Household energy consumption portrait generation method based on deep learning and macro-microscopic perspective combination

CN121412450BActive Publication Date: 2026-09-08Hangzhou Gongshu District University of Technology Future Technology Research Institute
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202511533728.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-10-25
Publication Date
2026-09-08
Estimated Expiration
2045-10-25

AI Technical Summary

Technical Problem

[0005]其二为自下而上的设备/行为生成或数据驱动负荷合成,依赖通用启停规则或黑箱模型,难以显式嵌入设备可用性、禁止组合、端用途时段窗口与设定温度等约束,也缺乏与地区宏观边界的一致性校核

Benefits of technology

[0033] (1) This application constructs a hierarchical causal heterogeneous graph containing macroscopic, mesoscopic, microscopic, behavioral knobs, and end-use result nodes, and embeds monotonicity and threshold conditions to enforceably transmit macroscopic influences such as energy structure, price, climate, building type, heating method, and energy accessibility to the household level. The macroscopic statistical boundary formed according to a unified standard is mapped to a set temperature range, end-use time window, equipment availability and prohibited combinations, and end-use energy consumption ratio envelope, and assembled into behavioral constraint cards that correspond one-to-one with the household identifier vector. Unlike black-box generation that only allocates according to statistical proportions or does not explicitly constrain, the design clarifies the feasible domain before generation, avoids behavior and equipment path out-of-bounds, and enhances the interpretability and scenario adaptability of the profile.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121412450B_ABST
    Figure CN121412450B_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of household energy consumption portrait, and proposes a household energy consumption portrait generation method based on deep learning and macro-micro perspective combination; in order to solve the inconsistency between micro portrait and macro statistics and the lack of constraints in generation, a causal diagram with energy structure, price, climate and other macro nodes is constructed and the macro statistical boundary is established; along the causal path, the behavior constraint card corresponding to the household identification vector is assembled; the constraint parameter is parameterized as a gate signal to drive the end-use multi-flow generator, and the occupancy, end-use behavior and power sequence are generated in turn, and the end-use energy consumption and peak-valley characteristics are obtained through physical constraint mapping; according to the region aggregation, the macro boundary is compared, the correction instruction is generated based on the causal diagram deviation, and the final portrait containing the equipment list, behavior sequence and end-use energy consumption is output; the method realizes the generation in the feasible region, is consistent with the macro boundary after aggregation, and has explainability and physical consistency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of household energy consumption profiling technology, and in particular to a method for generating household energy consumption profiles based on a combination of deep learning and macro and micro perspectives. Background Technology

[0002] Urban household energy consumption is influenced by multiple factors, including energy structure, price, climate, building type, and heating methods, resulting in variations in end-uses such as kitchen, hot water, cooling, heating, lighting, and electrical outlets at different times. For management and assessment, it is necessary to generate interpretable household-level energy consumption profiles.

[0003] Existing technologies mostly adopt two types of paths:

[0004] One approach is to allocate or break down the total regional energy consumption based on the statistical yearbook from top to bottom, quickly obtaining the end-use proportion and load curve, but lacking fine-grained transmission and behavioral constraints on price, climate and building differences.

[0005] Secondly, the bottom-up generation of equipment / behaviors or data-driven load synthesis relies on general start-stop rules or black-box models, making it difficult to explicitly embed constraints such as equipment availability, prohibited combinations, end-use time windows and set temperatures, and also lacks consistency verification with regional macro boundaries.

[0006] The above-mentioned schemes generally suffer from three shortcomings: First, the macro-statistical boundaries cannot be effectively mapped to the household level, behavioral and equipment constraints are lacking, and the generated results are prone to exceeding the boundaries; second, the consistency between the aggregated micro-profiles and regional energy share and peak and valley periods cannot be guaranteed; and third, the lack of causally explainable bias attribution and card-level correction mechanisms makes it difficult to achieve reasonable allocation of end-use co-occurrence, sequence, and mutual exclusion relationships while maintaining the feasible domain.

[0007] Therefore, those skilled in the art urgently need a method for generating household energy consumption profiles; Summary of the Invention

[0008] One objective of this invention is to propose a method for generating household energy consumption profiles based on deep learning and a combination of macro and micro perspectives. In the scenario of generating urban household energy consumption profiles, this method translates factors such as climate, price, building, heating method, and energy accessibility into actionable behavioral constraints at the household level along causal paths under the constraints of regional macro statistical boundaries. The multi-stream generation at the gating end ensures that the results are always within the feasible region, and a closed loop is formed through aggregation comparison and causal attribution, so that the micro profile is consistent with the macro boundary after aggregation.

[0009] A method for generating household energy consumption profiles based on a combination of deep learning and macro / micro perspectives, according to an embodiment of the present invention, is characterized by comprising the following steps:

[0010] S1. Establish causal dependence based on regional energy structure, price, climate, building type, heating method, energy accessibility, income and family size, establish macro-statistical boundaries, define behavioral knob variables including set temperature, cooking frequency and time, hot water segmentation and lighting usage preference, and output hierarchical causal heterogeneity diagram and macro-statistical boundaries.

[0011] S2. Following the causal path, the macro-statistical boundary is refined into household-level device availability, prohibited combinations, set temperature range, end-use time window, and energy consumption ratio envelope. Based on household attributes, a household identification vector is generated, and a set of behavioral constraint cards corresponding one-to-one with the household identification vector is output.

[0012] S3. Parameterize the behavior constraint card into a decoded gating signal, and set up a time windower, device combination filter, linkage coupling module and physical constraint mapping layer in the end-use multi-stream generator so that the generation can only be expanded within the feasible domain, and output the gating end-use multi-stream generator.

[0013] S4. Based on the behavior constraint card set and the gated end-use multi-stream generator, the occupancy sequence, end-use behavior sequence and power time series are generated sequentially. The end-use energy consumption and peak-valley characteristics are obtained through the physical constraint mapping layer, and the initial household energy consumption profile is output.

[0014] S5. Based on the initial household energy consumption profile, aggregate the data by region and end-use to form macro results. Compare the macro results with the macro statistical boundary and attribute deviations to specific behavioral knobs and constraints based on the hierarchical causal heterogeneity diagram. Generate a set of correction instructions for behavioral constraint cards and output the set of correction instructions.

[0015] S6. Adjust the set temperature range, time window and energy consumption ratio envelope in the behavior constraint card according to the calibration instruction set, and perform re-decoding on the end-use multi-stream generator after gating to output the final household energy consumption profile.

[0016] S7. Generate a home-level profile based on the final home energy consumption profile, including a list of devices, behavior sequences, and end-use energy consumption.

[0017] Optionally, the hierarchical causal heterogeneous graph is specifically a scenario-based causal structure serving the generation of urban household energy consumption profiles. Its construction method is as follows: using regional energy structure, energy price, and climate as macro-level nodes; building type, heating method, and energy accessibility as meso-level nodes; income and household size as micro-level nodes; behavioral knob variables as behavioral knob variable nodes; and kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, lighting energy consumption, and socket load energy consumption as end-use result nodes. Based on energy consumption business rules, causal dependencies are established between nodes. Specifically, climate and building type are directed to set temperature and end-use result nodes to constrain heat load, and energy price is directed to set temperature and end-use result nodes. Mealtimes are constrained by time period selection, energy accessibility is linked to equipment availability, and end-use results are indirectly affected through behavioral knob variable nodes. Income and family size are linked to cooking frequency and hot water segmentation to constrain activity intensity, and heating methods are linked to heating-related end-use results to limit equipment paths. Business-oriented monotonic directions and threshold conditions are embedded in causal dependencies, where temperature is set to monotonically increase heating energy consumption and monotonically decrease cooling energy consumption, energy price is set to monotonically decrease electricity use during high-price periods, the corresponding equipment path is disabled when energy accessibility is lacking, and the heating end-use result is limited to zero when the heating method is not covered, thus translating macro-level effects into actionable behavioral constraints.

[0018] The establishment of macro-statistical boundaries specifically involves: using statistical yearbooks and industry standards to uniformly describe the regional total energy volume, the proportion of different energy carriers, energy price structure, household number and attribute distribution, climate indicators, energy accessibility coverage, and building type proportion, forming the allowable range and consistency relationship for each indicator; mapping the macro-statistical boundaries to the feasible range of behavioral knob variables, specifically by using climate and energy prices to jointly limit the upper and lower limits of set temperature, using energy price structure to limit the selectable windows for cooking and lighting times, using household size to limit the range of values ​​for cooking frequency and hot water segmentation, using energy accessibility to limit the available set of equipment and accordingly limit the effective combination of behavioral knob variables, and limiting the proportion of different energy carriers to the energy consumption proportion envelope of end-use results.

[0019] Optionally, the refinement of the macroscopic statistical boundary along the causal path specifically involves: based on the hierarchical causal heterogeneity graph and the macroscopic statistical boundary, passing constraints downward from the macroscopic nodes in a topological order, where energy accessibility and heating methods are mapped to equipment availability to obtain a list of available equipment; based on equipment availability, generating prohibited combinations according to equipment mutual exclusion relationships and energy accessibility deficiency conditions, and writing prohibited combinations into the behavior constraint card; based on the impact of climate, building type, and building age on heat load, limiting the set temperature range to a feasible interval between the upper limit of heating temperature and the lower limit of cooling temperature, and projecting the impact of energy price structure on high-price and low-price periods onto the end-use period window, so that the kitchen The selectable time periods for energy consumption, hot water consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption meet the constraints of macroeconomic price signals and activity patterns; the macroeconomic statistical boundaries of the proportion of energy carriers and the distribution of the number and attributes of households in a region are mapped to an energy consumption proportion envelope; upper and lower bounds are set for the energy consumption proportion of each end use or each energy carrier at the household level, so that the household profile is consistent with the regional distribution after aggregation; a household identifier vector is generated based on region, income percentile, household size, housing type, building age, heating method, and energy accessibility, and the above equipment availability, prohibited combinations, set temperature range, end use time window, and energy consumption proportion envelope are assembled into a behavioral constraint card using the household identifier vector as an index.

[0020] Optionally, the time window controller is a module that controls the start and stop of time steps based on the time window of the terminal application;

[0021] The device combination filter is a module that verifies the legality of device activation status based on device availability and prohibited combinations;

[0022] The linkage coupling module is a constraint module used to apply constraints on simultaneous occurrence, sequential order and mutual exclusion between end uses. The constraints are based on the allocation relationship between the end use time window and the energy consumption ratio envelope from the behavior constraint card.

[0023] The physical constraint mapping layer is a module that maps the set temperature range and end-use behavior to power limits, power variation range and duration limits;

[0024] The feasible domain is a set of states that simultaneously satisfy the end-use time window, device availability, prohibited combinations, set temperature range, and energy consumption percentage envelope.

[0025] The parameterization of the behavior constraint card into a decoded gating signal specifically involves: reading the end-use time period window from the behavior constraint card and encoding it into a time mask vector, which is then input to the time period windower; reading the device availability from the behavior constraint card and combining it with prohibited combinations to generate a device legality mask, which is then input to the device combination filter; reading the set temperature range from the behavior constraint card and converting it into end-use power upper limit, power change amplitude, and minimum duration parameters, which are then input to the physical constraint mapping layer; and reading the energy consumption ratio envelope from the behavior constraint card and converting it into end-use allocation coefficients, which are then input to the linkage coupling module and the terminal decoding output layer.

[0026] The end-use multi-stream generator includes a structure of multi-stream decoding sub-modules. Each end-use corresponds to one decoding sub-module, and during the decoding process, it receives gating in the following order: a time-segment windower blocks the propagation of the state during prohibited time periods and allows it during permitted time periods based on the time mask vector; a device combination filter applies a device validity mask to the activation state of candidate devices at each time step, retaining only the state that is consistent with the device availability and has not triggered prohibited combinations; a linkage coupling module constrains and allocates the co-occurrence, sequence, and mutual exclusion between end-uses based on the overlap relationship of the end-use time-segment windows and the allocation coefficient of the energy consumption ratio envelope, so that the multi-streams maintain a consistent allocation relationship in terms of time and intensity; a physical constraint mapping layer maps and truncates the decoded power and duration based on the power upper limit, power change amplitude, and shortest duration corresponding to the set temperature range, so that the output falls into the feasible region.

[0027] Optionally, the sequential generation of the occupancy sequence, end-use behavior sequence, and power time sequence specifically involves: processing each behavior constraint card in the set; the gated end-use multi-stream generator, under the control of the time period windower, reads the end-use time period window to generate an occupancy sequence covering the day and quarter, ensuring that the occupancy status only falls within the permitted time period; triggering end-use behavior based on the occupancy sequence; reading equipment availability and prohibited combinations in the equipment combination filter to eliminate non-compliant equipment activation states; and reading the energy consumption ratio envelope in the linkage coupling module to allocate and restrict the co-occurrence, sequence, and mutual exclusion relationships of multiple end-uses, outputting the power consumption of each end-use. The system first obtains the end-use behavior sequence of the route; then it sends the end-use behavior sequence to the physical constraint mapping layer, reads the set temperature range and maps it to the power upper limit, power change amplitude and shortest duration, decodes the end-use behavior sequence to obtain the power time series aligned with the time step; based on the power time series, it accumulates by end use to obtain end-use energy consumption, and extracts peak power, peak occurrence period, valley power and valley occurrence period in the time dimension, calculates the peak-valley ratio to form peak-valley features; finally, it summarizes the occupancy sequence, end-use behavior sequence, power time series, end-use energy consumption and peak-valley features into an initial household energy consumption profile.

[0028] Optionally, the aggregation by region and end-use to form a macro result, and the comparison with the macro statistical boundary, specifically involves: reading the power time series and end-use energy consumption, grouping households by region, and accumulating by end-use in the time dimension to obtain regional end-use energy consumption and peak-valley characteristics; converting the regional end-use energy consumption into end-use proportion, comparing it with the corresponding proportion interval in the macro statistical boundary for each end-use, forming the difference value and interval boundary marker for each region, and checking the consistency of peak power and valley power in the time period to obtain the difference set between the macro result and the macro statistical boundary;

[0029] The method of attributing deviations to specific behavior knobs and constraints based on hierarchical causal heterogeneity graphs is as follows: Based on the difference set, determine the macroscopic node where the deviation occurs; enumerate the reachable paths from the macroscopic node to the fields within the behavior constraint card along the directed edges of the hierarchical causal heterogeneity graph; determine the allocation ratio according to the end-use and time range covered by the path; and allocate each deviation amount to the candidate adjustment amount of the corresponding behavior knob; under the premise that equipment availability and prohibited combinations are not violated, priority is given to allocating to the energy consumption ratio envelope and end-use time window; when the interval and time period have reached the upper or lower limit, allocation is then made to the set temperature range to adjust the power upper limit and continuous constraints; for deviations across end-uses, use the constraint relationships between end-uses in the hierarchical causal heterogeneity graph to map the allocation ratio to the behavior knob of the corresponding end-use according to co-occurrence, sequence, and mutual exclusion relationships, thus obtaining a set of executable adjustment amounts for the card level;

[0030] The process of generating a calibration instruction set for behavior constraint cards involves: locating the target card in the behavior constraint card set using the family identifier vector as an index; discretizing the executable adjustment quantities into instruction entries, with each instruction including at least the target field, adjustment direction and magnitude, applicable time range, and priority; for conflicts among multiple instructions within the same behavior constraint card, consistency is determined based on priority and prohibited combinations, retaining only combinations that do not compromise device availability; and finally, the final instruction entries are archived by region and terminal purpose to form a calibration instruction set for behavior constraint cards.

[0031] Optionally, the specific steps for adjusting the set temperature range, end-use time window, and energy consumption ratio envelope in the behavior constraint card are as follows: The target card is located in the behavior constraint card set using the household identifier vector; each correction instruction entry is read and adjusted within the specified end-use and time range; for the energy consumption ratio envelope, the target coefficient is increased or decreased within the lower and upper bounds according to the instructions, and the allocation coefficient is updated to ensure that the end-use allocation ratio is consistent with the energy consumption ratio envelope within the same time step; for the end-use time window, the coverage ratio of the allowed time period is modified according to the instructions, and a new time mask is generated, ensuring that the occupancy sequence and end-use behavior triggering only fall within the adjusted allowed time period; for the set temperature range, the upper and lower bounds of the temperature are tightened or loosened according to the instructions, and synchronously mapped to power upper limit, power change amplitude, and minimum duration parameters in the physical constraint mapping layer.

[0032] The beneficial effects of this invention are:

[0033] (1) This application constructs a hierarchical causal heterogeneous graph containing macroscopic, mesoscopic, microscopic, behavioral knobs, and end-use result nodes, and embeds monotonicity and threshold conditions to enforceably transmit macroscopic influences such as energy structure, price, climate, building type, heating method, and energy accessibility to the household level. The macroscopic statistical boundary formed according to a unified standard is mapped to a set temperature range, end-use time window, equipment availability and prohibited combinations, and end-use energy consumption ratio envelope, and assembled into behavioral constraint cards that correspond one-to-one with the household identifier vector. Unlike black-box generation that only allocates according to statistical proportions or does not explicitly constrain, the design clarifies the feasible domain before generation, avoids behavior and equipment path out-of-bounds, and enhances the interpretability and scenario adaptability of the profile.

[0034] (2) In the multi-stream generation stage of end-use applications, the behavior constraint card is parameterized into a decoded gating signal, which is then gated sequentially by the time period windower, the device combination filter, the linkage coupling module, and the physical constraint mapping layer. This ensures that the entire process of occupancy, end-use behavior, and power time series generation is constrained and maintains a consistent allocation relationship. Device availability and prohibition combinations ensure the legitimacy of device activation. The end-use time period window and energy consumption ratio envelope achieve a reasonable allocation of co-occurrence, sequence, and mutual exclusion among multiple streams. The physical constraint mapping layer converts the set temperature range into power upper limit, variation range, and minimum duration, suppressing unreasonable start-stop and load jumps. Compared to post-correction or synthesis lacking a mechanism, gating generation ensures that the output naturally falls within the feasible region, and end-use interaction and physical behavior are consistently expressed.

[0035] (3) In the aggregation and correction stage, the initial profiles are aggregated according to region and end-use, compared with the macro statistical boundaries, and the deviations are attributed to specific behavioral knobs and constraints based on the cause-effect graph. A correction instruction set for behavioral constraint cards is generated and then decoded. This closed-loop mechanism prioritizes adjusting the energy consumption ratio envelope and end-use time window without compromising equipment availability and prohibited combinations. If necessary, it tightens or relaxes the set temperature range, so that the household profile is consistent with the regional energy carrier ratio and peak-valley characteristics after aggregation. Compared with existing schemes that lack consistency verification and interpretable correction, this application achieves effective coupling between macro boundaries and micro generation, ensuring regional consistency of the profile and the implementation of behavioral constraints, and meeting the scenario-based needs of urban household energy consumption profile generation. Attached Figure Description

[0036] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0037] Figure 1 This is a flowchart of a method for generating household energy consumption profiles based on a combination of deep learning and macro / micro perspectives proposed in this invention.

[0038] Figure 2 This is a flowchart of a parameterized gating and multi-stream generator for a household energy consumption profile generation method based on deep learning and a combination of macro and micro perspectives proposed in this invention.

[0039] Figure 3 This is a flowchart illustrating the aggregation, comparison, and causal attribution process of a household energy consumption profile generation method based on deep learning and a combination of macro and micro perspectives proposed in this invention. Detailed Implementation

[0040] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0041] refer to Figures 1 to 3 A method for generating household energy consumption profiles based on a combination of deep learning and macro / micro perspectives includes:

[0042] S1. Establish causal dependence based on regional energy structure, price, climate, building type, heating method, energy accessibility, income and family size, establish macro-statistical boundaries, define behavioral knob variables including set temperature, cooking frequency and time, hot water segmentation and lighting usage preference, and output hierarchical causal heterogeneity diagram and macro-statistical boundaries.

[0043] S2. Following the causal path, the macro-statistical boundary is refined into household-level device availability, prohibited combinations, set temperature range, end-use time window, and energy consumption ratio envelope. Based on household attributes, a household identification vector is generated, and a set of behavioral constraint cards corresponding one-to-one with the household identification vector is output.

[0044] S3. Parameterize the behavior constraint card into a decoded gating signal, and set up a time windower, device combination filter, linkage coupling module and physical constraint mapping layer in the end-use multi-stream generator so that the generation can only be expanded within the feasible domain, and output the gating end-use multi-stream generator.

[0045] S4. Based on the behavior constraint card set and the gated end-use multi-stream generator, the occupancy sequence, end-use behavior sequence and power time series are generated sequentially. The end-use energy consumption and peak-valley characteristics are obtained through the physical constraint mapping layer, and the initial household energy consumption profile is output.

[0046] S5. Based on the initial household energy consumption profile, aggregate the data by region and end-use to form macro results. Compare the macro results with the macro statistical boundary and attribute deviations to specific behavioral knobs and constraints based on the hierarchical causal heterogeneity diagram. Generate a set of correction instructions for behavioral constraint cards and output the set of correction instructions.

[0047] S6. Adjust the set temperature range, time window and energy consumption ratio envelope in the behavior constraint card according to the calibration instruction set, and perform re-decoding on the end-use multi-stream generator after gating to output the final household energy consumption profile.

[0048] S7. Generate a home-level profile based on the final home energy consumption profile, including a list of devices, behavior sequences, and end-use energy consumption.

[0049] In this embodiment, the hierarchical causal heterogeneous graph is specifically a scenario-based causal structure serving the generation of urban household energy consumption profiles. Its construction method is as follows: regional energy structure, energy price, and climate are macro-level nodes; building type, heating method, and energy accessibility are meso-level nodes; income and household size are micro-level nodes; behavioral knob variables are behavioral knob variable nodes; and kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, lighting energy consumption, and socket load energy consumption are end-use result nodes. Based on energy consumption business rules, causal dependencies are established between nodes. Specifically, climate and building type are directed to the set temperature and end-use result nodes to constrain heat load, and energy price is directed to the set temperature and end-use result nodes. The cooking time period is constrained by the time period selection, and energy accessibility is linked to equipment availability. This indirectly affects the end-use results through behavioral knob variable nodes. Income and family size are linked to cooking frequency and hot water segmentation to constrain activity intensity, and heating method is linked to heating-related end-use results to limit equipment paths. Business-oriented monotonic directions and threshold conditions are embedded in the causal dependencies. Temperature is set to monotonically increase heating energy consumption and monotonically decrease cooling energy consumption. Energy price is set to monotonically decrease electricity use during high-price periods. When energy accessibility is lacking, the corresponding equipment path is disabled. When the heating method is not covered, the heating end-use result is limited to zero, thus translating macro-level influences into actionable behavioral constraints.

[0050] The establishment of macro-statistical boundaries specifically involves: using statistical yearbooks and industry standards to uniformly describe the regional total energy volume, the proportion of energy carriers, energy price structure, household number and attribute distribution, climate indicators, energy accessibility coverage, and building type proportion, forming the allowable range and consistency relationship of each indicator; mapping the macro-statistical boundaries to the feasible range of behavioral knob variables, specifically by using climate and energy prices to jointly limit the upper and lower limits of set temperature, using energy price structure to limit the selectable windows of cooking and lighting times, using household size to limit the range of values ​​for cooking frequency and hot water segmentation, using energy accessibility to limit the available set of equipment and accordingly limit the effective combination of behavioral knob variables, and limiting the proportion of energy carriers to the energy consumption proportion envelope of end-use results;

[0051] Specifically, in the implementation method, a hierarchical causal heterogeneity diagram is constructed and a macro-statistical boundary is established by taking regional energy structure, energy price structure, climate indicators, building type and age ratio, heating method, energy accessibility coverage, number of households and attribute distribution as inputs.

[0052] First, establish a node hierarchy:

[0053] Macro nodes are used to represent the region's total energy consumption, the proportion of different energy carriers, the energy price structure, climate indicators, energy accessibility coverage, and the proportion of different building types.

[0054] Mesoscopic nodes are used to represent building types and heating methods and their impact on heat load paths;

[0055] Micro-nodes are used to represent income and family size;

[0056] The behavior knob variable node is used to carry settings such as temperature, cooking frequency and time period, hot water segmentation, and lighting usage preferences.

[0057] The terminal application result node is used to carry energy for kitchen use, hot water use, cooling use, heating use, lighting use, and socket loads.

[0058] Secondly, directed edges are established between nodes according to business rules, allowing the influence of macro and meso nodes on behavioral control variable nodes and end-use result nodes to be tracked. For example, energy price structure points to end-use time windows and set temperatures to limit usage during high-price periods; climate indicators and building types point to set temperatures to limit feasible heating and cooling ranges; energy accessibility coverage points to equipment availability and restricts end-use paths through prohibited combinations; income and household size point to cooking frequency and hot water segmentation to constrain activity intensity. Finally, macro inputs are standardized into executable macro statistical boundaries, and the outputs are used to push down to the legal ranges at the household level in subsequent steps.

[0059] To clarify the generation of graph structure and boundaries, the hierarchical causal heterogeneous graph and macroscopic statistical boundaries are defined as follows:

[0060] ;

[0061] ;

[0062] ;

[0063] in: This is a hierarchical causal heterogeneity graph; A set of nodes; This is a set of directed edges used to represent the direction of influence; It is a set of macro nodes, including regional total energy, the proportion of energy carriers, energy price structure, climate indicators, energy accessibility coverage and building type proportion, etc. This is a set of meso-level nodes, including nodes related to building type and heating method; It is a set of micro-nodes, including nodes related to income and family size; This is a set of behavior knob variable nodes, including nodes for set temperature, cooking frequency and time period, hot water segmentation, and lighting usage preference. This is a set of end-use result nodes, including kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, lighting energy consumption, and socket load energy consumption nodes; For the set of macro-statistical boundaries, the constraint intervals are given item by item according to the end-use index; For end-use indexes; This is a collection of end-use indexes, including kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, lighting energy consumption, and socket load energy consumption; The upper and lower bounds of the proportion of end-use are used to limit the allocation of energy carriers and end-uses at the regional level; To set the upper and lower bounds of the temperature range, with the dimension of temperature, it is used to limit the values ​​of the knob variables related to heating and cooling. This is the upper and lower bounds of the coverage ratio for end-use time periods, ranging from zero to one, used to limit the allowable coverage ratio during high-price and low-price periods.

[0064] In implementation, macro-level data is first collected and standardized, then written into... Then, according to the business rules, , and arrive and Establish a set of directed edges between them Subsequently, driven by energy price structure, climate indicators, building type, and heating method, the usage at each end was calculated and registered. and And driven by the proportion of energy carriers and the number and attributes of households, registration Final output and As a result, S2 will refine the macroscopic statistical boundary along the causal path into direct inputs for equipment availability, prohibited combinations, set temperature range, end-use time window, and energy consumption ratio envelope.

[0065] In this embodiment, refining the macroscopic statistical boundary along the causal path specifically involves: based on the hierarchical causal heterogeneous graph output by S1 and the macroscopic statistical boundary, constraints are passed down from the macroscopic nodes in topological order, where energy accessibility and heating methods are mapped to equipment availability to obtain a list of available equipment; based on equipment availability, prohibited combinations are generated according to the mutual exclusion relationship between equipment and the lack of energy accessibility conditions, and these prohibited combinations are written into the behavior constraint card; based on the influence of climate, building type, and building age on heat load, the set temperature range is limited to a feasible interval between the upper limit of heating temperature and the lower limit of cooling temperature, and the influence of energy price structure on high-price and low-price periods is projected onto the end-use period window. The system ensures that the selectable time periods for kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption meet the constraints of macroeconomic price signals and activity patterns. It maps the macroeconomic statistical boundaries of the proportion of energy carriers and the distribution of the number and attributes of households in a region to an energy consumption proportion envelope. At the household level, it sets upper and lower bounds for the energy consumption proportion of each end use or each energy carrier, so that the household profile is consistent with the regional distribution after aggregation. It generates a household identifier vector based on region, income percentile, household size, housing type, building year, heating method, and energy accessibility. Using the household identifier vector as an index, it assembles the above-mentioned equipment availability, prohibited combinations, set temperature range, end use time window, and energy consumption proportion envelope into a behavior constraint card.

[0066] Specifically, in the implementation method, based on the hierarchical causal heterogeneity graph and macroscopic statistical boundary output in step S1, the macroscopic statistical boundary is refined along the causal path into executable equipment availability, prohibited combinations, set temperature range, end-use time window, and energy consumption ratio envelope at the household level, and assembled into a set of behavioral constraint cards. Specifically, the set of households within the region is first located using the regional index, and equipment availability is mapped according to energy accessibility and heating method, so that households with only the energy and equipment paths required for the end-use can use it for that end-use; prohibited combinations are generated under the mutual exclusion relationship of equipment compatibility and energy path, recording equipment or settings that cannot be simultaneously established; then, the set temperature range is limited to the feasible interval between the upper limit of heating temperature and the lower limit of cooling temperature by climate indicators and building type, and the energy price structure and activity patterns are projected as the end-use time window, so that allowed time periods and prohibited time periods can be distinguished and coded; then, a household identifier vector is constructed, which at least includes region, income percentile, household size, housing type, building year, heating method, and energy accessibility, and is used to index households and assemble behavioral constraint cards.

[0067] To ensure consistency with macro-statistical boundaries, the energy consumption share envelope for each household is calculated at the end-use level, using the following standardized allocation model:

[0068] ;

[0069] in: For family indexing; For end-use indexes; For regional indexing; For the region Family collection within; For family End-user applications The lower and upper bounds of the energy consumption ratio envelope (dimensionless). and For use in the middle range of macroeconomic statistical boundaries The lower and upper bounds of the proportion (dimensionless). For device availability indication, representing the household Does it have terminal applications? The required equipment and energy are available; The prohibition of combination compliance instructions indicates the use of the product under the already effective prohibition of combination constraints. Is it feasible? The percentage of time windows for end-use indicates the household usage. End-user applications The available time is the percentage of the maximum reference time for the region where this device is used; To set the temperature range factor, representing the home End-user applications The set temperature allowable span is the proportion of the regional reference span; for non-thermal applications, take... ; This is a coefficient based on family size, expressed as a ratio of family size to the regional average family size.

[0070] The above allocation is normalized for all households in the same region according to feasibility weights to ensure that the aggregated data is consistent with the macro-statistical boundaries.

[0071] During implementation, four basic quantities are first established at the home level for each end-user application: device availability indicator. Prohibition of combining compliance instructions Terminal usage time window ratio With set temperature range factor Simultaneous calculation of family size coefficient And register the region to which the family belongs. Then calculate according to the formula. The energy consumption percentage envelope field of the behavior constraint card is written into it; at the same time, the device availability, prohibited combinations, set temperature range, and end-use time window are written into the corresponding fields as is, completing the behavior constraint card assembly. Finally, a set of behavior constraint cards corresponding one-to-one with each household identifier vector is output, which serves as the direct source for parameterizing the behavior constraint cards into decoded gate control signals in step S3.

[0072] In this embodiment, the time-segment windower is a module that controls the start and stop of time steps based on the end-use time-segment window; the device combination filter is a module that filters the legality of device activation status based on device availability and prohibited combinations; the linkage coupling module is a module used to apply constraints on simultaneous occurrence, sequential order, and mutual exclusion relationships between end-uses, and the constraints are based on the allocation relationship between the end-use time-segment window and the energy consumption ratio envelope from the behavior constraint card; the physical constraint mapping layer is a module that maps the set temperature range and end-use behavior to power upper limit, power change amplitude, and duration limit; the feasible domain is the domain that simultaneously satisfies the end-use time-segment window and device availability. The set of states including usage, prohibited combinations, set temperature range, and energy consumption ratio envelope; the parameterization of the behavior constraint card into a decoded gating signal specifically involves: reading the end-use time period window from the behavior constraint card and encoding it into a time mask vector, which is then input to the time period windower; reading the device availability from the behavior constraint card and combining it with prohibited combinations to generate a device legality mask, which is then input to the device combination filter; reading the set temperature range from the behavior constraint card and converting it into end-use power upper limit, power change amplitude, and minimum duration parameters, which are then input to the physical constraint mapping layer; and reading the energy consumption ratio envelope from the behavior constraint card and converting it into end-use allocation coefficients, which are then input to the linkage coupling module and the terminal decoding output layer.

[0073] The end-use multi-stream generator has a structure that includes a multi-stream decoding submodule. Each end-use corresponds to one decoding submodule, and during the decoding process, it receives gating in the following order: the time windower blocks the propagation of the state during the prohibited time period based on the time mask vector and allows it during the allowed time period; the device combination filter applies a device validity mask to the activation state of candidate devices at each time step, retaining only the state that is consistent with the device availability and has not triggered the prohibited combination; the linkage coupling module constrains and allocates the co-occurrence, sequence, and mutual exclusion between each end-use according to the overlap relationship of the end-use time window and the allocation coefficient of the energy consumption ratio envelope, so that the multi-stream maintains a consistent allocation relationship in terms of time and intensity; the physical constraint mapping layer maps and truncates the decoded power and duration with the power upper limit, power change amplitude, and shortest duration corresponding to the set temperature range, so that the output falls into the feasible region.

[0074] Specifically, in this implementation, the behavioral constraint card set output in step S2 is parameterized into a decoded gating signal and assembled into a time windower, a device combination filter, a linkage coupling module, and a physical constraint mapping layer to form a gating end-use multi-stream generator. (In the home index) End-use index set (Including kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption), first, the usage time window is read from the behavior constraint card to generate a time mask. Its meaning is in the time index set inside, when When it is within the permitted time period ,otherwise Then read device availability. Compliance instructions that prohibit combination Generate a device validity mask This is used to filter out invalid device activation states at each time step; then it is used to encapsulate the energy consumption percentage. Select target coefficient And within the linkage coupling module, based on the active end usage set... Perform normalization allocation; finally, set the temperature range factor. Mapped to physical constraint parameters, including power limits Power variation range With the shortest duration These three elements are used to implement upper bound and slope control on the strength and duration of end-use behaviors in the physical constraint mapping layer.

[0075] To clarify the key quantities for gating, the active end usage set and allocation coefficients are defined as follows:

[0076] ;

[0077] ;

[0078] in: For family indexing; For end-use indexes; For time indexing; For end-use index set; A collection of time indexes; To occupy the sequence strength (zero to one), it is generated over time in the end-use multistream generator and works in conjunction with the gating signal; The time mask (zero or one) is obtained by encoding the end-use time window. The device validity mask (zero or one) is determined by the device availability. Compliance instructions that prohibit combination Multiplying them together yields the result. The active end usage set is determined when the occupancy, time, and device legality gates are simultaneously satisfied. The lower and upper bounds of the energy consumption ratio envelope (dimensionless) are derived from the canonical allocation in step S2; The target coefficient (dimensionless) for the energy consumption ratio envelope is taken from... ; The allocation factor (from zero to one) is used to normalize the intensity of active end applications within the same time step; This is the upper limit of power (in units of power). The power change amplitude (power per time). It represents the shortest duration (in units of time).

[0079] The assembly sequence of the gate control is as follows:

[0080] Firstly, Input a time-period windower so that end-use candidate states only propagate during permitted time periods;

[0081] Second, Input device combination filter, which performs legality screening based on device availability and prohibited combinations;

[0082] Thirdly The input linkage coupling module allocates and constrains the strength of each end application under co-occurrence, sequential, and mutual exclusion relationships, ensuring that the end applications at the same time step maintain consistent allocation rules; fourthly, it will... , and The input physical constraint mapping layer controls the power upper limit, slope projection, and minimum duration of the decoded intensity and duration. These gating signals are applied in parallel on the multi-stream decoding submodule according to end-use applications, forming a gated end-use multi-stream generator. Its output is expanded only within the feasible region (simultaneously satisfying the end-use time window, device availability, prohibited combinations, set temperature range, and energy consumption percentage envelope), providing a consistent gating basis for generating the occupancy sequence, end-use behavior sequence, and power time series in step S4.

[0083] In this embodiment, the sequential generation of the occupancy sequence, end-use behavior sequence, and power time sequence specifically involves: processing each behavior constraint card in the set; the gated end-use multi-stream generator, under the control of the time period windower, reads the end-use time period window to generate an occupancy sequence covering the day and quarter, ensuring that the occupancy status only falls within the permitted time period; triggering end-use behavior based on the occupancy sequence; reading equipment availability and prohibited combinations in the equipment combination filter to eliminate non-compliant equipment activation states; and reading the energy consumption percentage envelope in the linkage coupling module to allocate and restrict the co-occurrence, sequence, and mutual exclusion relationships of multiple end-uses, and outputting the power time sequence for each end-use. The end-use behavior sequence is generated for each application. This sequence is then fed into a physical constraint mapping layer, where a set temperature range is read and mapped to a power upper limit, power variation amplitude, and minimum duration. The end-use behavior sequence is then decoded to obtain a power time series aligned with the time step. Based on the power time series, end-use energy consumption is accumulated by application. Peak power, peak occurrence period, valley power, and valley occurrence period are extracted in the time dimension, and the peak-valley ratio is calculated to form peak-valley characteristics. The occupancy sequence, end-use behavior sequence, power time series, end-use energy consumption, and peak-valley characteristics are then combined to form an initial household energy consumption profile.

[0084] Specifically, in this implementation, the set of behavioral constraint cards output in step S2 is used as input, and occupancy sequences, end-use behavior sequences, and power time sequences are generated sequentially in the gated end-use multi-flow generator. Then, end-use energy consumption and peak-valley characteristics are calculated, and these are summarized to form an initial household energy consumption profile. For each household index... End-use index set (Including kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption), the end-use multi-stream generator after gating first reads the end-use time period window and expands it into a time mask. Read device availability Compliance instructions that prohibit combination And generate a device validity mask. Read the target coefficient of energy consumption ratio envelope. The allocation coefficient is calculated by combining the active end usage set. At the same time, the temperature range factor will be set. Mapped to power limit Power variation range With the shortest duration The above-mentioned gating quantity and the occupancy sequence generated internally by the end-use multi-stream generator. End-use behavior sequence strength Together they drive subsequent decoding.

[0085] The core calculation of power time series adopts the form of superimposing interval projection and shortest duration constraint:

[0086] ;

[0087] in: For family indexing; For end-use indexes; For time indexing; For family End-user applications time Power time series values ​​(power dimensions), initial value ; For interval projection operators, the input values ​​are restricted to a closed interval. Inside; The power variation amplitude (power per time dimension) is determined by a set temperature range factor. Mapped to obtain; The upper limit of power (in terms of power dimensions) is determined by the set temperature range factor. Mapped to obtain; Shortest duration operator The intensity of the post-activation end-use behavior sequence (zero to one) when the length of the consecutive activation segment is shorter than Extend it at that time; The shortest duration (time dimension). The occupied sequence strength (zero to one) is output by the gated end-use multistream generator; The time mask (zero or one) is obtained by expanding the end-use time window; The device validity mask (zero or one) is determined by the device availability. Compliance instructions that prohibit combination Multiplying them together yields the result. The allocation coefficient (from zero to one) is the target coefficient enveloped by the energy consumption ratio. The results are obtained by normalizing the set of active endpoints at the same time step. The lower and upper bounds of the energy consumption ratio envelope (dimensionless) are derived from step S2.

[0088] In obtaining Then, according to the time step length The energy consumption of the end-use applications is accumulated, and the total power is aggregated:

[0089] ;

[0090] ;

[0091] in: For family End-user applications Energy consumption (in the dimension of energy). The time step (time dimension). For family In time Total power.

[0092] based on Extracting peak and valley features:

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] in: Peak power (in units of power). This refers to the period during which the peak occurs; Valley power (power dimension); This refers to the period during which the trough value occurs; Peak-to-valley ratio (dimensionless); It is the stability constant.

[0099] The execution process is as follows: the behavior constraint card set is processed card by card, and the end-use multi-stream generator after gating first outputs the occupancy sequence. Then, under the combined effect of the time windower and the device filter, the end-use behavior sequence is generated. Subsequently, in the linkage coupling module according to To implement co-occurrence, sequence, and mutual exclusion relationships, the final step is to determine the physical constraint mapping layer based on... , and calculate Then, the results are accumulated according to the end-user's purpose. and from extract , , , and Together with the occupancy sequence, end-use behavior sequence, power time series, and end-use energy consumption, the data is summarized and output to form an initial profile of household energy consumption.

[0100] In this embodiment, the aggregation by region and end-use to form a macro result, and the comparison with the macro statistical boundary of S1, specifically involves: reading the power time series and end-use energy consumption from the output of S4, grouping households by region, and accumulating them by end-use in the time dimension to obtain regional end-use energy consumption and peak-valley characteristics; converting the regional end-use energy consumption into end-use proportion, comparing it with the corresponding proportion interval in the macro statistical boundary for end-use, forming the difference value and interval boundary marker for each region, and checking the consistency of peak power and valley power in the time period to obtain the difference set between the macro result and the macro statistical boundary.

[0101] The method of attributing deviations to specific behavior knobs and constraints based on hierarchical causal heterogeneity graphs is as follows: Based on the difference set, determine the macroscopic node where the deviation occurs; enumerate the reachable paths from the macroscopic node to the fields within the behavior constraint card along the directed edges of the hierarchical causal heterogeneity graph; determine the allocation ratio according to the end-use and time range covered by the path; and allocate each deviation amount to the candidate adjustment amount of the corresponding behavior knob; provided that equipment availability and prohibited combinations are not violated, priority is given to allocating to the energy consumption ratio envelope and end-use time window; when the interval and time period have reached the upper or lower limit, allocation is then made to the set temperature range to adjust the power upper limit and continuous constraints; for deviations across end-uses, use the constraint relationships between end-uses in the hierarchical causal heterogeneity graph to map the allocation ratio to the corresponding end-use behavior knob according to co-occurrence, sequence, and mutual exclusion relationships, thus obtaining a set of executable adjustment amounts for the card level.

[0102] The process of generating a calibration instruction set for behavior constraint cards involves: locating the target card in the behavior constraint card set using the family identifier vector as an index; discretizing the executable adjustment quantities into instruction entries, with each instruction including at least the target field, adjustment direction and magnitude, applicable time range, and priority; for conflicts among multiple instructions within the same behavior constraint card, consistency is determined based on priority and prohibited combinations, retaining only combinations that do not compromise device availability; and finally, the final instruction entries are archived by region and terminal purpose to form a calibration instruction set for behavior constraint cards.

[0103] Specifically, in this implementation, the end-use energy consumption and power time series output in step S4 are used as input. First, the data is aggregated by region and end-use to form a macroscopic result. Then, the consistency is checked with the macroscopic statistical boundary established in step S1. Subsequently, the deviation is attributed to specific behavior knobs and constraint items along the hierarchical causal heterogeneity diagram to generate a set of correction instructions for behavior constraint cards.

[0104] Specifically: in the regional index Below, family group End-use index set The scope includes (kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption). First, calculate the regional usage ratio. and the macro-statistical boundary The difference set is obtained by comparison; then, from the macroscopic node along the directed edge to the behavior constraint card field in the hierarchical causal heterogeneous graph, the deviation is assigned according to the end use and time range covered by the path, and under the premise that the equipment availability and prohibition combination are not destroyed, it is sequentially assigned to the energy consumption ratio envelope, the end use time window and the set temperature range, and finally the correction instruction entries are generated and archived for output.

[0105] To distinguish between aggregation and difference detection, the regional usage ratio is defined as follows:

[0106] ;

[0107] in: For regional indexing; For end-use indexes; For the region Family collection within; For end-use index set; For the family obtained in step S4 End-user applications Energy consumption (in terms of energy units). Based on this, the upper and lower limits are obtained. , Deviation With amplitude ;in , and For terminal use The lower and upper bounds (dimensionless) of the proportion within the macro-statistical boundary.

[0108] The core assignment calculation for deviation from the behavioral constraint card is as follows:

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] in: For family indexing; For end-use indexes; The correction amounts allocated to the behavior constraint card correspond to the adjustment amounts of the energy consumption ratio envelope target coefficient, the end-use time period window ratio, and the set temperature range factor (dimensionless, positive and negative indicate direction). It is a symbolic function; For feasible domain gating, For equipment availability indication, To prohibit the combination of compliance instructions; For truncation operators. Assigning weights along the hierarchical causal heterogeneity graph down to the card level, where For end-user use, the proportion of the time window. To set the temperature range factor, This represents the family size coefficient. This is the adjustable margin of the energy consumption ratio envelope; The target coefficient for the energy consumption ratio envelope. The lower and upper bounds of the energy consumption ratio envelope obtained in step S2 are: This represents the adjustable margin for the time window; Adjustable margin for the set temperature range; For indicator functions; , , The allocation ratio is based on the adjustable margin. As the normalization factor, It is the stability constant.

[0114] The above order reflects the rule of prioritizing allocation to the energy consumption percentage envelope and end-use time window, with the remaining amount then allocated to a set temperature range, and by... Ensure that equipment availability is not compromised and prohibit any combinations thereof.

[0115] During implementation, the execution process is as follows:

[0116] First, by region Terminal uses From step S4 calculate And obtained and ;

[0117] Second, locate regions in the hierarchical causal heterogeneity diagram. family collection Terminal Applications The corresponding behavior constraint card fields are calculated. And according to the current , , get , , and , , ;

[0118] Third, obtain the result using the above formula. This is interpreted as the adjustment amount of the card level, which is the "increase or decrease in magnitude and direction";

[0119] Fourth Write the adjustment instruction as the energy consumption ratio envelope target coefficient, and... Write an instruction to adjust the coverage ratio of the time window for terminal use. Write the adjustment command to set the temperature range factor, and include the target field, applicable time range, and priority;

[0120] Fifth, for the same card instruction, consistent decisions are made based on priority and prohibited combinations, retaining only combinations that do not compromise device availability;

[0121] Sixth, archive all instruction entries according to region and terminal use to form a set of correction instructions for behavior constraint cards, which serves as the direct input for targeted adjustment and re-decoding in step S6.

[0122] In this embodiment, the set temperature range, end-use time window, and energy consumption ratio envelope in the targeted adjustment behavior constraint card are as follows: The target card is located in the behavior constraint card set using the household identifier vector; each correction instruction entry is read and adjusted within the specified end-use and time range; for the energy consumption ratio envelope, the target coefficient is increased or decreased within the lower and upper bounds according to the instructions, and the allocation coefficient is updated to ensure that the end-use allocation ratio is consistent with the energy consumption ratio envelope within the same time step; for the end-use time window, the coverage ratio of the allowed time period is modified according to the instructions, and a new time mask is generated so that the occupancy sequence and end-use behavior triggering only fall within the adjusted allowed time period; for the set temperature range, the upper and lower bounds of the temperature are tightened or loosened according to the instructions, and synchronously mapped in the physical constraint mapping layer as power upper limit, power change amplitude, and minimum duration parameters.

[0123] Specifically, in this implementation, the correction instruction set generated in step S5 is used as input. Without changing the device availability and prohibition combinations, the target coefficient of energy consumption percentage envelope, the percentage of end-use time window, and the set temperature range factor in the behavior constraint card are adjusted in a targeted manner. Then, re-decoding is performed on the gated end-use multi-stream generator to output the final household energy consumption profile. For each household index... End-use index (End-use set) (Including kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption), read the card-level calibration amount in step S5. and feasible domain gating ,in For equipment availability indication, To prohibit the combination of compliance instructions.

[0124] The adjustment amount is projected across the boundaries to form the update field:

[0125] ;

[0126] in: The lower and upper bounds of the energy consumption ratio are defined (derived from step S2). The original target coefficient; The percentage of the time window for original end use; This is the original set temperature range factor; For interval projection operators, the input values ​​are restricted to a closed interval. Inside; the superscript "new" indicates the new value after adjustment.

[0127] Will Parameterized into decoded gated signals and injected into the gated end-use multistream generator: based on Generate a new time mask ; Retain the original equipment legitimacy mask Unchanged; based on Mapped to power upper limit in the physical constraint mapping layer Power variation range With the shortest duration The three are mapped by the corresponding end-use functions. , , Monotonically obtained;

[0128] The new allocation coefficient is obtained by normalizing the set of active end uses:

[0129] ;

[0130] ;

[0131] in: For time indexing; The occupied sequence strength (output by the gated end-use multistream generator during re-decoding, with a value between zero and one); Set a new time mask (zero or one); This is a device validity mask (zero or one). For the use of new active terminals; This is the new allocation coefficient (from zero to one).

[0132] In the physical constraint mapping layer, the power time series is recalculated with update gating as follows:

[0133] ;

[0134] in: The adjusted power time series (power dimension), initial values ; New power limit; This represents the magnitude of the new power change; Shortest duration operator The intensity of the applied behavior sequence (zero to one) after the action, when the continuous segment is shorter than Time extended; This represents the end-use behavior sequence strength (from zero to one) output by the gated end-use multistream generator during re-decoding. This ensures that the power upper limit and slope constraints are met at the same time step.

[0135] The decoding process is as follows: Time window application Controlling the propagation of status during permitted and prohibited periods, the device combines filters to... Ensure legitimate devices are activated, and the linkage coupling module is activated. To implement the co-occurrence, sequence, and mutual exclusion allocation relationships between end-uses, the physical constraint mapping layer uses... , and Complete power decoding to obtain Then, according to the time step length The energy consumption of the end application is obtained by summing them up. The data is then compiled by summarizing total household power, peak power, peak periods, valley power, valley periods, and peak-valley ratios to form the final peak-valley characteristics. The final occupancy sequence will then be... Terminal usage behavior sequence Power time series Energy consumption for end-use applications The peak and valley characteristics are summarized to form the final household energy consumption profile and output.

[0136] In this embodiment, the process of generating a family-level profile including a device list, behavior sequence, and end-use energy consumption based on the final family energy consumption profile output by S6 is described.

[0137] Specifically, in this implementation, without changing device availability and prohibited combinations, the final household energy consumption profile output in step S6 is archived and structurally encapsulated to generate a household-level profile containing a device list, behavior sequence, and end-use energy consumption. Each household is indexed. On-end use of index set (Including kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption) and time index set Next, read the occupancy sequence output in step S6. Terminal usage behavior sequence Power time series Energy consumption for end applications The device availability set is then read from the behavior constraint card corresponding to that household, serving as the basis for the device list. Since the directional adjustment in step S6 does not change device availability... Combinations with Prohibited Therefore, the device list is consistent with it and is only recorded as a static field in the profile.

[0138] To clarify the components of a family-level profile, a family-level profile record is defined as follows:

[0139] ;

[0140] in: For family Family-level portraits; The device list is equivalent to the set of devices available for that household, i.e., for each end use. In device availability indication Record the corresponding available equipment and energy path in real time; For a set of behavior sequences, ,in The occupied sequence strength (from zero to one). For end-use behavior sequence strength (zero to one); For end-use energy consumption vectors, ,in For step S6 The energy consumption of end-use applications is obtained by summing them up.

[0141] For easier searching, you can include derived terms such as total energy consumption and the percentage of end-use applications:

[0142] ;

[0143] ;

[0144] in: Total household energy consumption (energy dimension). Percentage of household-level usage (dimensionless).

[0145] The portrait generation process is as follows:

[0146] Equipment list generated. For households. Read the device availability set from its behavior constraint card to form This set is consistent with step S6, consisting of... Together with the corresponding equipment type and energy path, they form a static list to express "what can be used".

[0147] Summary of behavioral sequences. The household's energy consumption sequence from the final household energy consumption profile in step S6 is summarized. End-use behavior sequence Index by Time Merging This sequence directly reflects "when and with what intensity it is triggered".

[0148] Energy consumption extraction for end-use applications. Read the energy consumption calculated in step S6. And assembled as To display total consumption and percentage, calculate using the formula above. and The derived quantity does not change the original record.

[0149] Structured output. Using the family identifier vector as the index key, ... Write to the profile database; the fields only contain a list of devices. Behavioral sequence Energy consumption for end applications Three parts; the peak and valley characteristics of step S6 can be associated with the power time series in the metadata to support higher-level analysis.

[0150] Through the above encapsulation, a family-level portrait is created. It can provide three key types of information: "which devices can be used", "when to use them", and "how much energy is consumed", which can be directly called in subsequent hierarchical statistics, load alignment and policy distribution, while maintaining consistency with macro statistical boundaries and behavioral constraint cards.

[0151] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for generating household energy consumption profiles based on a combination of deep learning and macro / micro perspectives, characterized in that, Includes the following steps: S1. Establish causal dependence based on regional energy structure, price, climate, building type, heating method, energy accessibility, income and family size, establish macro-statistical boundaries, define behavioral knob variables including set temperature, cooking frequency and time, hot water segmentation and lighting usage preference, and output hierarchical causal heterogeneity diagram and macro-statistical boundaries. S2. Following the causal path, the macro-statistical boundary is refined into household-level device availability, prohibited combinations, set temperature range, end-use time window, and energy consumption ratio envelope. Based on household attributes, a household identification vector is generated, and a set of behavioral constraint cards corresponding one-to-one with the household identification vector is output. S3. Parameterize the behavior constraint card into a decoded gating signal, and set up a time windower, device combination filter, linkage coupling module and physical constraint mapping layer in the end-use multi-stream generator so that the generation can only be expanded within the feasible domain, and output the gating end-use multi-stream generator. S4. Based on the behavior constraint card set and the gated end-use multi-stream generator, the occupancy sequence, end-use behavior sequence and power time series are generated sequentially. The end-use energy consumption and peak-valley characteristics are obtained through the physical constraint mapping layer, and the initial household energy consumption profile is output. S5. Based on the initial household energy consumption profile, aggregate the data by region and end-use to form macro results. Compare the macro results with the macro statistical boundary and attribute deviations to specific behavioral knobs and constraints based on the hierarchical causal heterogeneity diagram. Generate a set of correction instructions for behavioral constraint cards and output the set of correction instructions. S6. Adjust the set temperature range, time window and energy consumption ratio envelope in the behavior constraint card according to the calibration instruction set, and perform re-decoding on the end-use multi-stream generator after gating to output the final household energy consumption profile. S7. Generate a home-level profile based on the final home energy consumption profile, including a list of devices, behavioral sequences, and end-use energy consumption.

2. The method for generating a household energy consumption profile based on deep learning and a combination of macro and micro perspectives as described in claim 1, characterized in that, Step S1 is as follows: The hierarchical causal heterogeneous graph is specifically a scenario-based causal structure serving the generation of urban household energy consumption profiles. Its construction method is as follows: regional energy structure, energy price, and climate are macro-level nodes; building type, heating method, and energy accessibility are meso-level nodes; income and household size are micro-level nodes; behavioral knob variables are behavioral knob variable nodes; and kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, lighting energy consumption, and socket load energy consumption are end-use result nodes. Based on energy consumption business rules, causal dependencies are established between nodes. Specifically, climate and building type are directed to set temperature and end-use result nodes to constrain heat load, and energy price is directed to set temperature and cooking time. The system uses time period constraints to link energy accessibility to equipment availability and indirectly influence end-use results through behavioral knob variable nodes. It links income and household size to cooking frequency and hot water segmentation to constrain activity intensity, and links heating methods to heating-related end-use results to limit equipment paths. It embeds business-oriented monotonic directions and threshold conditions into causal dependencies, setting temperature to monotonically increase heating energy consumption and monotonically decrease cooling energy consumption, and energy price to monotonically decrease electricity use during high-price periods. When energy accessibility is lacking, the corresponding equipment path is disabled, and when the heating method is not covered, the heating end-use result is limited to zero, thus translating macro-level impacts into actionable behavioral constraints. The establishment of macro-statistical boundaries specifically involves: using statistical yearbooks and industry standards to uniformly describe the regional total energy volume, the proportion of different energy carriers, energy price structure, household number and attribute distribution, climate indicators, energy accessibility coverage, and building type proportion, forming the allowable range and consistency relationship for each indicator; mapping the macro-statistical boundaries to the feasible range of behavioral knob variables, specifically by using climate and energy prices to jointly limit the upper and lower limits of set temperature, using energy price structure to limit the selectable windows for cooking and lighting times, using household size to limit the range of values ​​for cooking frequency and hot water segmentation, using energy accessibility to limit the available set of equipment and accordingly limit the effective combination of behavioral knob variables, and limiting the proportion of different energy carriers to the energy consumption proportion envelope of end-use results.

3. The method for generating a household energy consumption profile based on deep learning and a combination of macro and micro perspectives as described in claim 1, characterized in that, Step S2 is as follows: The refinement of the macroscopic statistical boundary along the causal path is specifically as follows: based on the hierarchical causal heterogeneous graph and the macroscopic statistical boundary, constraints are passed down from the macroscopic nodes in a topological order, wherein energy accessibility and heating mode are mapped to equipment availability to obtain a list of available equipment. Based on equipment availability, prohibited combinations are generated according to the mutual exclusion of equipment and the lack of energy accessibility, and these prohibited combinations are written into the behavior constraint card. According to the impact of climate, building type, and building age on heat load, the set temperature range is limited to a feasible interval between the upper limit of heating temperature and the lower limit of cooling temperature. The impact of energy price structure on high-price and low-price periods is projected onto the end-use time window, ensuring that the selectable time periods for kitchen energy consumption, hot water energy consumption, cooling energy consumption, heating energy consumption, and lighting energy consumption meet the constraints of macroeconomic price signals and activity patterns. The macroeconomic statistical boundary of the proportion of energy carriers and the distribution of the number and attributes of households in the region is mapped to an energy consumption proportion envelope. Upper and lower bounds are set for the energy consumption proportion of each end use or each energy carrier at the household level, ensuring that the household-level profile remains consistent with the regional distribution after aggregation. A household identifier vector is generated based on region, income percentile, household size, housing type, building age, heating method, and energy accessibility. Using the household identifier vector as an index, the above-mentioned equipment availability, prohibited combinations, set temperature range, end-use time window, and energy consumption proportion envelope are assembled into a behavior constraint card.

4. The method for generating household energy consumption profiles based on a combination of deep learning and macro / micro perspectives as described in claim 1, characterized in that, Step S3 is as follows: The time window controller is a module that controls the start and stop of time steps based on the time window of the terminal application; The device combination filter is a module that verifies the legality of device activation status based on device availability and prohibited combinations; The linkage coupling module is a constraint module used to apply constraints on simultaneous occurrence, sequential order and mutual exclusion between end uses. The constraints are based on the allocation relationship between the end use time window and the energy consumption ratio envelope from the behavior constraint card. The physical constraint mapping layer is a module that maps the set temperature range and end-use behavior to power limits, power variation range and duration limits; The feasible domain is a set of states that simultaneously satisfy the end-use time window, device availability, prohibited combinations, set temperature range, and energy consumption percentage envelope. The parameterization of the behavior constraint card into a decoded gating signal specifically involves: reading the end-use time period window from the behavior constraint card and encoding it into a time mask vector, which is then input to the time period windower; reading the device availability from the behavior constraint card and combining it with prohibited combinations to generate a device legality mask, which is then input to the device combination filter; reading the set temperature range from the behavior constraint card and converting it into end-use power upper limit, power change amplitude, and minimum duration parameters, which are then input to the physical constraint mapping layer; and reading the energy consumption ratio envelope from the behavior constraint card and converting it into end-use allocation coefficients, which are then input to the linkage coupling module and the terminal decoding output layer. The end-use multi-stream generator includes a structure of multi-stream decoding sub-modules. Each end-use corresponds to a decoding sub-module, and during the decoding process, it receives gating in the following order: the time windower blocks the propagation of the state in the prohibited time period based on the time mask vector and allows it in the allowed time period; the device combination filter applies a device legality mask to the active state of the candidate device at each time step, and only retains the state that is consistent with the device availability and has not triggered the prohibited combination. The linkage coupling module constrains and allocates the co-occurrence, sequence, and mutual exclusion of each end application based on the overlapping relationship of the end application time window and the allocation coefficient of the energy consumption ratio envelope, so that the multiple flows maintain a consistent allocation relationship in terms of time and intensity. The physical constraint mapping layer maps and truncates the decoded power and duration by setting the power upper limit, power change amplitude, and shortest duration corresponding to the set temperature range, so that the output falls into the feasible region.

5. The method for generating a household energy consumption profile based on deep learning and a combination of macro and micro perspectives as described in claim 1, characterized in that, Step S4 is as follows: The process of sequentially generating the occupancy sequence, end-use behavior sequence, and power time sequence is as follows: the behavior constraint card set is processed card by card, and the gated end-use multi-stream generator reads the end-use time window under the control of the time window controller to generate an occupancy sequence covering the day and quarter, so that the occupancy status only falls within the allowed time period. Based on the usage behavior triggered by the occupancy sequence, the device availability and prohibited combinations are read in the device combination filter to eliminate non-compliant device activation states. The energy consumption ratio envelope is read in the linkage coupling module to allocate and restrict the co-occurrence, sequence, and mutual exclusion relationships of multiple end uses, and output the end use behavior sequence of each end use. The end use behavior sequence is sent to the physical constraint mapping layer to read the set temperature range and map it to the power upper limit, power change amplitude, and minimum duration. The power of the end use behavior sequence is decoded to obtain the power time series aligned with the time step. The end use energy consumption is obtained by accumulating the power time series by end use, and the peak power, peak occurrence period, valley power, and valley occurrence period are extracted in the time dimension to calculate the peak-valley ratio and form peak-valley features. The occupancy sequence, end-use behavior sequence, power time series, end-use energy consumption, and peak-valley characteristics are summarized to form an initial household energy consumption profile.

6. The method for generating a household energy consumption profile based on deep learning and a combination of macro and micro perspectives as described in claim 1, characterized in that, Step S5 is as follows: The aggregation by region and end-use to form macro results, and the comparison with macro statistical boundaries, specifically involves: reading power time series and end-use energy consumption, grouping households by region, and accumulating end-use energy consumption and peak-valley characteristics over time to obtain regional end-use energy consumption; converting the regional end-use energy consumption into end-use proportions, comparing them with the corresponding proportion intervals in the macro statistical boundaries for each end-use, forming difference values ​​and interval boundary markers for each region, and verifying the consistency of peak power and valley power in the time periods to obtain the set of differences between the macro results and the macro statistical boundaries; The method of attributing deviations to specific behavior knobs and constraints based on hierarchical causal heterogeneity graphs is as follows: determine the macroscopic node where the deviation occurs based on the set of differences, enumerate the reachable paths from the macroscopic node to the fields in the behavior constraint card along the directed edges of the hierarchical causal heterogeneity graph, determine the allocation ratio according to the end use and time range covered by the path, and allocate each deviation amount to the candidate adjustment amount of the corresponding behavior knob; under the premise that the equipment availability and prohibition combination are not violated, priority is given to allocating to the energy consumption ratio envelope and end use time window; when the interval and time period have reached the upper or lower limit, it is then allocated to the set temperature range to adjust the power upper limit and continuous constraints. For deviations from cross-end uses, the constraint relationships between end uses in the hierarchical causal heterogeneity diagram are used to map the allocation ratio to the behavior knobs of the corresponding end uses according to co-occurrence, sequence and mutual exclusion relationships, so as to obtain a set of executable adjustment quantities for card level; The process of generating a calibration instruction set for behavior constraint cards involves: locating the target card in the behavior constraint card set using the family identifier vector as an index; discretizing the executable adjustment quantities into instruction entries, with each instruction including at least the target field, adjustment direction and magnitude, applicable time range, and priority; for conflicts among multiple instructions within the same behavior constraint card, consistency is determined based on priority and prohibited combinations, retaining only combinations that do not compromise device availability; and finally, the final instruction entries are archived by region and terminal purpose to form a calibration instruction set for behavior constraint cards.

7. The method for generating household energy consumption profiles based on a combination of deep learning and macro / micro perspectives as described in claim 1, characterized in that, Step S6 is as follows: The specific steps for setting the temperature range, end-use time window, and energy consumption ratio envelope in the targeted adjustment behavior constraint card are as follows: locate the target card in the behavior constraint card set using the household identification vector, read the correction instruction entries one by one, and perform the adjustment within the specified end-use and time range; for the energy consumption ratio envelope, increase or decrease the target coefficient within the lower and upper bounds according to the instructions and update the allocation coefficient so that the end-use allocation ratio and energy consumption ratio envelope are consistent within the same time step; For end-use time periods, the coverage ratio of the allowed time periods is modified according to the instructions, and a new time mask is generated so that the occupancy sequence and end-use behavior triggering only fall within the adjusted allowed time periods; for the set temperature range, the upper and lower temperature boundaries are tightened or loosened according to the instructions, and synchronously mapped to power upper limit, power change amplitude and minimum duration parameters in the physical constraint mapping layer.

Citation Information

Patent Citations

  • Modeling method and system for user charging behavior portrait

    CN120146915A

  • Clinical research data analysis method based on machine learning

    CN120409633A