Building load prediction and flexible control system and device

Abstract

The invention belongs to the technical field of load regulation and control, and discloses a building load prediction and flexible control system and device.The system comprises a mechanism twinborn reference simulation module, a data driving load prediction module, a load sequence fusion module, a flexible strategy mapping module and a strategy preferential coding module; a mechanism type digital twinborn body is constructed, and a theoretical reference load sequence is obtained through standard working condition simulation; based on historical operation data and real-time input data, constructing a data-driven digital twin, and performing dynamic load prediction to obtain a data-driven load prediction sequence; adaptively fusing the two types of sequences to obtain a load prediction sequence; performing multi-target flexible strategy mapping on the load prediction sequence to obtain a flexible candidate strategy; and preferentially coding the candidate flexible strategy efficiency based on the two types of digital twin bodies to obtain a flexible control instruction. According to the invention, the prediction and flexible control efficiency of the building load can be improved.

Description

Technical Field

[0001] This invention belongs to the field of load regulation technology, and particularly relates to a building load prediction and flexible control system and device. Background Technology

[0002] As the construction industry transforms towards intelligence and energy conservation, building load forecasting and flexible regulation have become the core directions for improving energy efficiency and optimizing operating costs. Digital twin technology, with its entity mapping and dynamic simulation capabilities, is widely used in this field. Its derived mechanistic and data-driven models provide key support for load analysis. By integrating and optimizing forecast results and mapping regulation strategies, multi-objective optimization of building energy conservation, cost control, and comfort assurance can be achieved.

[0003] Existing building load forecasting and flexible control technologies typically employ either mechanistic or data-driven digital twin models. When forecasting loads, it's difficult to simultaneously balance theoretical rigor with adaptability to actual operating conditions. Mechanistic models, constrained by fixed physical rules, lack adaptability to dynamic changes in actual operation. Data-driven models rely on historical data distribution and lack support for the fundamental laws governing building energy transfer. This results in load forecast accuracy and reliability failing to meet actual control requirements, and the load sequence fusion and flexible strategy formulation processes lack dynamic adaptive capabilities. A few hybrid forecasting methods, when fusing different forecast sequences, fail to adequately consider the real-time reliability of forecast results and deviations from actual operating conditions in weight allocation. Furthermore, strategy formulation lacks a dynamic correlation with forecast characteristics and operating condition changes, leading to insufficient targeting of control strategies and difficulty in achieving flexible control effectiveness for multi-objective building optimization. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a building load prediction and flexible control system and device, which can improve the efficiency of building load prediction and flexible control.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A building load forecasting and flexible control system, comprising: The mechanistic twin benchmark simulation module is used to construct a mechanistic digital twin of the target building based on the design parameters and physical laws of the target building, and to perform standard working condition simulation on the mechanistic digital twin to obtain the theoretical benchmark load sequence of the target building. The data-driven load forecasting module is used to construct a data-driven digital twin of the target building based on its historical operating data and real-time input data, and to perform dynamic load forecasting on the data-driven digital twin to obtain the data-driven load forecasting sequence of the target building. The load sequence fusion module is used to adaptively fuse the theoretical baseline load sequence and the data-driven load forecast sequence to obtain the load forecast sequence of the target building. The flexible strategy mapping module is used to perform multi-objective flexible strategy mapping on the load forecast sequence to obtain flexible candidate strategies for the target building. The strategy optimization coding module is used to perform performance optimization coding on candidate flexible strategies based on mechanistic digital twins and data-driven digital twins to obtain flexible control instructions for the target building.

[0006] Preferably, in the mechanism twin benchmark simulation module, the geometric information of the target building is topologically processed according to the design parameters of the target building to obtain the geometric topology of the target building; Based on the geometric topology, the material properties in the design parameters are identified item by item to obtain the material property entries of the target building. Based on the structural hierarchy description in the design parameters, the material property items are hierarchically assigned to obtain the thermal property list of the target building. The heat exchange logic between the interior space of the target building and the external environment is defined by rules, resulting in the energy transfer rule set of the target building. By using a thermal property list and an energy transfer rule set, dynamic energy interaction is performed on the geometric topology to obtain a mechanistic digital twin of the target building. Standard operating condition simulations were performed on the mechanistic digital twin to obtain the theoretical baseline load sequence of the target building.

[0007] Preferably, in the mechanistic twin baseline simulation module, standard load condition simulation is performed on the mechanistic digital twin to obtain the theoretical baseline load sequence of the target building, specifically: Based on pre-defined building codes and climate zones, standard operating conditions for the mechanism-based digital twin are set. Based on standard operating conditions, the mechanistic digital twin is dynamically simulated according to a preset simulation cycle to obtain the building load dataset within the simulation cycle. The theoretical baseline load sequence of the target building is obtained by performing time-series smoothing aggregation on the building load dataset.

[0008] Preferably, in the data-driven load forecasting module, the historical operating data of the target building is analyzed to obtain the historical feature sequence of the target building; By performing operational mode decomposition on the historical feature sequence, the steady-state operation law and dynamic response law of the target building are obtained; By performing aggregated analysis on the steady-state operation law and the dynamic response law, the twin mapping relationship of the target building is obtained; By injecting real-time input data of the target building into the twin mapping relationship, the dynamic evolution relationship of the target building can be obtained; Virtual entities are constructed based on dynamic evolutionary relationships to obtain a data-driven digital twin of the target building; Dynamic load forecasting is performed on the data-driven digital twin to obtain the data-driven load forecast sequence for the target building.

[0009] Preferably, in the data-driven load forecasting module, dynamic load forecasting is performed on the data-driven digital twin to obtain the data-driven load forecasting sequence for the target building, specifically as follows: Obtain the real-time operating status and future weather forecast of the target building to obtain the real-time predictive characteristics of the target building; By inputting real-time forecast features into a data-driven digital twin, preliminary load values ​​for the target building are obtained. Based on historical forecast deviations, the initial load value is reliably calibrated to obtain the calibrated load value for the target building. The formula for calculating the calibrated load value is as follows: ; In the formula, The calibrated load value obtained at time t, The initial load value obtained at time t, Let be the dynamic weighted average of the historical prediction biases at time t. Let be the standard deviation obtained at time t based on historical prediction bias. The moving average of the actual load at time t, obtained from historical operating data. The preset uncertainty suppression coefficient, The preset saturation adjustment parameters are: The standard deviation correction factor is the preset value. For symbolic functions, It is an exponential decay factor. The dynamic weighted average of historical prediction bias The absolute value; The calibrated load values ​​are serialized and normalized to obtain the data-driven load prediction sequence for the target building.

[0010] Preferably, in the load sequence fusion module, the process of obtaining the load forecast sequence of the target building is as follows: The confidence level of the data-driven load forecast sequence is quantified to obtain the real-time confidence index of the data-driven load forecast sequence; Based on the real-time operating condition information of the target building, the fit deviation of the theoretical reference load sequence is evaluated to obtain the operating condition deviation index of the theoretical reference load sequence. Based on real-time confidence index and load condition deviation index, the theoretical baseline load sequence and the data-driven load prediction sequence are dynamically weighted and matched to obtain the initial fused load sequence of the target building. The initial fused load sequence is continuously smoothed to obtain the load prediction sequence for the target building.

[0011] Preferably, in the load sequence fusion module, based on real-time confidence indicators and load condition deviation indicators, the theoretical baseline load sequence and the data-driven load prediction sequence are dynamically weighted and matched to obtain the initial fused load sequence of the target building, specifically as follows: The real-time confidence index and the operating condition deviation index are normalized to obtain the quantitative values ​​of the confidence index and the quantitative values ​​of the deviation of the target building. Based on the confidence metric and the offset metric, the weighting relationship between the theoretical baseline load series and the data-driven load forecast series is jointly determined; The weight allocation relationship is normalized to obtain the fused weight vector of the weight allocation relationship; Based on the fusion weight vector, the theoretical baseline load sequence and the data-driven load prediction sequence are weighted and fused to obtain the initial fused load sequence of the target building.

[0012] Preferably, in the flexible strategy mapping module, the process of obtaining the flexible candidate strategy for the target building is as follows: Multi-objective reconstruction of the load forecast sequence yields multi-dimensional optimization objectives for the target building. Based on multi-dimensional optimization objectives, the load forecast sequence is deconstructed according to time-series features to obtain the load dynamic characteristics of the target building; By associating load dynamic characteristics and multi-dimensional optimization objectives with strategy orchestration, a preliminary flexible control scheme for the target building is obtained. Based on multi-dimensional optimization objectives, a comprehensive performance evaluation of the preliminary flexible control scheme is conducted to obtain flexible candidate strategies for the target building.

[0013] Preferably, in the strategy optimization coding module, the process of obtaining the flexible control instructions for the target building is as follows: Based on the mechanistic digital twin, the mechanistic consistency of candidate flexible strategies is evaluated, and the mechanistic compliance sequence of candidate flexible strategies is obtained. Based on the data-driven digital twin, the historical performance of candidate flexible strategies is predicted to obtain the data-driven performance sequence of candidate flexible strategies. The fusion performance evaluation of the mechanism compliance sequence and the data-driven performance sequence is carried out to obtain the comprehensive performance score of the candidate flexible strategy; Based on the comprehensive performance score, the candidate flexible strategies are ranked and selected to obtain the optimal flexible strategy identifier for the target building. The optimal flexibility strategy identifier is encoded as the flexibility control command for the target building.

[0014] A building load prediction and flexible control device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The system described above is implemented by the processor executing the program.

[0015] The present invention has the following beneficial effects: This invention constructs a dual digital twin consisting of a mechanistic model and a data-driven model. The mechanistic model relies on physical laws to ensure its theoretical foundation, while the data-driven model combines historical and real-time data to adapt to dynamic operating conditions. Through an adaptive fusion mechanism, the real-time confidence of the data-driven prediction is quantified, the deviation between the benchmark sequence of the evaluation mechanism and the actual operating conditions is assessed, and the weighted fusion sequence is dynamically allocated to compensate for the shortcomings of a single model, improve the accuracy and reliability of load prediction, and balance theoretical rigor with adaptability to actual operating conditions.

[0016] This invention is based on a precisely fused load forecast sequence, dynamically correlates the forecast time series characteristics with multi-dimensional optimization objectives to arrange a preliminary control scheme, and combines dual twins to evaluate candidate strategies from two dimensions: mechanistic consistency and historical operating efficiency. The optimal strategy is then encoded into control commands, enabling the strategy to adapt to changes in operating conditions, ensuring the targeted control and multi-objective optimization, enhancing the targetedness and effectiveness of flexible control strategies, and achieving multi-objective optimization efficiency. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of the system of the present invention. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0019] Example 1: A building load forecasting and flexible control system can be set up in a cloud server. In terms of implementation, it can be used as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed into a website. Figure 1 As shown, based on the implemented functions, a building load forecasting and flexible control system includes a mechanistic twin benchmark simulation module, a data-driven load forecasting module, a load sequence fusion module, a flexible strategy mapping module, and a strategy optimization coding module. The modules of this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform fixed functions, stored in the memory of the electronic device. Each of the above modules can be implemented independently and can call other modules.

[0020] The mechanistic twin baseline simulation module is used to construct a mechanistic digital twin of the target building based on its design parameters and physical laws, and to perform standard load simulations on the mechanistic digital twin to obtain the theoretical baseline load sequence of the target building. Based on the design parameters of the target building, the geometric information of the target building is topologically processed to obtain the geometric topology of the target building; Based on the geometric topology, the material properties in the design parameters are identified item by item to obtain the material property entries of the target building. Based on the structural hierarchy description in the design parameters, the material property items are hierarchically assigned to obtain the thermal property list of the target building. The heat exchange logic between the interior space of the target building and the external environment is defined by rules, resulting in the energy transfer rule set of the target building. By using a thermal property list and an energy transfer rule set, dynamic energy interaction is performed on the geometric topology to obtain a mechanistic digital twin of the target building. Standard load simulations were performed on the mechanistic digital twin to obtain the theoretical baseline load sequence of the target building: Based on pre-defined building codes and climate zones, standard operating conditions for the mechanism-based digital twin are set. Based on standard operating conditions, the mechanistic digital twin is dynamically simulated according to a preset simulation cycle to obtain the building load dataset within the simulation cycle. The theoretical baseline load sequence of the target building is obtained by performing time-series smoothing aggregation on the building load dataset.

[0021] The design parameters of the target building include geometric information such as building dimensions, wall layout, room distribution, and door and window positions. Based on these design parameters, the spatial connection relationships of each geometric part are sorted out, the adjacency relationships between each room, the connection methods between walls and the ground and roof, and the assembly position relationship between doors and windows and walls are clarified. These scattered geometric information are organized into a structured expression, and finally a geometric topology that can clearly reflect the spatial relationship and composition structure of each geometric part of the building is formed.

[0022] The geometric topology has clearly defined the spatial structure of each geometric part of the building. The design parameters also include information on the materials used in each geometric part, such as the type of bricks in the walls, the type of insulation material in the roof, and the glass material of the doors and windows. In conjunction with the geometric topology, the material information corresponding to each geometric part is identified one by one, and the information of each geometric part and its corresponding material is organized separately to form material attribute entries.

[0023] The structural hierarchy description in the design parameters clarifies the material stacking order of each part of the building. For example, the description of the exterior wall from the outside to the inside is as follows: decorative layer, insulation layer, and load-bearing layer. According to this structural hierarchy description, the previously obtained material attribute entries are classified according to the corresponding hierarchy. For example, the material attribute entries corresponding to each layer in the exterior wall structural hierarchy are grouped together to clarify the materials corresponding to each structural hierarchy and integrate them to form a thermal attribute list that can reflect the thermal properties of materials related to building heat transfer.

[0024] The interior space of the target building includes enclosed or semi-enclosed areas such as rooms and corridors. The external environment includes external conditions such as atmospheric temperature, solar radiation, and wind. The heat exchange logic refers to the ways and paths of heat transfer between the interior space and the external environment, such as heat conduction through walls, radiative heat exchange through doors and windows, and convective heat exchange through ventilation. The conditions for the occurrence and the transfer path of each heat exchange method are clearly defined, such as the heat exchange requirements for solar radiation entering the interior space through doors and windows when there is no obstruction, forming a set of energy transfer rules.

[0025] The thermal property list provides the material thermal properties of each part of the building, the energy transfer rule set clarifies the mode and path of heat exchange, and the geometric topology reflects the spatial architecture of the building. By combining these three, the heat exchange process of each geometric part of the building under different energy transfer rules is simulated based on its own material thermal properties. For example, according to the thermal conductivity of the wall material, the heat transfer between the wall and the internal and external environment is simulated according to the thermal conductivity rules and the spatial position of the wall. Finally, a mechanism-based digital twin that can accurately replicate the building's geometry, material thermal properties and energy transfer process is constructed.

[0026] Building codes are industry-specific standards for building operation, such as suitable indoor temperature ranges and ventilation requirements. Climate zones are regions divided according to the climate characteristics of different areas, such as tropical, temperate, and frigid zones. By combining these pre-defined building codes and the climate zone in which the target building is located, the basic conditions for the operation of the mechanism-based digital twin are determined. For example, in a temperate climate zone, standard indoor temperature, standard humidity, and standard meteorological parameters of the external environment are set according to building codes to form standard operating conditions.

[0027] Standard operating conditions provide a fixed basic environment for dynamic simulation. The preset simulation cycle is a set simulation time interval, such as one simulation cycle per hour. The mechanistic digital twin simulates the energy consumption of the building according to the standard operating conditions in each simulation cycle, such as the amount of heat or cooling required by the building to maintain the standard indoor temperature each hour. Relevant data are collected in each simulation cycle to form a building load dataset.

[0028] The building load dataset consists of scattered data from different simulation periods. These scattered data are organized in chronological order to eliminate random fluctuations. For example, data from multiple consecutive simulation periods are arranged chronologically, and data from adjacent periods are processed to ensure a smooth transition, making the data changes more continuous and reasonable. Finally, a stable and continuous theoretical benchmark load sequence is formed in chronological order, which can reflect the load change pattern of buildings under standard working conditions.

[0029] The mechanistic twin benchmark simulation module clarifies the spatial relationships between various geometric parts of a building, providing a precise spatial foundation for subsequent material property identification and energy interaction simulation. It achieves accurate correspondence between building geometry and materials, providing accurate basic data for structural hierarchy attribution and thermal property list formation. It clearly presents the material composition of each structural level of the building, providing targeted material basis for the application of energy transfer rules and dynamic energy interaction relationships. It provides clear normative basis for simulating building energy transfer processes, ensuring the accuracy and standardization of energy transfer simulation. It replicates the physical characteristics of the building and the energy transfer process, providing high-precision virtual model support for standard operating condition simulation. It ensures that the simulation operating environment conforms to actual scenarios and industry standards, providing a scientific basis environment for dynamic simulation. It comprehensively reflects building load changes within the simulation cycle, providing rich basic data for data aggregation and the formation of theoretical benchmark load sequences, eliminating random data fluctuations, accurately reflecting the building load change patterns under standard operating conditions, and providing reliable benchmark data for load forecasting and flexible control strategy formulation.

[0030] The data-driven load forecasting module is used to construct a data-driven digital twin of the target building based on its historical operating data and real-time input data, and to perform dynamic load forecasting on the data-driven digital twin to obtain the data-driven load forecast sequence of the target building. The historical operational data of the target building is analyzed to obtain the historical feature sequence of the target building. By performing operational mode decomposition on the historical feature sequence, the steady-state operation law and dynamic response law of the target building are obtained; By performing aggregated analysis on the steady-state operation law and the dynamic response law, the twin mapping relationship of the target building is obtained; By injecting real-time input data of the target building into the twin mapping relationship, the dynamic evolution relationship of the target building can be obtained; Virtual entities are constructed based on dynamic evolutionary relationships to obtain a data-driven digital twin of the target building; Dynamic load forecasting is performed on the data-driven digital twin to obtain the data-driven load forecast sequence for the target building: Obtain the real-time operating status and future weather forecast of the target building to obtain the real-time predictive characteristics of the target building; By inputting real-time forecast features into a data-driven digital twin, preliminary load values ​​for the target building are obtained. Based on historical forecast deviations, the initial load value is reliably calibrated to obtain the calibrated load value for the target building. The formula for calculating the calibrated load value is as follows: ; In the formula, The calibrated load value obtained at time t, The initial load value obtained at time t, Let be the dynamic weighted average of the historical prediction biases at time t. Let be the standard deviation obtained at time t based on historical prediction bias. The moving average of the actual load at time t, obtained from historical operating data. The preset uncertainty suppression coefficient, The preset saturation adjustment parameters are: The standard deviation correction factor is the preset value. For symbolic functions, It is an exponential decay factor. The dynamic weighted average of historical prediction bias The absolute value of.

[0031] The calibrated load values ​​are serialized and normalized to obtain the data-driven load prediction sequence for the target building.

[0032] The historical operational data of the target building consists of all relevant data recorded during the actual operation of the building over a period of time, including energy consumption, equipment operating parameters, and internal environmental data. Feature analysis involves comprehensively sorting through this historical operational data and extracting key information that reflects the building's operational status and load correlation. This information includes data trends, numerical ranges, and frequency of occurrence. The resulting historical feature sequence is a collection of all extracted key information arranged in chronological order, with each data point corresponding to a specific time point and related characteristics of the building's operation.

[0033] Historical feature sequence is a set of key information about building operation arranged in chronological order. Operation mode decomposition breaks down this sequence according to its inherent change patterns. Based on the stability and response characteristics of data fluctuations, the parts of the sequence that change smoothly, follow consistent patterns, and have small fluctuation amplitudes are identified to form steady-state operation patterns. At the same time, the parts of the sequence that fluctuate significantly, are greatly affected by external factors, and change rapidly are identified to form dynamic response patterns. These two patterns fully demonstrate the building's operation characteristics under stable and dynamically changing conditions, respectively.

[0034] Steady-state operation law reflects the operational characteristics of a building under stable operating conditions, while dynamic response law reflects the operational characteristics of a building under dynamically changing operating conditions. Aggregate analysis combines these two laws for comprehensive analysis, explores the intrinsic relationship between their interaction and mutual influence, clarifies the correspondence between various characteristics under different operating states, and ultimately forms a set of twin mapping relationships that can accurately reflect the corresponding relationship between building operation laws and loads, providing core support for the subsequent construction of digital twins.

[0035] Twin mapping is a set of corresponding associations between building operation patterns and loads. Real-time input data consists of various operation-related data of the target building at the current moment, including current equipment operating parameters and current internal environmental data. By integrating these real-time input data into the twin mapping, and adjusting and updating the original mapping relationship in combination with the current actual operation, the resulting dynamic evolution relationship can reflect the correlation between the building's operating status and load change trends under the influence of current real-time data, thus embodying the real-time dynamic attributes of building operation.

[0036] Dynamic evolution relationships are the correlations that reflect the real-time dynamic changes of a building. Virtual entity construction is based on this relationship, comprehensively simulating the building's operating logic, data interaction mode, load generation mechanism, and other aspects to build a virtual model that highly matches the actual operating state of the target building. This data-driven digital twin can accurately replicate the operating process of the target building, respond to changes in various input data in real time, and completely match the operating characteristics of the actual building.

[0037] Real-time operational status refers to the actual operational data of the target building at the current moment, such as equipment operation, energy consumption, and internal environmental parameters. Future weather forecasts are meteorological data released by authoritative meteorological agencies for the area where the target building is located over a future period, including information such as temperature, humidity, and wind speed. By integrating these two types of data and extracting key information related to the building load, the resulting real-time forecast features are a comprehensive set of information that reflects the current operational foundation of the building and the impact of the future external environment, providing both real-time and future-oriented basis for subsequent load forecasting.

[0038] Real-time predictive features are a comprehensive collection of information on the current operational foundation and future environmental impacts. A data-driven digital twin is a virtual model that replicates the actual operational characteristics of a building. Real-time predictive features are transmitted to this virtual model, which then processes the input information comprehensively based on the building's replicated operational logic and relationships. This simulates the building's load generation process in the current and future environments, and the final output preliminary load value is the expected load value generated by the building at the corresponding time point, directly presenting the initial prediction results.

[0039] The preliminary load value is the initial predicted load value output by the virtual model. The historical prediction deviation is the difference between the predicted load value and the actual load value of the same period over a period of time. By analyzing the overall situation of these differences, the deviation pattern and range of the prediction results are clarified. Then, the preliminary load value is adjusted according to this pattern to eliminate the impact of the deviation. The calibrated load value obtained is a load value that is closer to the actual situation, ensuring the accuracy of the prediction results.

[0040] In the formula for calculating the load value after calibration It is the result of the preliminary load value after correction of historical forecast deviations, which is used to form a data-driven load forecast sequence to match the actual building load situation; It is derived from a digital twin simulation calculation driven by real-time predictive feature input data; The index-based decay factor is calculated by weighting the difference between historical forecast load and actual load according to time distance and degree of impact, reflecting the overall trend of historical forecast deviation. It is obtained by calculating the correlation between the uncertainty suppression coefficient and the standard deviation, and is used to reduce the impact of historical prediction bias as the standard deviation increases; Based on the building's operational characteristics, load fluctuation patterns, and prediction accuracy requirements, the influence of the standard deviation on the calibration results is controlled. This data is derived by statistically analyzing all historical prediction deviation data, reflecting the range and dispersion of deviation fluctuations. Based on the range of building load changes and the preset stability requirements, the impact of historical prediction deviations on the calibration results is limited to avoid overcorrection; It only measures the magnitude of the deviation, without considering the direction; The correction level for standard deviation is adjusted based on the sensitivity of building load to standard deviation and the characteristics of operational data fluctuations. It is calculated continuously over a fixed time window based on historical actual loads and serves as a reference benchmark for standard deviation correction. The sign function is derived from the positive or negative attribute of the dynamic weighted average of historical prediction deviations. It is used to adjust the calibration direction in a targeted manner. The overall formula achieves the reliability calibration of the initial load value by multiplying the initial load value with a correction term that combines factors such as historical prediction deviations and standard deviations, and finally obtains a load value that fits the actual situation.

[0041] The calibrated load values ​​are accurate load values ​​after deviation adjustment. Serialization and normalization arrange these values ​​in chronological order to ensure that each value corresponds to the correct prediction time point, forming an ordered sequence. The resulting data-driven load prediction sequence is a set of values ​​arranged in chronological order that accurately reflects the load changes of the building over a period of time in the future, fully presenting the temporal distribution characteristics of the load.

[0042] The data-driven load forecasting module retains core building operation data characteristics, providing accurate foundational data for subsequent analysis and avoiding deviations caused by missing key information. It clearly distinguishes the characteristics of different building operating conditions, identifies the features of stable and dynamically changing components, lays a solid foundation for constructing mapping relationships, clarifies the intrinsic correlation between building operation patterns and load, transforming it into a directly applicable predictive basis. This enhances the logic and relevance of the forecasting process, ensuring that building operation correlations closely follow actual conditions. The digital twin possesses real-time dynamic adjustment capabilities, guaranteeing synchronization with the actual building's operating status, accurately replicating building operation characteristics, and providing a virtual simulation platform highly consistent with the actual building. This improves the reliability of load forecasting results, taking into account both the building's current operational foundation and future external environmental influences. It provides comprehensive forecast input information, avoiding one-sided forecast results. Through the virtual model, it completes preliminary load simulation calculations, outputting clear load values, providing specific operational objects for subsequent calibration steps, eliminating the impact of historical forecast deviations, improving the accuracy of load values, and preventing unreliable final results due to initial forecast deviations. The calibrated load values ​​are arranged in chronological order, fully presenting the building's future load change trends, providing clear and systematic data support for flexible strategy formulation.

[0043] The load sequence fusion module is used to adaptively fuse the theoretical baseline load sequence and the data-driven load forecast sequence to obtain the load forecast sequence for the target building. The confidence level of the data-driven load forecast sequence is quantified to obtain the real-time confidence index of the data-driven load forecast sequence; Based on the real-time operating condition information of the target building, the fit deviation of the theoretical reference load sequence is evaluated to obtain the operating condition deviation index of the theoretical reference load sequence. Based on real-time confidence indices and load condition deviation indices, the theoretical baseline load sequence and the data-driven load forecast sequence are dynamically weighted to obtain the initial fused load sequence for the target building: The real-time confidence index and the operating condition deviation index are normalized to obtain the quantitative values ​​of the confidence index and the quantitative values ​​of the deviation of the target building. Based on the confidence metric and the offset metric, the weighting relationship between the theoretical baseline load series and the data-driven load forecast series is jointly determined; The weight allocation relationship is normalized to obtain the fused weight vector of the weight allocation relationship; Based on the fusion weight vector, the theoretical baseline load sequence and the data-driven load prediction sequence are weighted and fused to obtain the initial fused load sequence of the target building; The initial fused load sequence is continuously smoothed to obtain the load prediction sequence for the target building.

[0044] When quantifying the confidence of a data-driven load forecast sequence, the generated data-driven load forecast sequence is used as a basis. This is combined with the sequence's generation criteria, including the completeness of historical operational data, the accuracy of real-time input data, and the stability of the dynamic load forecasting process. The confidence level of each forecast value is assessed by verifying whether historical operational data fully covers relevant operational scenarios, whether real-time input data accurately reflects the current operational status, and whether the operations of each stage in the dynamic forecasting process are consistent and without anomalies. The confidence level of each forecast value is then determined. Finally, the confidence assessment results of all time points are integrated to form a real-time confidence index. The real-time confidence index is a set of quantified results of the confidence level of each forecast value in the corresponding data-driven load forecast sequence, with each forecast value at each time point associated with a corresponding confidence level label.

[0045] When assessing the fit deviation of the theoretical baseline load sequence based on the real-time operating information of the target building, the content of the real-time operating information is first clarified, namely, the current equipment operating status, internal environmental parameters, and external real-time meteorological conditions of the target building. Then, using the standard operating conditions corresponding to the theoretical baseline load sequence as a reference, the standard operating conditions and the real-time operating conditions are compared one by one at each time node. The specific differences in operating conditions between the two are analyzed, such as the difference between the actual indoor temperature and the standard temperature, and the differences between the actual external wind speed, temperature, and other meteorological parameters and the standard meteorological parameters. Based on these differences, the fit deviation of the theoretical baseline load sequence under the real-time operating conditions is calculated. Finally, the deviations at all time nodes are summarized to form the operating condition deviation index. The operating condition deviation index is a set of results reflecting the fit deviation between the theoretical baseline load sequence and the real-time operating conditions, and each time node corresponds to a deviation degree indicator.

[0046] When normalizing the real-time confidence index and the operating condition deviation index, the acquired real-time confidence index and the operating condition deviation index are used as the objects. The confidence values ​​of different ranges in the real-time confidence index are adjusted to a unified numerical range. At the same time, the deviation values ​​of different degrees in the operating condition deviation index are also adjusted to this unified numerical range to ensure that the two indices are consistent in numerical range and have the conditions for direct comparison. After adjustment, the quantified confidence value and the quantified deviation value are obtained. The quantified confidence value is a specific value within the unified range that reflects the credibility of the data-driven load forecast sequence, and the quantified deviation value is a specific value within the unified range that reflects the degree of deviation between the theoretical benchmark load sequence and the real-time operating condition.

[0047] When determining the weight allocation relationship based on the combined confidence and offset quantification values, these two quantification values ​​are the core analytical basis for comprehensive consideration. When both the confidence and offset quantification values ​​are high, it indicates that the data-driven load prediction sequence has high reliability and the theoretical baseline load sequence deviates significantly from real-time operating conditions. In this case, the data-driven load prediction sequence is assigned a higher weight, and the theoretical baseline load sequence a lower weight. Conversely, when both the confidence and offset quantification values ​​are low, it indicates that the data-driven load prediction sequence has low reliability and the theoretical baseline load sequence deviates significantly from real-time operating conditions. In this case, the theoretical baseline load sequence is assigned a higher weight, and the data-driven load prediction sequence a lower weight. When the two quantification values ​​are at an intermediate level, the corresponding weights are allocated according to their specific numerical ratios, ultimately forming a clear weight allocation relationship. This relationship is the association rule that determines the proportion of each type of sequence in the fusion process.

[0048] When normalizing the weight allocation relationship, based on the established weight allocation relationship, the weight values ​​of the theoretical baseline load sequence and the data-driven load prediction sequence are adjusted to a range where the sum is 1. This ensures that the weight ratio of the two types of sequences is accurate and meets the basic requirements of fusion calculation. After adjustment, a fusion weight vector is obtained. This vector is a numerical combination containing the weights of the two types of sequences, and the sum of the two weight values ​​is 1, clearly reflecting the proportion of the two in the fusion process.

[0049] When performing weighted fusion based on the fusion weight vector, the load value at each time node in the theoretical baseline load sequence and the data-driven load prediction sequence are used as the basis. At the same time, the load value at each time node in the data-driven load prediction sequence is multiplied by the corresponding weight. Then, the results of the two multiplications at each time node are added together to obtain the fused load value at that time node. Finally, the fused load values ​​of all time nodes are arranged in chronological order to form the initial fused load sequence. This sequence is a set of load values ​​that combines the advantages of the two types of sequences, and each time node has a corresponding fused load value.

[0050] When performing continuous smoothing on the initial fused load sequence, the initial fused load sequence is used as the processing object. The load value changes of adjacent time nodes in the sequence are checked. For data points with large changes or abrupt changes, the load values ​​before and after them are referenced for transitional adjustments to keep the load value changes stable and continuous, avoiding abrupt fluctuations. After adjustment, the load prediction sequence is obtained. This sequence is a stable and continuous set of load values ​​arranged in chronological order, which can accurately reflect the load change trend of the target building under actual working conditions.

[0051] The load sequence fusion module provides a reliable basis for data-driven load forecast sequences to ensure the weights match actual reliability, avoids unreliable sequences from excessively affecting the accuracy of fusion results, clarifies the adaptation deviation between the theoretical benchmark load sequence and real-time operating conditions, provides an adaptability basis for weight allocation, ensures that the weights conform to actual operating conditions, unifies the numerical ranges of real-time confidence indicators and operating condition deviation indicators, ensures their comparability, avoids weight judgment deviations due to differences in numerical ranges, improves the rationality of weight allocation, integrates the reliability and adaptability of the two types of sequences, formulates scientific weight rules, highlights the role of advantageous sequences, makes the fusion results more in line with actual needs, standardizes the numerical range of weights, ensures the accuracy of the weight ratio of the two types of sequences, provides a precise basis for weighted fusion, ensures the accuracy of fusion calculation, combines the characteristics of the two types of sequences to complement each other, retains theoretical rationality and actual adaptability, forms high-quality initial fusion results, lays the foundation for the final load forecast sequence, eliminates data mutations in the initial fused load sequence, ensures the sequence is stable and continuous, makes the forecast results conform to the actual change law of building load, provides reliable data support for flexible control strategies, and improves the practicality of the strategy.

[0052] The flexible strategy mapping module is used to perform multi-objective flexible strategy mapping on the load forecast sequence to obtain flexible candidate strategies for the target building. Multi-objective reconstruction of the load forecast sequence yields multi-dimensional optimization objectives for the target building. Based on multi-dimensional optimization objectives, the load forecast sequence is deconstructed according to time-series features to obtain the load dynamic characteristics of the target building; By associating load dynamic characteristics and multi-dimensional optimization objectives with strategy orchestration, a preliminary flexible control scheme for the target building is obtained. Based on multi-dimensional optimization objectives, a comprehensive performance evaluation of the preliminary flexible control scheme is conducted to obtain flexible candidate strategies for the target building.

[0053] When reconstructing a load forecast sequence for multiple objectives, it is necessary to use the already obtained load forecast sequence as a basis, combined with the core needs of building operation, including energy conservation, operating cost control, and ensuring indoor environmental comfort, to decompose and redefine the load changes reflected by the load forecast sequence, clarify the specific direction of each objective, and finally form a multi-dimensional optimization objective. This objective is a set of multiple optimization directions that are closely related to the actual operating needs of the building and can be implemented. Each optimization direction can correspond to a specific time period or load characteristic in the load forecast sequence.

[0054] When deconstructing the time-series features of the load forecast sequence based on multi-dimensional optimization objectives, it is necessary to rely on the determined multi-dimensional optimization objectives and the load forecast sequence, and analyze the numerical changes in the load forecast sequence in time sequence. Combined with the specific requirements of each optimization objective for the load in different time periods, key information such as the load change trend, change magnitude, peak occurrence time, and valley duration are extracted in different time stages. After these information are organized according to the time dimension, they form the load dynamic features. These features are a set of key information reflecting the load change pattern, and each feature can directly correspond to the realization requirements of the multi-dimensional optimization objectives.

[0055] When orchestrating strategies that correlate load dynamics and multi-dimensional optimization objectives, it is necessary to take these as a premise, accurately match each load dynamic characteristic with its corresponding optimization objective, and design specific control actions and execution logic for different load changes and optimization needs. For example, when the load shows a rapid upward trend and the optimization objective is to control energy consumption, control actions to reduce the operation of unnecessary energy-consuming equipment are designed. When the load is in a stable range and the optimization objective is to ensure comfort, control actions to maintain the current operating parameters are designed. Then, all targeted control actions are organized and integrated according to time sequence and logical relationship to form a preliminary flexible control scheme. This scheme contains a set of control actions executed in a specific order for different load scenarios, and each control action clearly corresponds to a specific load dynamic characteristic and optimization objective.

[0056] When evaluating the comprehensive effectiveness of a preliminary flexible control scheme based on multi-dimensional optimization objectives, it is necessary to analyze each control action in the preliminary flexible control scheme one by one, based on the preliminary flexible control scheme and the multi-dimensional optimization objectives, to determine its actual effect in achieving the corresponding optimization objective, and to assess whether the control action can effectively achieve the objective without affecting the achievement of other optimization objectives. For example, can a certain control action reduce energy consumption while ensuring indoor comfort, or can it control costs while meeting load operation requirements? Subsequently, control actions that cannot achieve the optimization objective or have a negative impact on other objectives are eliminated, and control actions that achieve the target are retained and integrated. Finally, flexible candidate strategies are obtained. These strategies are a set of control schemes that have been screened and can effectively achieve multi-dimensional optimization objectives. Each strategy has a clear execution logic and expected effect.

[0057] The flexible strategy mapping module provides clear guidance for subsequent work by defining multi-dimensional optimization objectives, aligning with the actual operational needs of buildings, ensuring the practicality of control strategies, accurately extracting dynamic characteristics of load time series, making the matching of characteristics with optimization objectives more targeted, providing a solid basis for the arrangement of related strategies, establishing a precise correlation between load characteristics and optimization objectives, forming a systematic and orderly preliminary control plan, laying the foundation for subsequent performance evaluation, screening effective control plans, eliminating ineffective or negatively impactful actions, improving the effectiveness and feasibility of strategies, and providing high-quality alternatives for subsequent selection.

[0058] The strategy optimization coding module is used to perform performance optimization coding on candidate flexible strategies based on mechanistic digital twins and data-driven digital twins to obtain the flexible control instructions for the target building. Based on the mechanistic digital twin, the mechanistic consistency of candidate flexible strategies is evaluated, and the mechanistic compliance sequence of candidate flexible strategies is obtained. Based on the data-driven digital twin, the historical performance of candidate flexible strategies is predicted to obtain the data-driven performance sequence of candidate flexible strategies. The fusion performance evaluation of the mechanism compliance sequence and the data-driven performance sequence is carried out to obtain the comprehensive performance score of the candidate flexible strategy; Based on the comprehensive performance score, the candidate flexible strategies are ranked and selected to obtain the optimal flexible strategy identifier for the target building. The optimal flexibility strategy identifier is encoded as the flexibility control command for the target building.

[0059] Mechanistic digital twins are virtual models built based on target building design parameters and physical laws, capable of replicating building geometry, material thermal properties, and energy transfer processes. Candidate flexible strategies are multiple potential control schemes output by the flexible strategy mapping module. During the evaluation process, for each control link of each candidate flexible strategy, the physical laws and building operation mechanisms followed by the mechanistic digital twin are compared one by one to determine whether the link is reasonable and feasible and whether there is any violation of the relevant mechanisms. The mechanism compliance sequence is a sequence formed by arranging the evaluation results of each candidate flexible strategy in order. Each sequence item corresponds to a candidate flexible strategy, clarifying the compliance of the strategy with the mechanism at each control link, including specific situations such as complete compliance and basic compliance.

[0060] A data-driven digital twin is a virtual model built on the target building's historical operating data and real-time input data. It can replicate the building's actual operating characteristics and reflect real-time dynamic changes. Historical operating performance prediction uses the building's historical operating patterns and dynamic response patterns stored in the data-driven digital twin to simulate the effects of each candidate flexible strategy after implementation in similar past operating scenarios. It analyzes the energy consumption, load regulation effects, and other performance-related factors that the strategy may bring. The data-driven performance sequence is a sequence formed by arranging the historical operating performance prediction results of each candidate flexible strategy in order. Each sequence item corresponds to a candidate flexible strategy and includes the specific details of various operating performance indicators of the strategy in the simulated historical scenario, such as the expected energy savings and the accuracy of load regulation.

[0061] The integrated performance evaluation combines the assessment and prediction results of the same candidate flexible strategy in two sequences to conduct a comprehensive evaluation of its effectiveness. During the evaluation process, it considers both the degree to which the strategy conforms to the building mechanism, ensuring its physical feasibility, and its operational effectiveness in similar historical scenarios, ensuring that the strategy achieves the expected control objectives. By comprehensively weighing these two aspects, a comprehensive evaluation result is obtained. The integrated performance score is a specific score given to each candidate flexible strategy based on the integrated performance evaluation results. The score directly reflects the overall quality of the candidate flexible strategy; a higher score indicates that the strategy better conforms to the mechanism requirements and has better operational effectiveness.

[0062] The optimal ranking is to arrange all candidate flexible strategies in descending order of comprehensive performance score. The candidate flexible strategy with the highest score is the optimal flexible strategy. The optimal flexible strategy identifier is the information used to uniquely identify the optimal flexible strategy. It may be a specific number or name. The corresponding optimal flexible strategy can be quickly located through this identifier.

[0063] Encoding is the process of converting the optimal flexibility strategy identifier into an instruction form that the building control system can recognize and execute. This instruction form conforms to the signal transmission and execution specifications of the control system, ensuring that the control system accurately understands the optimal flexibility strategy corresponding to the identifier. The flexible control instruction is the final result after encoding. It is an instruction that the building control system can directly execute, used to control the relevant equipment in the building to operate according to the optimal flexibility strategy, thereby achieving flexible regulation of the building load.

[0064] The strategy optimization coding module ensures the feasibility of strategy mechanisms, avoids unreasonable solutions, provides reliable support for selection, ensures smooth implementation, predicts the actual application effectiveness of strategies, provides quantitative references, improves the accuracy of strategy selection, ensures the achievement of control and energy optimization goals, comprehensively considers the feasibility of mechanisms and operational efficiency, avoids decision-making biases, provides a scientific basis for optimization ranking, ensures the quality of selected strategies, quickly screens the comprehensive optimal strategy, clarifies selection objectives, improves the efficiency and accuracy of strategy selection, transforms the optimal strategy into executable instructions, builds a bridge between strategies and control actions, and ensures the smooth implementation of flexible control.

[0065] Example 2: In the load sequence fusion module, when dynamically weighting the theoretical baseline load sequence and the data-driven load prediction sequence based on real-time confidence indicators and operating condition deviation indicators, the module also includes operating condition scenario classification and dynamic calibration of indicator sensitivity steps, specifically: Based on the real-time meteorological parameters, building operating load level, and energy supply status of the target building, a three-dimensional operating condition scenario classification model is constructed. The input parameters of the three-dimensional operating condition scenario classification model include: the difference between the real-time ambient temperature and the standard operating temperature ΔT, the ratio of the actual operating load of the building to the rated load K, and the energy supply stability coefficient S (S = actual supply fluctuation variance / preset allowable fluctuation variance). The operating conditions are divided into three categories: normal operating conditions, extreme operating conditions, and transitional operating conditions according to preset thresholds. Configure corresponding sensitivity coefficients for different operating conditions: Sensitivity coefficients for real-time confidence indicators under normal operating conditions. Sensitivity coefficient of operating condition deviation index Under extreme working conditions, , Under transitional operating conditions, , ; The confidence and offset metrics are calibrated based on the sensitivity coefficient, using the following calibration formula: Post-calibration confidence quantification ; Calibrated offset quantization ; In the formula, C is the original confidence metric, and D is the original offset metric. The absolute value of the temperature difference. The standardized temperature reference value is set at 25℃. Based on the post-calibration confidence quantification Quantized value of offset after calibration The weight allocation relationship is updated using a weighted summation method. The updated weight allocation formula is as follows: Data-driven sequence weights ; Theoretical benchmark sequence weights ; Based on the updated weight allocation relationship, normalization and weighted fusion are performed to obtain the initial fused load sequence of the target building.

[0066] Under extreme operating conditions (such as high temperature / cold wave), the sensitivity of the operating condition deviation index is increased, the weight of the theoretical benchmark sequence is strengthened, and the prediction deviation caused by insufficient extreme data in the data-driven sequence is avoided. Under transitional operating conditions (such as seasonal change), the influence of the two types of indicators is balanced to adapt to the uncertainty of load fluctuations. The energy supply stability coefficient S is introduced to incorporate the energy side status into the weight adjustment, so that the fused sequence is more in line with the actual energy dispatch needs.

[0067] Example 3: In the data-driven load forecasting module, a building area load heterogeneity calibration mechanism is introduced, specifically as follows: First, based on the geometric topology and functional zoning design of the target building, the building is divided into multiple load-independent areas (such as office area, computer room area, conference room area, etc.), and the area ratio of each area is calculated. For each area, its historical operating data is extracted separately to form a regional feature sequence. Based on this, the region-specific historical prediction deviation dynamic weighted average, standard deviation, and actual load moving average are recalculated. Furthermore, the region-specific uncertainty suppression coefficient, saturation adjustment parameter, and standard deviation correction coefficient are obtained through training with the region's historical data. In dynamic load forecasting, the real-time operating status of each region and local meteorological microenvironment data are combined as regional real-time forecasting features. The initial load value of the region is obtained by inputting a data-driven digital twin, and then the calibrated load value of the region is calculated by using a region-specific calibration formula. Subsequently, based on the area proportion of each region, the calibrated load values ​​of all regions are weighted and aggregated to obtain the overall data-driven load prediction sequence for the building. Simultaneously, the load forecast deviation of each region is monitored in real time. When the deviation exceeds the preset threshold, the exclusive calibration parameters of the corresponding region are dynamically updated according to the simulation cycle, realizing refined load forecasting and dynamic self-adaptive calibration of the region. This effectively solves the problem that a single parameter cannot adapt to the load characteristics of different regions and that the local micro-environmental differences cause prediction distortion, and significantly improves the accuracy and adaptability of the data-driven load forecast sequence.

[0068] Example 4: A building load prediction and flexible control device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The processor executes the program to implement the system in Example 1, Example 2, or Example 3.

Claims

1. A building load prediction and flexible control system, characterized in that, include: The mechanistic twin benchmark simulation module is used to construct a mechanistic digital twin of the target building based on the design parameters and physical laws of the target building, and to perform standard working condition simulation on the mechanistic digital twin to obtain the theoretical benchmark load sequence of the target building. The data-driven load forecasting module is used to construct a data-driven digital twin of the target building based on its historical operating data and real-time input data, and to perform dynamic load forecasting on the data-driven digital twin to obtain the data-driven load forecasting sequence of the target building. The load sequence fusion module is used to adaptively fuse the theoretical baseline load sequence and the data-driven load forecast sequence to obtain the load forecast sequence of the target building. The flexible strategy mapping module is used to perform multi-objective flexible strategy mapping on the load forecast sequence to obtain flexible candidate strategies for the target building. The strategy optimization coding module is used to perform performance optimization coding on candidate flexible strategies based on mechanistic digital twins and data-driven digital twins to obtain flexible control instructions for the target building.

2. The building load prediction and flexible control system as described in claim 1, characterized in that, In the mechanism twin benchmark simulation module, the geometric information of the target building is topologically processed according to the design parameters of the target building to obtain the geometric topology of the target building; Based on the geometric topology, the material properties in the design parameters are identified item by item to obtain the material property entries of the target building. Based on the structural hierarchy description in the design parameters, the material property items are hierarchically assigned to obtain the thermal property list of the target building. The heat exchange logic between the interior space of the target building and the external environment is defined by rules, resulting in the energy transfer rule set of the target building. By using a thermal property list and an energy transfer rule set, dynamic energy interaction is performed on the geometric topology to obtain a mechanistic digital twin of the target building. Standard operating condition simulations were performed on the mechanistic digital twin to obtain the theoretical baseline load sequence of the target building.

3. The building load prediction and flexible control system as described in claim 2, characterized in that, In the mechanistic twin baseline simulation module, standard load condition simulation is performed on the mechanistic digital twin to obtain the theoretical baseline load sequence of the target building, specifically: Based on pre-defined building codes and climate zones, standard operating conditions for the mechanism-based digital twin are set. Based on standard operating conditions, the mechanistic digital twin is dynamically simulated according to a preset simulation cycle to obtain the building load dataset within the simulation cycle. The theoretical baseline load sequence of the target building is obtained by performing time-series smoothing aggregation on the building load dataset.

4. The building load prediction and flexible control system as described in claim 1, characterized in that, In the data-driven load forecasting module, the historical operating data of the target building is analyzed to obtain the historical feature sequence of the target building; By performing operational mode decomposition on the historical feature sequence, the steady-state operation law and dynamic response law of the target building are obtained; By performing aggregated analysis on the steady-state operation law and the dynamic response law, the twin mapping relationship of the target building is obtained; By injecting real-time input data of the target building into the twin mapping relationship, the dynamic evolution relationship of the target building can be obtained; Virtual entities are constructed based on dynamic evolutionary relationships to obtain a data-driven digital twin of the target building; Dynamic load forecasting is performed on the data-driven digital twin to obtain the data-driven load forecast sequence for the target building.

5. A building load prediction and flexible control system as described in claim 4, characterized in that, In the data-driven load forecasting module, dynamic load forecasting is performed on the data-driven digital twin to obtain the data-driven load forecasting sequence for the target building, specifically: Obtain the real-time operating status and future weather forecast of the target building to obtain the real-time predictive characteristics of the target building; By inputting real-time forecast features into a data-driven digital twin, preliminary load values ​​for the target building are obtained. Based on historical forecast deviations, the initial load value is reliably calibrated to obtain the calibrated load value for the target building. The formula for calculating the calibrated load value is as follows: ； In the formula, The calibrated load value obtained at time t, The initial load value obtained at time t, The dynamic weighted average of the historical prediction biases at time t. Let be the standard deviation obtained at time t based on historical prediction bias. The moving average of the actual load at time t, obtained from historical operating data. The preset uncertainty suppression coefficient, The preset saturation adjustment parameters are: The standard deviation correction factor is the preset value. For symbolic functions, It is an exponential decay factor. The dynamic weighted average of historical prediction bias The absolute value; The calibrated load values ​​are serialized and normalized to obtain the data-driven load prediction sequence for the target building.

6. The building load prediction and flexible control system as described in claim 1, characterized in that, In the load sequence fusion module, the process of obtaining the load forecast sequence for the target building is as follows: The confidence level of the data-driven load forecast sequence is quantified to obtain the real-time confidence index of the data-driven load forecast sequence; Based on the real-time operating condition information of the target building, the fit deviation of the theoretical reference load sequence is evaluated to obtain the operating condition deviation index of the theoretical reference load sequence. Based on real-time confidence index and load condition deviation index, the theoretical baseline load sequence and the data-driven load prediction sequence are dynamically weighted and matched to obtain the initial fused load sequence of the target building. The initial fused load sequence is continuously smoothed to obtain the load prediction sequence for the target building.

7. A building load prediction and flexible control system as described in claim 6, characterized in that, In the load sequence fusion module, based on real-time confidence indices and load condition deviation indices, the theoretical baseline load sequence and the data-driven load forecast sequence are dynamically weighted and matched to obtain the initial fused load sequence for the target building, specifically: The real-time confidence index and the operating condition deviation index are normalized to obtain the quantitative values ​​of the confidence index and the quantitative values ​​of the deviation of the target building. Based on the confidence metric and the offset metric, the weighting relationship between the theoretical baseline load series and the data-driven load forecast series is jointly determined; The weight allocation relationship is normalized to obtain the fused weight vector of the weight allocation relationship; Based on the fusion weight vector, the theoretical baseline load sequence and the data-driven load prediction sequence are weighted and fused to obtain the initial fused load sequence of the target building.

8. A building load prediction and flexible control system as described in claim 1, characterized in that, In the flexible strategy mapping module, the process of obtaining the flexible candidate strategy for the target building is as follows: Multi-objective reconstruction of the load forecast sequence yields multi-dimensional optimization objectives for the target building. Based on multi-dimensional optimization objectives, the load forecast sequence is deconstructed according to time-series features to obtain the load dynamic characteristics of the target building; By associating load dynamic characteristics and multi-dimensional optimization objectives with strategy orchestration, a preliminary flexible control scheme for the target building is obtained. Based on multi-dimensional optimization objectives, a comprehensive performance evaluation of the preliminary flexible control scheme is conducted to obtain flexible candidate strategies for the target building.

9. A building load prediction and flexible control system as described in claim 1, characterized in that, In the strategy optimization coding module, the process of obtaining the flexible control instructions for the target building is as follows: Based on the mechanistic digital twin, the mechanistic consistency of candidate flexible strategies is evaluated, and the mechanistic compliance sequence of candidate flexible strategies is obtained. Based on the data-driven digital twin, the historical performance of candidate flexible strategies is predicted to obtain the data-driven performance sequence of candidate flexible strategies. The fusion performance evaluation of the mechanism compliance sequence and the data-driven performance sequence is carried out to obtain the comprehensive performance score of the candidate flexible strategy; Based on the comprehensive performance score, the candidate flexible strategies are ranked and selected to obtain the optimal flexible strategy identifier for the target building. The optimal flexibility strategy identifier is encoded as the flexibility control command for the target building.

10. A building load prediction and flexible control device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the system according to any one of claims 1-9.

Images