time scale. Perform name merging and hierarchical expansion on each object identifier. Output the time scale alignment result and the identifier alignment result. S1-2. Read the time scale alignment result and the identifier alignment result. Write the device status before the action to the device status bit according to the unified time scale and the unified object identifier. Write the energy flow data to the energy flow bit, write the carbon flow data to the carbon flow bit, write the meteorological data to the meteorological bit, write the load data to the load bit, write the boundary data to the boundary bit, and output the object mapping table. S1-3. For the same unified object identifier in the object mapping table, perform a before-and-after continuity check on the corresponding equipment status, energy flow data, carbon flow data, meteorological data, load data and boundary data under adjacent unified time scales. If the continuity is successful, retain the corresponding record. If the continuity is unsuccessful, delete the broken record and reconnect the before-and-after continuity records according to the unified object identifier, and output the status before the action.

[0007] In a preferred embodiment, S2 includes: S2-1. Read the equipment status, energy flow data, carbon flow data, meteorological data, load data and boundary data in the state before the action. In the digital twin, establish a succession relationship diagram according to the unified object identifier and unified time scale. Based on the target action corresponding to the target object, action period and action field, delineate the candidate rewrite edge set in the succession relationship diagram. At the same time, write the meteorological data, load data and boundary data into a fixed invariant item set. S2-2. For each candidate rewriting path in the candidate rewriting edge set, calculate the equipment state continuity cost, energy flow continuity cost, and carbon flow continuity cost between adjacent edges in the path. Under the condition that the fixed invariant itemset remains consistent item by item and the unified object identifier remains continuous, sort the candidates by total cost from smallest to largest and retain the candidate rewriting path that is ranked first and whose ranking results are consistent in two consecutive rounds as the target rewriting path.

[0008] In a preferred embodiment, S2 further includes: S2-3. Replace the original continuation edges at the corresponding positions in the state before the action with the target rewriting path, and perform time-scaled deduction on the associated states before and after the action along the target rewriting path under the condition that the fixed itemset remains unchanged. Stop the deduction when the rewriting position set and the rewritten state value obtained in two adjacent rounds of deduction are consistent, and output the entity state sequence. S2-4. Delete the rewriting edges corresponding to the target rewriting path from the previous and subsequent continuity relationship graph. Under the condition that the fixed and unchanging itemset remains unchanged, reconnect the original continuity edges before and after the deletion position according to the unified object identifier and unified time scale. Calculate the equipment state breakage, energy flow breakage, and carbon flow breakage for each mirror candidate path formed after reconnection. Sort the equipment state breakage, energy flow breakage, and carbon flow breakage in lexicographical order. Select the mirror candidate path with the first sorting number as the mirror path and output the mirror state sequence.

[0009] In a preferred embodiment, S3 includes: S3-1. Read the corresponding state items in the entity state sequence and mirror state sequence according to the unified time scale and unified object identifier, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and extract the meteorological data, load data and boundary data in the corresponding state items respectively, and output the difference pairs and impact item groups. S3-2. For meteorological data, load data and boundary data in the impact item group, compare the field values ​​and field positions of the entity state sequence and the mirror state sequence in the corresponding state item. When the field values ​​and field positions are consistent, the corresponding impact items are recorded as common source impact items, and they are merged into a common source impact item table according to the unified object identifier and unified time scale. S3-3. Based on the table of common source influence items, delete the synchronous change part corresponding to each common source influence item from the difference pair, retain the energy flow difference and carbon flow difference after deletion, and arrange them in a unified time scale order to form a difference sequence.

[0010] In a preferred embodiment, S4 includes: S4-1. Read each difference item in the difference sequence according to the unified time scale order. Form a difference vector by combining the energy flow difference and carbon flow difference of each difference item. Perform symbol mapping on each difference vector to obtain the direction vector. Construct an adjacency matrix, directional inner product matrix and difference ratio matrix with the same unified object identifier and adjacent time scales as objects. The directional inner product matrix is ​​formed by summing the products of the preceding and following direction vectors. The difference ratio matrix is ​​composed of the ratio of the absolute values ​​of the preceding and following energy flow differences and the ratio of the absolute values ​​of the preceding and following carbon flow differences. When the directional inner product matrix is ​​two and both terms in the difference ratio matrix are written in the same rational interval, the corresponding preceding and following difference items are recorded as a continuable pair. When the directional inner product matrix is ​​not two or the two terms in the difference ratio matrix are written in different rational intervals, the corresponding preceding and following difference items are recorded as a cut-off pair. Output the initial continuation graph.

[0011] In a preferred embodiment, S4 further includes: S4-2. Extract each candidate continuation chain based on the initial continuation graph. Construct an energy flow sequence matrix and a carbon flow sequence matrix for each candidate continuation chain. Perform first-order difference, second-order difference, and discrete convolution on the energy flow sequence matrix and the carbon flow sequence matrix respectively to obtain the energy flow change matrix and the carbon flow change matrix. Then calculate the Gram matrix, eigenvalue sequence, and sign-preserving sequence for the energy flow change matrix and the carbon flow change matrix respectively. When the eigenvalue sequence does not change the positive and negative order after being expanded according to the time scale and the sign-preserving sequence is continuous, retain the corresponding candidate continuation chain. When the positive and negative order of the eigenvalue sequence changes or the sign-preserving sequence is interrupted, cut off the candidate continuation chain at the corresponding position and output the corrected continuation chain. S4-3. Construct concatenation cost triplets for two adjacent concatenation chains before and after the time scale in the correction concatenation chain. The concatenation cost triplets include the difference between the cumulative energy flow value after concatenation and the cumulative energy flow value of the two chains before concatenation, the difference between the cumulative carbon flow value after concatenation and the cumulative carbon flow value of the two chains before concatenation, and the difference in encoding length between the sequence matrix after concatenation and the sequence matrix before concatenation. Sort the concatenation cost triplets in lexicographical order. For the concatenation result with the first sorting number, retain and recalculate the adjacent concatenation cost triplets. Repeat this process until there are no new concatenation pairs. Output the attribution fragment.

[0012] In a preferred embodiment, S5 includes: S5-1. Read the energy flow difference sequence and carbon flow difference sequence in each attribution segment, perform time-scale integration and accumulation on the energy flow difference sequence in each attribution segment to obtain the segment energy efficiency change, perform time-scale integration and accumulation on the carbon flow difference sequence in each attribution segment to obtain the segment carbon footprint change, and generate a segment result table with a unified object identifier, segment start time scale and segment end time scale. S5-2. Based on the start and end timestamps of each attribution segment in the segment result table, establish the segment succession relationship according to time sequence, and calculate the succession difference of energy efficiency change and carbon footprint change for adjacent segments with consistent object identification. When both are zero, the corresponding segment results are directly merged; when there are non-zero values ​​in either case, the corresponding segment results are retained, and the merge result table is output. S5-3. Read each merge result in the merge result table according to the unified object identifier and time order. Perform a second accumulation on the energy efficiency change and carbon footprint change in each merge result. Write the second accumulation result into the result record along with the target action identifier, time range and boundary data. Output the attributable emission reduction result corresponding to the target action.

[0013] An AI-powered integrated energy efficiency and carbon footprint management system for zero-carbon industrial parks, comprising a mapping modeling module, a mirror generation module, a difference stripping module, a fragment attribution module, and a result merging module. The mapping modeling module is used to acquire real-time data of the park corresponding to the target action. The real-time data of the park includes the equipment status before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data. The module performs synchronous mapping of the real-time data of the park in the digital twin according to a unified time scale and a unified object identifier, and outputs the status before the action. The mirror generation module is used to take the state before the action as input, retain the state rewriting process corresponding to the target action in the digital twin to form an entity state sequence, and remove the state rewriting process corresponding to the target action to form a mirror state sequence while keeping the meteorological data, load data and boundary data unchanged. The difference stripping module is used to read the corresponding state items in the entity state sequence and the mirror state sequence at the same time position, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and remove the common external influences based on the same meteorological data, load data and boundary data in the corresponding state items, and output the difference sequence. The segment attribution module is used to compare the energy flow difference direction and the carbon flow difference direction of adjacent difference items in the difference sequence. When the two directions are consistent, the corresponding difference items are continuously connected. When either direction is reversed, the connection is cut off and the attribution segment is output. The result merging module is used to accumulate the energy flow difference and carbon flow difference in each attribution segment to obtain the energy efficiency change and carbon footprint change corresponding to each attribution segment. The energy efficiency change and carbon footprint change corresponding to each attribution segment are then merged in chronological order, and the attributable emission reduction results corresponding to the target action are output.

[0014] The technical effects and advantages of this invention are as follows: 1. This scheme reduces reliance on comparisons between previous and subsequent time periods and historical baseline estimations by simultaneously forming entity state sequences and mirror state sequences within a digital twin and comparing them under the same time scale, the same object, and the same boundaries. This relatively improves the accuracy of identifying attributable emission reduction results of target actions. 2. Writing meteorological data, load data, and boundary data into a fixed and invariant itemset, and deleting the synchronously changing parts corresponding to the same source of external influences after the difference calculation, can relatively reduce the mixing of external disturbances with the attribution results, and make the energy flow difference and carbon flow difference more concentrated in reflecting the effect of the target action itself. 3. Establish a sequential relationship diagram, filter target rewriting paths, and generate entity state sequences and mirror state sequences accordingly. This can maintain the connection between the preceding and following states under a unified object identifier and a unified time scale, thereby relatively improving the continuity and verifiability of complex multi-source data entering the subsequent inference process. 4. Performing direction mapping, continuation chain correction, and concurrency cost comparison on the difference sequence can extract attribution fragments from continuous changes and cut off unstable continuation relationships, thereby relatively suppressing misattribution caused by local jumps, short-term fluctuations, or discontinuous disturbances. 5. Time-scaled integral accumulation, segment continuation merging, and result recording are performed on the energy flow difference and carbon flow difference in the attribution segment. This can transform the local difference results into attributable emission reduction results corresponding to the target action, time range, and boundary data, which is relatively convenient for subsequent review, settlement, and result reporting. Attached Figure Description

[0015] Figure 1 This is a flowchart outlining the method steps of the present invention; Figure 2 This is a schematic diagram of the system module structure of the present invention. Detailed Implementation

[0016] 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, and 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.

[0017] Refer to the instruction manual appendix Figure 1-2 The present invention provides an integrated AI-based energy efficiency and carbon footprint management method for zero-carbon industrial parks, comprising: S1. Obtain the real-time data of the park corresponding to the target action. The real-time data of the park includes the equipment status before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data. Perform synchronous mapping on the real-time data of the park in the digital twin according to a unified time scale and a unified object identifier, and output the status before the action. In this embodiment, S1 is used to organize the real-time park data that has been scattered across different acquisition links, different metering standards, and different object levels before the target action occurs into a pre-action state that can be directly read by the digital twin. The purpose is to first eliminate problems such as inconsistent time bases, inconsistent object identifiers, and discontinuous connection between previous and subsequent records. Then, equipment status, energy flow data, carbon flow data, meteorological data, load data, and boundary data are written into a unified data structure, providing a common entry boundary for subsequent target action rewriting, mirror state construction, and differential attribution. This implementation process includes the following steps: The purpose of S1-1 is to unify data from different sources onto the same time coordinate and the same object coordinate, so that subsequent writing and continuous verification have a unique reading basis. The input includes the original time value and object identifier corresponding to the equipment status before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data respectively. The original time value includes at least one of the acquisition time, reporting time or recording start and end time. The object identifier includes at least one of the source system identifier, name identifier and location identifier. The processing actions include: firstly, selecting the unified time scale currently used by the digital twin as the reference time scale. The unified time scale is preset by the park's energy and carbon management platform or determined by statistics according to the minimum common sampling granularity. When the original time value is a single point time, it is directly converted to the corresponding unified time scale position. When the original time value is a start and end period, the period is divided into multiple continuous time scale positions according to the unified time scale and the corresponding record index is copied. The converted records are then rearranged in ascending order according to a unified time scale to form a time scale alignment result. The processing of object identifiers includes: first, merging identical names in the order of identical names, identical installation locations, and identical connection relationships to obtain unified object identifiers; then, performing hierarchical expansion on the unified object identifiers according to fixed hierarchical paths of park level, region level, and equipment level to form a unique and continuous object hierarchy chain from top to bottom, and outputting the identifier alignment result; the outputs are the time scale alignment result and the identifier alignment result, and both are written into the subsequent object mapping table generation process; when the original time value is missing, the missing record is deleted and a missing flag is written; when an object identifier corresponds to multiple unified object identifiers at the same time, unique retention is performed in the order of identical installation location, identical connection relationship, and identical name, and records that fail to pass the unique retention are not read in the next step; The purpose of S1-2 is to write the real-time data of the park that has completed time-scale alignment and identifier alignment into a unified field structure, so that each record can be directly called by the digital twin with a unified time scale and unified object identifier as an index. The input includes time-scale alignment results and identifier alignment results. The processing actions include: establishing an object mapping table according to the unified time scale and unified object identifier. Each object mapping record includes fixed equipment status bit, energy flow bit, carbon flow bit, weather bit, load bit and boundary bit. Among them, the equipment status bit writes the on / off status, operating level or control status of the equipment before the action. The energy flow bit writes the corresponding fields of electricity, cooling, heating or fuel flow. The carbon flow bit writes the corresponding fields of carbon emissions, carbon factor or carbon emission conversion result. The weather bit writes the corresponding fields of temperature, humidity, irradiance or wind speed. The load bit writes the corresponding fields of production load, energy load or tenant occupancy load. The boundary bit writes the spatial boundary, equipment ownership boundary, energy supply boundary, accounting object boundary and time validity boundary. For the same type of field from different sources under the same unified object identifier and unified time scale, write different field items in the same position according to the preset field order, without changing the field position; the output is an object mapping table, and write the object mapping table into the preceding and following verification process for the next step to read; when a corresponding field is missing in a certain position, write an empty field mark in that position and retain its unified time scale and unified object identifier, and do not delete the entire object mapping record due to the missing single field; The purpose of S1-3 is to remove records with time breaks or object breaks from the object mapping table and restore the continuous relationship of the same unified object identifier under adjacent unified time scales, so that the output pre-action state has predictable continuity. The input is the object mapping table. The processing actions include: performing continuity checks on the equipment status, energy flow data, carbon flow data, meteorological data, load data and boundary data corresponding to the same unified object identifier under adjacent unified time scales in the object mapping table. The continuity check is performed in the following order: the unified object identifier is continuous, the unified time scale is adjacent, and no object switching occurs at the boundary position. When all three conditions are met, the corresponding records are retained and a continuity relationship is established. When the unified object identifier is not continuous, the unified time scale is not adjacent, or an object switching occurs at the boundary position, the next record is recorded as a break record, the continuity relationship of the break record is deleted from the object mapping table, and the continuous records before and after the break position are searched again according to the unified object identifier. The searched continuous records are then reconnected. When reconnecting, only records with the same unified object identifier and consistent boundary bits are allowed to be connected. Connection is not performed across objects or boundaries. The output is the pre-action state, which is a set of states organized according to the unified object identifier and unified time scale, and is written into the digital twin for S2 to read. When there are no records that can be continued before and after the break position, the verified record before the break position is retained, and the record after the break position is written separately into the pre-action state as a new starting record. Through the above processing, the pre-action equipment status, energy flow data, carbon flow data, meteorological data, load data and boundary data are first unified to the same time scale and the same object coordinates, then written into a unified field structure, and finally formed into a pre-action status that can be directly entered into the digital twin after sequential verification. This provides unique, continuous and verifiable data to enter the boundary for subsequent target action rewriting, entity state sequence formation and mirror state sequence formation. In practical applications: For example, before the park adjusts the group control strategy for the chiller plant, the equipment status uploaded by the chiller controller is sampled at the second level, the energy flow data uploaded by the electricity meter is sampled at the fifteen-minute level, the carbon flow data is calculated from the electricity consumption and the carbon factor of the current period, the meteorological data comes from the hourly records of the meteorological station, the load data comes from the tenant's energy consumption plan, and the boundary data comes from the equipment ownership ledger. In this implementation method, the above different original time values ​​are first converted to a unified time scale, for example, a unified fifteen-minute time scale. Then, the chiller number, power distribution circuit number, and energy station name are merged into the same unified object identifier and its hierarchical path. Subsequently, each type of field is written into the corresponding position in the object mapping table. Finally, the broken records caused by communication interruption are deleted and the records before and after can be continued, so as to obtain the pre-action status of the chiller plant before the adjustment of the group control strategy, which can be directly called by the subsequent digital twin to build the entity state sequence and mirror state sequence.

[0018] S2. Using the state before the action as input, retain the state rewriting process corresponding to the target action in the digital twin to form an entity state sequence, and remove the state rewriting process corresponding to the target action while keeping the meteorological data, load data and boundary data unchanged to form a mirror state sequence. In this embodiment, S2 is used to expand the pre-action state within the digital twin into a physical state sequence containing the target action rewriting result, and a mirror state sequence that does not contain the target action rewriting result under the same meteorological data, load data, and boundary data conditions. The overall purpose is to first identify the continuation position of the actual impact of the target action from the pre-action state, then determine a unique rewriting path from multiple rewriting paths, and generate the physical state sequence and mirror state sequence accordingly. This provides dual-sequence inputs of the same source, same boundary, and same target for subsequent difference calculation and removal of similar external influences. This implementation process includes the following steps: The purpose of S2-1 is to convert the pre-action state into a succession relationship structure that allows the target action to be inserted, replaced, and deleted, while simultaneously fixing the common input items between the entity state sequence and the mirror state sequence that cannot be modified. The input quantities are the equipment status, energy flow data, carbon flow data, meteorological data, load data, and boundary data in the pre-action state, as well as the target action's target object, action period, and action field. The processing actions include: firstly, reading each state record in the pre-action state according to the unified object identifier and unified time scale, and constructing each state record into a state node. A raw succession edge is constructed between every two state nodes with the same unified object identifier and adjacent unified time scales, thereby forming a succession relationship graph in the digital twin. The state node at least includes equipment status bit, energy flow bit, carbon flow bit, meteorological bit, load bit, and boundary bit, and the raw succession edge at least includes edge start point, edge end point, and succession field index. Subsequently, the target action's object, time period, and field of action are read, and a delineation operation is performed in the preceding and following relationship graph. The delineation rules are as follows: if the unified object identifier corresponding to the edge's start and end points falls within the target object range, the unified time scale covered by the edge falls within the time period range, and the index of the corresponding following field falls within the field of action range, then the edge is recorded as a candidate rewrite edge, and all candidate rewrite edges form a candidate rewrite edge set. Simultaneously, the meteorological data, load data, and boundary data in the pre-action state are directly copied according to the unified object identifier and unified time scale to form a fixed invariant itemset. The fixed invariant itemset is used... The values ​​of meteorological position, load position and boundary position are kept consistent between the constrained entity state sequence and the mirror state sequence at the same object and the same time scale position; the output is a sequential relationship diagram, a candidate rewriting edge set and a fixed invariant itemset, and is written into the candidate rewriting path filtering process for the next step of reading; when the target action does not have a corresponding unified object identifier in the state before the action, no candidate rewriting edge is generated and the target action is recorded as an invalid action; when the unified time scale covered by the action period is missing, only the existing unified time scale position is circled as a candidate rewriting edge, and no non-existent state node is created; The purpose of S2-2 is to determine the target rewriting path that best matches the target action and has the most continuous sequence from multiple candidate rewriting paths corresponding to the candidate rewriting edge set, so that the generation of subsequent entity state sequences has a unique rewriting basis. The inputs are the candidate rewriting edge set, the continuity graph, and the fixed invariant itemset. The processing actions include: firstly, in the continuity graph, sequentially connecting the candidate rewriting edges that are connected end to end and cover the same object identifier and the same time scale to form multiple candidate rewriting paths. Each candidate rewriting path includes at least the path start node, the path end node, and the entire path interior. The candidate rewrite edges are then calculated. Subsequently, for each candidate rewrite path, the equipment state continuity cost, energy flow continuity cost, and carbon flow continuity cost are calculated for the two adjacent edges. The equipment state continuity cost is obtained by comparing the equipment state field value at the end of the previous edge with the equipment state field value at the beginning of the next edge to obtain the number of inconsistent fields. The energy flow continuity cost is obtained by subtracting the energy flow field value at the end of the previous edge with the energy flow field value at the beginning of the next edge and then summing the absolute values. The carbon flow continuity cost is obtained by subtracting the carbon flow field value at the end of the previous edge with the carbon flow field value at the beginning of the next edge and then summing the absolute values. The three types of costs for all adjacent edges within the same path are accumulated to form a total cost triplet for that path. The path is retained under the condition that the fixed invariant itemset is consistent item by item and the unified object identifier corresponding to the path is continuous before and after. The consistent fixed invariant itemset means that the meteorological position, load position and boundary position within the path coverage area are completely consistent with the corresponding field values ​​of the same position in the state before the action. The continuous unified object identifier means that the start and end points of all edges within the path correspond to the same unified object identifier or continuous object identifiers under the same level of expansion chain. Among the retained candidate rewriting paths, the path is first sorted by equipment status, then by energy flow, and then by carbon flow to obtain the current round sorting result. The total cost triplet is then recalculated using the retained path set of the current round as input to form the next round sorting result. When the candidate rewriting path with the first order number in two consecutive rounds of sorting is consistent, the candidate rewriting path is determined as the target rewriting path. The output is the target rewriting path, and the target rewriting path is written into the entity state sequence generation process for the next step to read. When there is only one candidate rewriting path, the path is directly determined as the target rewriting path. When all candidate rewriting paths do not meet the requirements of the fixed invariant itemset being consistent item by item or the unified object identifier being continuous before and after, the current round of rewriting is terminated and the state before the action is returned as the initial input of the entity state sequence. The purpose of S2-3 is to actually write the target rewriting path into the continuity structure of the digital twin and propagate it along the rewriting result to the preceding and following associated states, forming a sequence of entity states containing the influence of the target action. The input quantities are the state before the action, the target rewriting path, and the fixed invariant itemset. The processing actions include: first, finding the original continuity edge corresponding to the position of the target rewriting path in the preceding and following continuity relationship graph, and replacing the corresponding original continuity edge one by one with the rewriting edge in the target rewriting path, so that the target action obtains the actual continuity position in the graph structure; then, performing time-scale deduction based on the replaced relationship graph. The deduction order is to read the state records of the rewriting start position and its subsequent associated positions in ascending order of unified time scale. At each unified time scale position, first read the equipment state, energy flow data, and carbon flow data after the rewriting of the previous time scale, and then write the corresponding field value of the current time scale according to the action field indicated by the target rewriting path, and use the written result as the input state of the next time scale. Among them, meteorological data, load data, and boundary data are always directly taken from the fixed invariant itemset and do not participate in the rewriting propagation. After each round of simulation, the positions where fields were written in this round are recorded to form a set of rewritten positions, and the field values ​​after writing at each rewritten position are recorded to form the rewritten state values. The set of rewritten positions and the rewritten state values ​​obtained in the current round of simulation are compared item by item with the results of the previous round of simulation. When the set of rewritten positions is completely the same and the rewritten state values ​​at the corresponding positions are completely the same, the simulation stops and the state records at all unified time scale positions at the time of stopping are output in chronological order as an entity state sequence. The output quantity is the entity state sequence and is written into the subsequent difference calculation process for the next step to read. When a unified time scale position lacks the previous time scale input state, the original record of the unified time scale position in the state before the action is used as the current simulation input. When a rewritten field does not exist at the current unified time scale position, the original field value at that position is retained and no new field is generated. The purpose of S2-4 is to restore the original continuity after deleting the target rewritten edge without introducing the target action rewriting result, and to determine a unique mirror path from multiple reconnected paths to form a mirror state sequence corresponding to the entity state sequence. The inputs are the preceding and following continuity graph, the target rewritten path, and a fixed invariant itemset. The processing actions include: first, deleting all rewritten edges corresponding to the target rewritten path from the preceding and following continuity graph, while retaining the state nodes before and after the deletion position; then, re-searching for the original continuity edges before and after the deletion position according to the unified object identifier and unified time scale. The search rule is to only connect edges with the same unified object identifier and unified time scale before and after the deletion position. Adjacent state nodes are used to form multiple mirror candidate paths. Then, for each mirror candidate path, the device state breakage, energy flow breakage, and carbon flow breakage are calculated. The device state breakage is obtained by comparing the device state field value of the node before the deletion position with the device state field value of the first node after reconnection, and the number of inconsistent fields is obtained. The energy flow breakage is obtained by subtracting the energy flow field value of the node before the deletion position with the energy flow field value of the first node after reconnection, and then summing the absolute values. The carbon flow breakage is obtained by subtracting the carbon flow field value of the node before the deletion position with the carbon flow field value of the first node after reconnection, and then summing the absolute values. All candidate mirror paths are sorted lexicographically by the three types of fracture quantities: equipment status fracture quantity, energy flow fracture quantity, and carbon flow fracture quantity. The candidate mirror path with the first sorting number is determined as the mirror path. Then, the equipment status, energy flow data, and carbon flow data within the coverage area of ​​the mirror path are written out in ascending order according to a unified time scale, and directly written into the meteorological data, load data, and boundary data in the fixed invariant term set to obtain the mirror status sequence. The output quantity is the mirror status sequence, which is written into the subsequent difference calculation process for the next step of reading. When only one candidate mirror path is formed after deleting the target and rewriting the path, the candidate mirror path is directly determined as the mirror path. When there are no reconnectable original continuation edges before and after the deletion position, the original status record before the deletion position is retained, and the original status record after the deletion position is used as the new mirror starting position to continue writing. Through the above processing, the state before the action is first constructed into a succession relationship graph, then the candidate rewriting edge set is delineated according to the target action, and the unique target rewriting path is screened out under the constraint of the fixed invariant itemset, thus forming an entity state sequence containing the influence of the target action; at the same time, the system deletes the rewriting edge corresponding to the target rewriting path, restores the original succession relationship and screens out the unique mirror path, thus forming a mirror state sequence that does not contain the influence of the target action but maintains the same meteorological data, load data and boundary data conditions, so that the subsequent difference calculation has the same source input, same boundary input and same standard input; In practical applications: For example, when a park executes a target action on the energy station chiller group control logic, the target action applies to the unified object identifier corresponding to the chiller unit, the time period is from 10:00 to 12:00 on the same day, and the affected fields are the start / stop field in the equipment status bit and the cooling capacity allocation field in the energy flow bit. In this implementation method, a sequential relationship diagram of the chiller units within the range of 10:00 to 12:00 is first established in the digital twin, and all candidate rewrite edges involving the start / stop field and cooling capacity allocation field within this time range are identified. Then, the equipment status continuity cost is calculated for each of the multiple candidate rewrite paths. Energy flow continuity cost and carbon flow continuity cost are used to determine a target rewriting path, which is written into the relationship graph and the entity state sequence is derived time-by-time. Then, the rewriting edge corresponding to the target rewriting path is deleted, and the original continuity edges before and after the deletion position are reconnected. The unique mirror path is selected by lexicographically arranging the equipment state breakage amount, energy flow breakage amount, and carbon flow breakage amount, and a mirror state sequence is generated. This results in two state sequences that differ only in the target action rewriting result and are completely consistent in other meteorological data, load data, and boundary data, which are used for subsequent difference stripping and attribution fragment extraction.

[0019] S3. Read the corresponding state items in the entity state sequence and mirror state sequence at the same time position, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and remove the same external influences based on the same meteorological data, load data and boundary data in the corresponding state items, and output the difference sequence. In this embodiment, S3 is used to establish a difference correspondence between the entity state sequence and the mirror state sequence under the same object, the same time scale, and the same boundary, and to separate the synchronous change part caused by the same meteorological data, the same load data, and the same boundary data. The overall purpose is to separate the changes caused by the target action itself from the changes caused by external common disturbances, so that the subsequent attribution segment extraction only focuses on the energy flow difference and carbon flow difference actually formed by the target action. This implementation process includes the following steps: The purpose of S3-1 is to first align the entity state sequence and the mirror state sequence to the same reading position, and form difference pairs and influence item groups at the same reading position that can directly enter the subsequent recognition process; the input is the entity state sequence and the mirror state sequence; the processing actions include: firstly, establishing a corresponding reading relationship according to the unified object identifier and the unified time scale. The reading rule is that the entity state sequence and the mirror state sequence form a set of corresponding state items at the same position of unified object identifier and unified time scale. Each set of corresponding state items retains only one pair of records. It is not allowed that the same unified object identifier and the same unified time scale correspond to multiple pairs of records. When there are multiple records under the same unified object identifier and the same unified time scale, the unique record that has been retained in the previous state generation process is read. Subsequently, the energy flow and carbon flow bits in the corresponding state items of this group are read respectively. The energy flow field values ​​at the same field position are subtracted to obtain the energy flow difference, and the carbon flow field values ​​at the same field position are subtracted to obtain the carbon flow difference. The energy flow difference is obtained by subtracting the corresponding field value of the mirror state sequence from the corresponding field value of the entity state sequence, and the carbon flow difference is obtained by subtracting the corresponding field value of the mirror state sequence from the corresponding field value of the entity state sequence. Then, the meteorological bit, load bit, and boundary bit in the corresponding state items of this group are extracted to form an influence item group. The influence item group includes at least the field values ​​and field positions of the meteorological data field, load data field, and boundary data field in the entity state sequence and the mirror state sequence. The output consists of difference pairs and influence item groups. The difference pairs must include at least a uniform object identifier, a uniform time scale, an energy flow difference, and a carbon flow difference. The difference pairs and influence item groups are written into the same-source influence item identification process for the next step of reading. When the entity state sequence and the mirror state sequence have records on only one side at a certain uniform object identifier and uniform time scale position, the record on that side is deleted and no difference pair is generated. When the energy flow field or carbon flow field in the corresponding state item is missing, a missing marker is written into the difference pair and the uniform object identifier and uniform time scale are retained for subsequent skip processing. The purpose of S3-2 is to identify common impact items from the impact item group that share the same source, values, and field positions between the entity state sequence and the mirror state sequence, so that subsequent deletion of synchronous changes has a clear field basis. The input is the impact item group. The processing actions include: performing field value comparison and field position comparison on the meteorological data, load data, and boundary data in each impact item group. The field value comparison is to perform a field-by-field comparison of the same-named field values ​​in the entity state sequence and the mirror state sequence at the same unified object identifier and the same unified time scale position. The field position comparison is to perform a field-by-field comparison of the fixed field positions of the corresponding fields in the meteorological position, load position, and boundary position. When the field values ​​and field positions are consistent, the impact item corresponding to that field is recorded as a common-source impact item. When the field values ​​or field positions are inconsistent, it is not recorded as a common-source impact item. Subsequently, all source-related impact items are merged according to the unified object identifier and unified time scale. Meteorological data source-related impact items, load data source-related impact items, and boundary data source-related impact items identified under the same unified object identifier and unified time scale are written into the same merge record to form a source-related impact item table. The source-related impact item table includes at least a unified object identifier, a unified time scale, a set of meteorological fields from the same source, a set of load fields from the same source, and a set of boundary fields from the same source. The output is the source-related impact item table, which is written into the synchronous change deletion process for the next step. When all meteorological data, load data, or boundary data in a certain impact item group do not meet the requirements of consistent field values ​​and consistent field positions, an empty field set is written to the corresponding position in the source-related impact item table without interrupting the generation of the merge record at the current unified object identifier and unified time scale position. The purpose of S3-3 is to subtract the synchronous changes caused by common external conditions from the original difference pairs based on the same source influence item table, retain only the remaining differences formed by the target action, and organize the remaining differences into a difference sequence that can be directly input into the attribution fragment for extraction according to a unified time scale. The inputs are difference pairs and the same source influence item table. The processing actions include: first, matching the same source influence item table with the difference pairs one by one according to the unified object identifier and unified time scale; for each difference pair, determining the synchronous change part according to the same source meteorological field set, same source load field set and same source boundary field set recorded in the same source influence item table. The synchronous change part refers to the same-direction change of meteorological data, load data and boundary data with the same field position and the same field value in the entity state sequence and the mirror state sequence, which corresponds to the current difference pair. Subsequently, the corresponding synchronous change portions are deleted from the original energy flow difference and the original carbon flow difference in the difference pair. The deletion rules are as follows: when a corresponding common source influence item exists in the common source influence item table, the corresponding synchronous change in energy flow is deducted from the original energy flow difference, and the corresponding synchronous change in carbon flow is deducted from the original carbon flow difference; when no corresponding common source influence item exists in the common source influence item table, the original energy flow difference and the original carbon flow difference remain unchanged. After processing all difference pairs, the energy flow difference and carbon flow difference after deleting the synchronous change portions are grouped according to a unified object identifier and arranged in ascending order according to a unified time scale to form a difference sequence. The output is the difference sequence, which is written into the subsequent attribution fragment extraction process for the next step of reading. When the synchronous change portion corresponding to a certain difference pair cannot be calculated, the original energy flow difference and the original carbon flow difference of the difference pair are retained, and an unstripped mark is written to the difference sequence for the subsequent attribution fragment extraction process to skip the position or process it separately. Through the above processing, the entity state sequence and the mirror state sequence first form a one-to-one difference pair under the unified object identifier and unified time scale. Then, the common source influence items formed by the same meteorological data, the same load data and the same boundary data are identified from the difference pairs. Finally, the synchronous change parts corresponding to these common source influence items are deleted to obtain the difference sequence that retains only the influence of the target action, thus providing the input basis for the subsequent extraction of attribution fragments after the common source external influences have been stripped away. In practical applications: For example, after the park executes the target action on the chiller group control logic, at a certain chiller unit corresponding to the unified object identifier and the unified time stamp position of 11:15, if the outdoor temperature field value in the meteorological position of the entity state sequence and the mirror state sequence are the same, the tenant load field value in the load position is the same, and the energy supply boundary field value in the boundary position is the same, then these fields are written into the same source influence item table; if the original energy flow difference at this position is the cooling capacity difference 12, the original carbon flow difference is the carbon emission difference 3, and the synchronous changes corresponding to the same source influence item are the cooling capacity change 2 and the carbon emission change 0.5 respectively, then 2 is subtracted from the original energy flow difference 12 to obtain the deleted energy flow difference 10, and 0.5 is subtracted from the original carbon flow difference 3 to obtain the deleted carbon flow difference 2.5, and written into the difference sequence in the unified time stamp order for subsequent extraction of attribution fragments according to the energy flow difference direction and the carbon flow difference direction.

[0020] S4. Compare the energy flow difference direction and carbon flow difference direction of adjacent difference items in the difference sequence. When the two directions are consistent, the corresponding difference items are continuously connected. When either direction is reversed, the connection is cut off and the attribution segment is output. In this embodiment, S4 is used to extract the range of differences that can truly be continuously attributed to the target action from the difference sequence after the external influences of the same source have been removed. Its overall purpose is to first establish an initial continuation relationship based on the relationship between the direction and magnitude of the difference, then cut off the continuation part that is superficially in the same direction but internally unstable through structural change verification, and finally perform parallel cost comparison and unique retention on adjacent correction continuation chains, thereby outputting the attribution fragment that can be directly entered into the result merging process; the implementation process includes the following steps: The purpose of S4-1 is to first use a unified object identifier, unified time scale, directional relationship, and difference ratio relationship to screen out the initial continuation relationship that can enter the subsequent correction process, so that adjacent difference terms in the difference sequence have a clear continuation basis; the input is the difference sequence; the processing actions include: reading each difference term in the difference sequence in ascending order of the unified time scale, forming a two-dimensional difference vector from the energy flow difference and carbon flow difference in each difference term, and then performing sign mapping on the two components in the two-dimensional difference vector, where positive values ​​are mapped to one, negative values ​​are mapped to negative one, and zero values ​​are mapped to zero, thereby obtaining a two-dimensional direction vector; Subsequently, an adjacency matrix is ​​constructed using adjacent difference terms with consistent object identifiers and consistent time scales. Positions with adjacent relationships are written as 1, and positions without adjacent relationships are written as 0. The directional inner product matrix and the difference ratio matrix are then calculated at the corresponding positions in the adjacency matrix. The directional inner product matrix is ​​obtained by summing the products of the two directional vectors. The difference ratio matrix consists of the ratio of the absolute values ​​of the energy flow difference and the ratio of the absolute values ​​of the carbon flow difference. The absolute value ratio is calculated with the numerator being the absolute value of the latter difference term and the denominator being the absolute value of the former difference term. When the denominator is zero and the numerator is not zero, it is recorded as an unbounded ratio. When both the numerator and denominator are zero, it is recorded as a constant ratio. To ensure the accuracy of the difference ratio matrix, this implementation writes all difference ratios into a preset rational interval table. This rational interval table is preset by the system configuration file and includes at least five interval types: zero to half, half to one, one to two, two to four, and four or more. When a difference ratio falls into a corresponding interval, it is written as the corresponding interval identifier. When the directional inner product matrix is ​​two and the ratio of the absolute values ​​of the energy flow difference before and after the difference to the absolute values ​​of the carbon flow difference before and after the difference falls into the same rational interval, the corresponding difference terms are recorded as a contiguous pair. When the value of the directional inner product matrix is ​​not two, or when the ratio of the absolute values ​​of the energy flow difference and the carbon flow difference are written in different rational intervals, the corresponding difference terms are recorded as cut-off pairs. The output is the initial continuation graph, which includes at least difference term nodes, continuable pairs, and cut-off pairs, and is written into the candidate continuation chain extraction process for the next step. When the energy flow difference or carbon flow difference is missing in a certain difference term, the difference term is directly recorded as an isolated node and does not participate in the calculation of the directional inner product matrix and the difference ratio matrix. The purpose of S4-2 is to re-verify the connectable relationships in the initial continuity graph by utilizing the difference change structure rather than just the sign consistency relationship, in order to eliminate candidate continuity chains with unstable internal change trends. The input is the initial continuity graph. The processing actions include: firstly, extracting each candidate continuity chain from the initial continuity graph, the extraction rule being to continuously traverse along the connectable pairs in a uniform time-scale increasing direction until a cut-off pair or isolated node is encountered, thus obtaining each candidate continuity chain; then, constructing an energy flow sequence matrix and a carbon flow sequence matrix for each candidate continuity chain, where the row square... The matrix is ​​arranged in ascending order according to a unified time scale, and the columns are arranged in the order of the difference terms within the chain. The matrix elements are written with the energy flow difference and carbon flow difference at the corresponding time scale positions. First-order difference is performed on the energy flow sequence matrix and the carbon flow sequence matrix along the unified time scale direction to obtain the change in adjacent time scales. Then, second-order difference is performed on the first-order difference result to obtain the change acceleration. Subsequently, the first-order difference result is discretely convolved with a fixed convolution kernel. The fixed convolution kernel is preset by the system configuration file as a three-term convolution kernel with a coefficient sum of one, which is used to smooth local jitter and output the energy flow change matrix and the carbon flow change matrix. Next, the Gram matrix is ​​calculated for the energy flow variation matrix and the carbon flow variation matrix respectively. The Gram matrix is ​​obtained by multiplying the variation matrix with its transpose. Eigenvalue decomposition is then performed on the Gram matrix to obtain the eigenvalue sequence expanded according to the unified time scale. At the same time, the sign of each position in the variation matrix is ​​read row by row along the unified time scale direction to form a sign-preserving sequence. The sign of adjacent time scale positions remains unchanged and is recorded as preserved, while the sign changes and is recorded as interrupted. When the eigenvalue sequence maintains the original positive and negative order after being expanded according to the unified time scale and the sign-preserving sequence is continuous, the corresponding candidate continuation chain is retained. When the positive and negative order of the eigenvalue sequence changes after being expanded according to the unified time scale, or the sign-preserving sequence is interrupted, the candidate continuation chain is cut off at the corresponding time scale position to form at least one new corrected continuation chain. The output is the corrected continuation chain and is written into the concatenation cost calculation process for the next step. When a candidate continuation chain contains only one difference term, difference, discrete convolution and eigenvalue decomposition are not performed, and the candidate continuation chain is directly written as the corrected continuation chain. The purpose of S4-3 is to perform parallel comparison and unique retention on adjacent correction continuation chains before and after the time scale, so that the attribution fragments maintain the consistency of internal changes and avoid fragmented results that can be parallelized but not merged under the same unified object identifier; the input is the correction continuation chain; the processing actions include: firstly, finding two adjacent correction continuation chains before and after the time scale under the same unified object identifier, and constructing a parallel cost triplet for each pair of adjacent correction continuation chains; the first cost value is the difference between the cumulative energy flow value after parallelization and the cumulative energy flow value of the two chains before parallelization. The calculation method is to first connect the two correction continuation chains end to end to form a parallel chain, then accumulate all energy flow differences in the parallel chain to obtain the cumulative energy flow value after parallelization, and then accumulate all energy flow differences in the two correction continuation chains before parallelization and sum them to obtain the cumulative energy flow value of the two chains before parallelization. The difference between the two is the first generation value; The second cost value is the difference between the cumulative carbon flow value after merging and the cumulative carbon flow value of the two chains before merging, calculated in the same way as the first generation value. The third cost value is the difference in encoding length between the sequence matrix after merging and the sequence matrix before merging. The calculation method is as follows: first, write the energy flow sequence matrix and carbon flow sequence matrix after merging into fixed-length encoding strings item by item according to a unified time scale order; then, write the energy flow sequence matrix and carbon flow sequence matrix corresponding to the two corrective continuation chains before merging into fixed-length encoding strings item by item; calculate the difference after counting the length of the encoding strings respectively to obtain the third generation value; then, sort all the merging cost triplets in lexicographical order according to the first generation value, the second generation value, and the third cost value; the merging result with the first sequence number is used as the current round retention result; and the corresponding two corrective continuation chains are replaced with the current round retention result; then, the new merging cost triplet is recalculated with the set of corrective continuation chains after replacement. The above concatenation, replacement, and recalculation process is repeated until there are no new concatenation pairs in the current round, or the set of concatenation pairs in the current round is completely consistent with the set of concatenation pairs in the previous round. Attribution fragments are then output. The output is the attribution fragment and is written into the subsequent energy efficiency change and carbon footprint change merging process for the next step. When two correction concatenation chains are adjacent in time scale but have different unified object identifiers, or when the concatenation chain crosses the boundary data change position, the concatenation cost triplet is not constructed, and the two correction concatenation chains are directly output separately. Through the above processing, the difference sequence is first screened out for the initial continuation relationship by the direction vector, the direction inner product matrix and the difference ratio matrix, and then internal structure is checked by the energy flow change matrix, the carbon flow change matrix, the Gram matrix, the eigenvalue sequence and the sign preservation sequence. Finally, the attribution fragment is formed by the parallel cost triplet and the lexicographical unique preservation rule, so that the output result simultaneously meets the three conditions of consistent direction, amplitude connection and continuous internal change, avoiding erroneous paralleling caused by relying solely on the same direction relationship; In practical applications: For example, after the park executes the target action on the refrigeration unit group control logic, a certain refrigeration unit forms four adjacent difference terms between 11:00 and 11:45. The energy flow difference and carbon flow difference of the four difference terms are all positive, the directional inner product matrix is ​​two in adjacent positions, and the ratio of the absolute values ​​of the energy flow difference and the absolute values ​​of the carbon flow difference before and after each fall into the interval of one to two. Then, these four difference terms are first written into the connectable pairs in the initial continuity diagram; after further constructing the energy flow sequence matrix and carbon flow sequence matrix for this continuity chain, if the first-order difference, second-order difference, and If the eigenvalue sequences corresponding to the energy flow change matrix and carbon flow change matrix obtained by discrete convolution maintain their original positive and negative order and the signs remain continuous, then this continuation chain is retained as a corrected continuation chain. If there is another corrected continuation chain with adjacent time scales between 11:45 and 12:00, then the concatenation cost triplet of the two corrected continuation chains is calculated, and the concatenation result with the first order number is retained in lexicographical order. Finally, the attribution segment of the chiller unit within this time range is output for subsequent merging to form the attributable emission reduction result corresponding to the target action.

[0021] S5. Accumulate the energy flow difference and carbon flow difference in each attribution segment to obtain the energy efficiency change and carbon footprint change corresponding to each attribution segment. Merge the energy efficiency change and carbon footprint change corresponding to each attribution segment in chronological order and output the attributable emission reduction results corresponding to the target action. In this embodiment, S5 is used to convert the local difference results corresponding to the attribution fragments into final results that can be directly used for the determination of the effect of the target action. Its overall purpose is to first perform time-scaled integration and accumulation on the energy flow difference sequence and carbon flow difference sequence within each attribution fragment to obtain the fragment's energy efficiency change and fragment's carbon footprint change. Then, the fragment results that are continuous in time and have the same object are merged. Finally, under the constraints of unified object identification, time range, and boundary data, the attributable emission reduction results corresponding to the target action are generated, thereby transforming the results of the previous fragment extraction into verifiable, recordable, and callable result outputs. This implementation process includes the following steps: The purpose of S5-1 is to transform the difference sequence in each attribution segment into segment energy efficiency change and segment carbon footprint change with physical and accounting significance, and record the segment results using a unified field structure. The input is the energy flow difference sequence and carbon flow difference sequence in each attribution segment. The processing actions include: firstly, reading the energy flow difference and carbon flow difference at each unified time scale position of each attribution segment, and reading the time scale interval between adjacent unified time scales, where the time scale interval is directly taken as the adjacent time difference of the unified time scale; then, for the same attribution... The energy flow difference sequence within a segment is time-scaled integrally accumulated. The calculation method is to multiply the energy flow difference at each unified time scale position by the corresponding time scale interval, and then sequentially accumulate all the product results to obtain the segment energy efficiency change of the attribution segment. The same calculation method is used for the carbon flow difference sequence within the same attribution segment to obtain the segment carbon footprint change of the attribution segment. When the unified time scale interval is fixed, each time scale interval takes the same value. When the unified time scale interval is not fixed, each time scale interval takes the value according to the actual time difference of the corresponding adjacent unified time scales. After the calculation is completed, a segment result table is generated using a unified object identifier, segment start time scale, segment end time scale, segment energy efficiency change, and segment carbon footprint change. The segment start time scale is the first unified time scale of the attribution segment, and the segment end time scale is the last unified time scale of the attribution segment. The output is the segment result table, which is written into the segment continuation relationship establishment process for further reading. When an attribution segment contains only one unified time scale position, the segment energy efficiency change and segment carbon footprint change are generated directly by multiplying the difference corresponding to that unified time scale position by the unified time scale interval. When some unified time scale positions in an attribution segment lack energy flow difference or carbon flow difference, the missing positions are skipped and the calculation results of the remaining positions are retained without interrupting the generation of the segment result for that attribution segment. The purpose of S5-2 is to identify time-sequentially connected segment results with consistent unified object identifiers from the segment result table, and to determine whether these segment results can be directly merged into the same continuous result. The input is the segment result table. The processing actions include: first, based on the segment start time and segment end time of each attribution segment in the segment result table, establishing segment continuity relationships according to time sequence. The establishment rule is that when the segment start time of the later segment result and the segment end time of the earlier segment result are adjacent and their unified object identifiers are consistent, they are recorded as having a segment continuity relationship. Subsequently, for each group of segment results with segment continuity relationships, the continuity difference of energy efficiency change and the continuity difference of carbon footprint change are calculated respectively. The continuity difference of energy efficiency change is obtained by subtracting the continuity value of the segment energy efficiency change of the earlier segment result from the segment energy efficiency change start continuity value of the later segment result, and the continuity difference of carbon footprint change is obtained by subtracting the continuity value of the segment carbon footprint change of the earlier segment result from the segment carbon footprint change start continuity value of the later segment result. In this embodiment, the starting continuation value of the segment energy efficiency change is taken as the starting point value of the energy flow difference integration corresponding to the first unified time scale position of the subsequent segment result, and the ending continuation value of the segment energy efficiency change is taken as the ending point value of the energy flow difference integration corresponding to the last unified time scale position of the preceding segment result. The starting continuation value and the ending continuation value of the segment carbon footprint change are determined in the same way. When both the energy efficiency change difference and the carbon footprint change difference are zero, the corresponding preceding and following segment results are directly merged. The starting time scale of the merged segment is taken as the starting time scale of the preceding segment result, and the ending time scale of the merged segment is taken as the ending time scale of the following segment result. The change in energy efficiency is the sum of the changes in energy efficiency of the two segments before and after the merger. The change in carbon footprint of the merged segment is the sum of the changes in carbon footprint of the two segments before and after the merger. When there are non-zero values ​​in the continuity difference of energy efficiency change and the continuity difference of carbon footprint change, the corresponding segment results are retained separately and the merger is not performed. The output is a merged result table, which is written into the final result record generation process for the next step of reading. When there are multiple segment results with the same time sequence under the same unified object identifier, the segment continuity relationship is established first according to the segment with the earlier start time. When the segment start time is the same but the segment end time is different, the segment with the earlier end time is read first. The purpose of S5-3 is to further organize the results of the segment merging already completed in the merge result table into the final result record corresponding to the target action according to the unified object identifier and time sequence, and output the attributable emission reduction results; the input quantities are the merge result table, the target action identifier, and the boundary data; the processing actions include: firstly, reading each merge result in the merge result table in groups according to the unified object identifier, then reading the segment energy efficiency change and segment carbon footprint change in each merge result in time sequence within each group, and performing the summation again to obtain the total energy efficiency change and total carbon footprint change under the unified object identifier within the scope of the target action; then extracting the earliest segment start time and the latest segment end time of the merge result covered by the group to form a time range, and reading the boundary data to which the corresponding merge result belongs, wherein the boundary data includes at least the spatial boundary, equipment ownership boundary, energy supply boundary, accounting object boundary, and time effective boundary; Finally, the accumulated results are written into the result record along with the target action identifier, time range, and boundary data. The result record includes at least the target action identifier, unified object identifier, time range, total energy efficiency change, total carbon footprint change, and boundary data. The output is the attributable emission reduction result corresponding to the target action, and it is written into the result record area of ​​the digital twin for subsequent display, verification, and retrieval. When there is only one merged result under a unified object identifier, the merged result is directly written into the result record, and cross-result re-accumulation is not performed. When the boundary data is inconsistent among multiple merged results of the same unified object identifier, they are grouped in the order of consistent time effective boundary, consistent accounting object boundary, and consistent energy supply boundary, and then written into the result record respectively. Merged results that do not have inconsistent boundary data are re-accumulated. Through the above processing, each attribution fragment is first converted into fragment energy efficiency change and fragment carbon footprint change, then fragment merging is performed according to the time sequence and unified object identifier, and finally attributable emission reduction results are formed under the constraints of target action identifier, time range and boundary data, so that the results of preceding difference stripping and attribution fragment extraction are organized into the final output that can directly correspond to the target action, object range and action time. In practical applications: For example, after a park executes a target action on the refrigeration unit group control logic, a certain refrigeration unit forms three attribution segments under a unified object identifier. The first attribution segment covers 10:00 to 10:30, the second attribution segment covers 10:30 to 11:00, and the third attribution segment covers 11:15 to 11:45. First, time-scaled integrals are performed on the energy flow difference sequence and carbon flow difference sequence within the three attribution segments to obtain the three segment results. Subsequently, it is found that the first attribution segment and the second attribution segment are adjacent in time and belong to the same object. If the identifiers are consistent and the continuity differences in energy efficiency changes and carbon footprint changes are both zero, then the first attribution segment and the second attribution segment are directly merged, while the third attribution segment is retained separately because it is separated from the second attribution segment by a unified time scale position. Finally, the two merged results corresponding to the chiller unit in the merged result table are summed again, and a result record is generated by combining the target action identifier, the time range from 10:00 to 11:45, and the spatial boundary, equipment ownership boundary, and accounting object boundary corresponding to the chiller unit, as the attributable emission reduction result corresponding to the target action.

[0022] Furthermore, the present invention also includes an AI-integrated energy efficiency and carbon footprint management system for zero-carbon industrial parks, the system comprising a mapping modeling module, a mirror generation module, a difference stripping module, a fragment attribution module, and a result merging module: The mapping modeling module is used to acquire real-time data of the park corresponding to the target action. The real-time data of the park includes the equipment status before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data. The module performs synchronous mapping of the real-time data of the park in the digital twin according to a unified time scale and a unified object identifier, and outputs the status before the action. The mirror generation module is used to take the state before the action as input, retain the state rewriting process corresponding to the target action in the digital twin to form an entity state sequence, and remove the state rewriting process corresponding to the target action to form a mirror state sequence while keeping the meteorological data, load data and boundary data unchanged. The difference stripping module is used to read the corresponding state items in the entity state sequence and the mirror state sequence at the same time position, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and remove the common external influences based on the same meteorological data, load data and boundary data in the corresponding state items, and output the difference sequence. The segment attribution module is used to compare the energy flow difference direction and the carbon flow difference direction of adjacent difference items in the difference sequence. When the two directions are consistent, the corresponding difference items are continuously connected. When either direction is reversed, the connection is cut off and the attribution segment is output. The result merging module is used to accumulate the energy flow difference and carbon flow difference in each attribution segment to obtain the energy efficiency change and carbon footprint change corresponding to each attribution segment. The energy efficiency change and carbon footprint change corresponding to each attribution segment are then merged in chronological order, and the attributable emission reduction results corresponding to the target action are output.

[0023] The working principle of this scheme can be summarized as follows: first, reconstruct the true state before the action; then, construct two comparable operational results: one with action and one without; finally, identify the emission reduction portion truly caused by the action. Specifically, first, acquire the equipment status, energy flow data, carbon flow data, meteorological data, load data, and boundary data before the target action occurs, and organize them into a pre-action state in a digital twin using a unified time scale and unified object identifier. Then, based on the pre-action state, on the one hand, retain the state rewriting process of the target action to form an entity state sequence; on the other hand, remove the state rewriting process of the target action to form a mirror state sequence, thereby obtaining a set of results for both actions performed and those not performed. The synchronous comparison results of the actions are then used to calculate the difference between the positions of the two state sequences at the same object and at the same time, and to eliminate synchronous changes caused by the same meteorological conditions, the same load conditions, and the same boundary conditions. On this basis, the continuous, consistent, and stable difference parts are connected into attribution segments. Finally, the energy flow difference and carbon flow difference corresponding to each attribution segment are accumulated and merged to output the attributable emission reduction results corresponding to the target action. In other words, this scheme does not simply look at the total change before and after the transformation, but specifically identifies how much energy efficiency change and carbon footprint change this action itself brings under the same time, the same boundary, and the same external conditions. For example, in a zero-carbon industrial park, the management adjusts the group control strategy for the energy station's chillers and wants to determine the actual emission reductions resulting from this adjustment. According to this solution, the system first organizes the chiller unit's operating status, power and cooling capacity data, carbon emission conversion data, current temperature, park load, and energy supply boundaries into a digital twin. Then, it generates two operating chains: one is the physical state sequence after operating according to the new group control strategy, and the other is a mirrored state sequence under the same weather, load, and boundaries without this adjustment. The two sequences are then compared moment-by-moment, removing common changes that would otherwise occur due to weather or tenant load variations, retaining only the differences directly caused by the group control strategy adjustment. Finally, the continuous and consistent portions of these differences are extracted and summed to obtain the actual energy savings and emission reductions resulting from the chiller group control strategy adjustment. This clearly demonstrates that the emission reduction results are due to the action itself, rather than being a coincidence caused by external environmental changes, whether used for energy-saving revenue sharing, service provider settlements, or low-carbon achievement applications.

[0024] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks, characterized by: include: S1. Obtain the real-time data of the park corresponding to the target action. The real-time data of the park includes the equipment status before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data. Perform synchronous mapping on the real-time data of the park in the digital twin according to a unified time scale and a unified object identifier, and output the status before the action. S2. Using the state before the action as input, retain the state rewriting process corresponding to the target action in the digital twin to form an entity state sequence, and remove the state rewriting process corresponding to the target action while keeping the meteorological data, load data and boundary data unchanged to form a mirror state sequence. S2 includes: S2-1. Read the equipment status, energy flow data, carbon flow data, meteorological data, load data and boundary data in the state before the action. In the digital twin, establish a succession relationship diagram according to the unified object identifier and unified time scale. Based on the target action corresponding to the target object, action period and action field, delineate the candidate rewrite edge set in the succession relationship diagram. At the same time, write the meteorological data, load data and boundary data into a fixed invariant item set. S2-2. For each candidate rewriting path in the candidate rewriting edge set, calculate the equipment state continuity cost, energy flow continuity cost, and carbon flow continuity cost between adjacent edges in the path. Under the condition that the fixed invariant itemset remains consistent item by item and the unified object identifier remains continuous before and after, sort the candidate rewriting paths by total cost from smallest to largest and retain the candidate rewriting path with the highest sorting result and consistent results in two consecutive rounds as the target rewriting path. S2-3. Replace the original continuation edges at the corresponding positions in the state before the action with the target rewriting path, and perform time-scaled deduction on the associated states before and after the action along the target rewriting path under the condition that the fixed itemset remains unchanged. Stop the deduction when the rewriting position set and the rewritten state value obtained in two adjacent rounds of deduction are consistent, and output the entity state sequence. S2-4. Delete the rewriting edges corresponding to the target rewriting path from the previous and subsequent continuity relationship graph, and reconnect the original continuity edges before and after the deletion position according to the unified object identifier and unified time scale under the condition that the fixed and unchanging itemset remains unchanged. Calculate the equipment state breakage, energy flow breakage, and carbon flow breakage for each mirror candidate path formed after reconnection. Sort the equipment state breakage, energy flow breakage, and carbon flow breakage in lexicographical order, select the mirror candidate path with the first sorting number as the mirror path, and output the mirror state sequence. S3. Read the corresponding state items in the entity state sequence and mirror state sequence at the same time position, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and remove the same external influences based on the same meteorological data, load data and boundary data in the corresponding state items, and output the difference sequence. S4. Compare the energy flow difference direction and carbon flow difference direction of adjacent difference terms in the difference sequence. When both directions are consistent, perform continuous continuation of the corresponding difference terms. When either direction is reversed, cut off the continuation relationship and output the attribution segment.

2. The AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks according to claim 1, characterized in that: Also includes: S5. Accumulate the energy flow difference and carbon flow difference in each attribution segment to obtain the energy efficiency change and carbon footprint change corresponding to each attribution segment. Then merge the energy efficiency change and carbon footprint change corresponding to each attribution segment in chronological order and output the attributable emission reduction results corresponding to the target action.

3. The AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks according to claim 2, characterized in that: S1 includes: S1-1. Obtain the original time values ​​and object identifiers corresponding to the equipment status, energy flow data, carbon flow data, meteorological data, load data and boundary data before the action. Perform reference time scale conversion on each original time value and rearrange them according to the unified time scale. Perform name merging and hierarchical expansion on each object identifier. Output the time scale alignment result and the identifier alignment result. S1-2. Read the time scale alignment result and the identifier alignment result. Write the device status before the action to the device status bit according to the unified time scale and the unified object identifier. Write the energy flow data to the energy flow bit, write the carbon flow data to the carbon flow bit, write the meteorological data to the meteorological bit, write the load data to the load bit, write the boundary data to the boundary bit, and output the object mapping table. S1-3. Perform a pre- and post-continuity check on the equipment status, energy flow data, carbon flow data, meteorological data, load data, and boundary data corresponding to the same unified object identifier under adjacent unified time scales in the object mapping table. If the continuation is successful, retain the corresponding record. If the continuation is unsuccessful, delete the broken record and reconnect the pre- and post-continuity records according to the unified object identifier, and output the status before the action.

4. The AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks according to claim 3, characterized in that: S3 includes: S3-1. Read the corresponding state items in the entity state sequence and mirror state sequence according to the unified time scale and unified object identifier, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and extract the meteorological data, load data and boundary data in the corresponding state items respectively, and output the difference pairs and impact item groups. S3-2. For meteorological data, load data and boundary data in the impact item group, compare the field values ​​and field positions of the entity state sequence and the mirror state sequence in the corresponding state item. When the field values ​​and field positions are consistent, the corresponding impact items are recorded as common source impact items, and they are merged into a common source impact item table according to the unified object identifier and unified time scale. S3-3. Based on the table of common source influence items, delete the synchronous change part corresponding to each common source influence item from the difference pair, retain the energy flow difference and carbon flow difference after deletion, and arrange them in a unified time scale order to form a difference sequence.

5. The AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks according to claim 4, characterized in that: S4 includes: S4-1. Read each difference item in the difference sequence according to the unified time scale order. Form a difference vector by combining the energy flow difference and carbon flow difference of each difference item. Perform symbol mapping on each difference vector to obtain the direction vector. Construct an adjacency matrix, directional inner product matrix and difference ratio matrix with the same unified object identifier and adjacent time scales as objects. The directional inner product matrix is ​​formed by summing the products of the preceding and following direction vectors. The difference ratio matrix is ​​composed of the ratio of the absolute values ​​of the preceding and following energy flow differences and the ratio of the absolute values ​​of the preceding and following carbon flow differences. When the directional inner product matrix is ​​two and both terms in the difference ratio matrix are written in the same rational interval, the corresponding preceding and following difference items are recorded as a continuable pair. When the directional inner product matrix is ​​not two or the two terms in the difference ratio matrix are written in different rational intervals, the corresponding preceding and following difference items are recorded as a cut-off pair. Output the initial continuation graph.

6. The AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks according to claim 5, characterized in that: S4 also includes: S4-2. Extract each candidate continuation chain based on the initial continuation graph. Construct an energy flow sequence matrix and a carbon flow sequence matrix for each candidate continuation chain. Perform first-order difference, second-order difference, and discrete convolution on the energy flow sequence matrix and the carbon flow sequence matrix respectively to obtain the energy flow change matrix and the carbon flow change matrix. Then calculate the Gram matrix, eigenvalue sequence, and sign-preserving sequence for the energy flow change matrix and the carbon flow change matrix respectively. When the eigenvalue sequence does not change the positive and negative order after being expanded according to the time scale and the sign-preserving sequence is continuous, retain the corresponding candidate continuation chain. When the positive and negative order of the eigenvalue sequence changes or the sign-preserving sequence is interrupted, cut off the candidate continuation chain at the corresponding position and output the corrected continuation chain. S4-3. Construct concatenation cost triplets for two adjacent concatenation chains before and after the time scale in the correction concatenation chain. The concatenation cost triplets include the difference between the cumulative energy flow value after concatenation and the cumulative energy flow value of the two chains before concatenation, the difference between the cumulative carbon flow value after concatenation and the cumulative carbon flow value of the two chains before concatenation, and the difference in encoding length between the sequence matrix after concatenation and the sequence matrix before concatenation. Sort the concatenation cost triplets in lexicographical order. For the concatenation result with the first sorting number, retain and recalculate the adjacent concatenation cost triplets. Repeat this process until there are no new concatenation pairs. Output the attribution fragment.

7. The AI-based integrated management method for energy efficiency and carbon footprint in zero-carbon industrial parks according to claim 6, characterized in that: S5 includes: S5-1. Read the energy flow difference sequence and carbon flow difference sequence in each attribution segment, perform time-scale integration and accumulation on the energy flow difference sequence in each attribution segment to obtain the segment energy efficiency change, perform time-scale integration and accumulation on the carbon flow difference sequence in each attribution segment to obtain the segment carbon footprint change, and generate a segment result table with a unified object identifier, segment start time scale and segment end time scale. S5-2. Based on the start and end timestamps of each attribution segment in the segment result table, establish the segment succession relationship according to time sequence, and calculate the succession difference of energy efficiency change and carbon footprint change for adjacent segments with consistent object identification. When both are zero, the corresponding segment results are directly merged; when there are non-zero values ​​in either case, the corresponding segment results are retained, and the merge result table is output. S5-3. Read each merge result in the merge result table according to the unified object identifier and time order. Perform a second accumulation on the energy efficiency change and carbon footprint change in each merge result. Write the second accumulation result into the result record corresponding to the target action identifier, time range and boundary data. Output the attributable emission reduction result corresponding to the target action.

8. An AI-based integrated energy efficiency and carbon footprint management system for zero-carbon industrial parks, used to implement the AI-based integrated energy efficiency and carbon footprint management method for zero-carbon industrial parks as described in any one of claims 1-7, the system comprising a mapping modeling module, a mirror generation module, a difference stripping module, a fragment attribution module, and a result merging module, characterized in that: The mapping modeling module is used to acquire the real-time data of the park corresponding to the target action. The real-time data of the park includes the equipment status before the action, energy flow data, carbon flow data, meteorological data, load data and boundary data. The module performs synchronous mapping on the real-time data of the park in the digital twin according to a unified time scale and a unified object identifier, and outputs the status before the action. The mirror generation module is used to take the state before the action as input, retain the state rewriting process corresponding to the target action in the digital twin to form an entity state sequence, and remove the state rewriting process corresponding to the target action to form a mirror state sequence while keeping the meteorological data, load data and boundary data unchanged. The difference stripping module is used to read the corresponding state items in the entity state sequence and the mirror state sequence at the same time position, calculate the energy flow difference and carbon flow difference in the corresponding state items respectively, and remove the common external influences based on the same meteorological data, load data and boundary data in the corresponding state items, and output the difference sequence. The segment attribution module is used to compare the energy flow difference direction and the carbon flow difference direction of adjacent difference items in the difference sequence. When the two directions are consistent, the corresponding difference items are continuously connected. When either direction is reversed, the connection is cut off and the attribution segment is output. The result merging module is used to accumulate the energy flow difference and carbon flow difference in each attribution segment to obtain the energy efficiency change and carbon footprint change corresponding to each attribution segment. The energy efficiency change and carbon footprint change corresponding to each attribution segment are then merged in chronological order, and the attributable emission reduction results corresponding to the target action are output.

Images