A public building virtual power plant adjustable load evaluation method, system and storage medium

Abstract

This invention discloses a method, system, and storage medium for assessing the adjustable load of a virtual power plant in public buildings. The method includes: acquiring raw influencing factor data, including at least air conditioning system operating parameters reflecting equipment aging and building thermal parameters reflecting building thermophysical characteristics; constructing hysteresis features characterizing thermal inertia based on the air conditioning system operating parameters, and constructing multi-parameter coupling features based on their coupling relationship with building thermal parameters; inputting the obtained data into a pre-trained adjustable load assessment model to obtain the adjustable load assessment result; wherein the model is a multi-model fusion model and integrates a rolling mixed-integer linear programming solver for synchronously generating a full-chain control scheme for the air conditioning system. This invention, by introducing a quantitative expression of the thermal inertia of the air conditioning system and multi-parameter coupling features, achieves an integrated closed loop of assessment and control, significantly improving the response accuracy and overall benefits of public buildings participating in virtual power plants.

Description

Technical Field

[0001] This invention relates to the field of virtual power plant technology, and in particular to a method, system and storage medium for assessing the adjustable load of a public building virtual power plant. Background Technology

[0002] With the accelerated construction of new power systems, virtual power plants, as the core carrier for integrating distributed energy and improving the flexibility of power grid dispatch, are seeing their participants gradually expand from industrial loads to public buildings (hereinafter referred to as "public buildings"). Public buildings, characterized by large total load and diverse types of adjustable equipment, are important adjustable load resources for virtual power plants.

[0003] However, existing methods for assessing adjustable loads in public buildings have the following technical shortcomings:

[0004] First, there is a lack of detailed characterization of the aging degree of air conditioning system equipment. Existing methods mostly use static equipment parameters and do not incorporate parameters that reflect the aging state of the equipment, such as the service life of the air conditioner and the performance degradation coefficient, resulting in a large deviation in the assessment of the actual adjustability of old equipment.

[0005] Second, the impact of building thermophysical properties on adjustable loads is not considered. Traditional assessments neglect thermal parameters such as building external insulation parameters and building age, making it impossible to accurately quantify the impact of the building itself on air conditioning load transmission and losses, resulting in assessment results that are out of sync with physical reality.

[0006] Third, there is a lack of modeling for the thermal inertia and multi-parameter coupling effects of air conditioning systems. Conventional methods do not construct hysteresis characteristics (such as previous period heat load) to characterize thermal inertia, nor do they construct coupling characteristics between meteorological parameters and building and equipment parameters (such as the product of cold and hot temperature difference and external insulation coefficient), resulting in the evaluation model being unable to capture the dynamic response characteristics and multi-factor synergistic effects of the air conditioning system.

[0007] Fourth, the assessment model and the generation of control schemes are disconnected. Existing methods only output adjustable load capacity and cannot simultaneously generate a complete chain of control schemes for the main unit, auxiliary units, and terminals of the air conditioning system. This makes it difficult for the assessment results to directly guide actual operation and meet the real-time dispatching requirements of virtual power plants. Summary of the Invention

[0008] To address the problems in existing technologies regarding the assessment of adjustable loads in public building air conditioning systems, such as the lack of detailed characterization of equipment aging and building thermophysical characteristics, failure to consider thermal inertia and multi-parameter coupling effects, and the disconnect between assessment and control, this invention proposes a method, system, and storage medium for assessing adjustable loads in a virtual power plant for public buildings.

[0009] The specific technical solution is as follows: A method for assessing the adjustable load of a public virtual power plant, including:

[0010] Obtain the original impact factor data of public buildings participating in the virtual power plant within the target period. The original impact factor data includes at least the air conditioning system operating parameters reflecting the degree of equipment aging and the building thermal parameters reflecting the building thermophysical properties.

[0011] Based on the operating parameters of the air conditioning system, a hysteresis feature characterizing the thermal inertia of the air conditioning system is constructed. Based on the coupling relationship between the operating parameters of the air conditioning system and the building thermal parameters, a multi-parameter coupling feature is constructed to obtain the data of the first main influencing factor.

[0012] Input the data of the first main influencing factor into the pre-trained adjustable load assessment model to obtain the adjustable load assessment results within the target period;

[0013] The adjustable load assessment model is a multi-model fusion model, integrating a rolling mixed-integer linear programming solver to synchronously generate full-chain control schemes for the air conditioning system. By introducing air conditioning system operating parameters reflecting equipment aging and building thermal parameters reflecting building thermophysical properties, hysteresis characteristics and multi-parameter coupling characteristics characterizing thermal inertia are constructed, achieving a quantitative expression of the dynamic response characteristics and multi-factor synergistic effects of the air conditioning system. Simultaneously, by integrating the rolling mixed-integer linear programming solver, the assessment results are directly converted into executable control schemes, solving the problem of traditional separation between assessment and control, and significantly improving the response accuracy and operability of public buildings participating in virtual power plants.

[0014] Furthermore, the hysteresis characteristic includes at least one of the following: heat load in the previous 15 minutes, average heat load in the previous hour, and heat load trend in the previous 2 hours;

[0015] The multi-parameter coupling features include at least one of the following: the product of the cold and hot temperature difference and the building's external insulation correction coefficient; the product of solar irradiance and the air conditioning performance attenuation coefficient; and the product of time-period characteristics and building occupancy rate. By constructing lag features across multiple time scales, the continuous impact of the air conditioning system's thermal inertia on adjustable loads was quantified. By constructing coupling features between meteorological parameters and building and equipment parameters, the nonlinear influence mechanism of the synergistic effect of multiple factors on adjustable loads was revealed, further improving the assessment accuracy.

[0016] Furthermore, the adjustable load assessment model is constructed through the following steps:

[0017] Train the XGBoost sub-model, LSTM sub-model, and GraphSAGE sub-model separately;

[0018] The outputs of the XGBoost, LSTM, and GraphSAGE sub-models are fused using adaptive weights, which are dynamically adjusted based on real-time scene features. By integrating the nonlinearity capture capability of XGBoost, the temporal dependency modeling capability of LSTM, and the multi-parameter coupling relationship mining capability of GraphSAGE, a complementary multi-model fusion architecture is constructed. Through adaptive weight adjustment based on real-time scene features, the model can dynamically optimize the contribution of each sub-model according to different weather conditions, equipment status, and other scenarios, significantly improving the generalization ability and robustness of the evaluation model under different operating conditions.

[0019] Furthermore, the application scenarios include meteorological conditions and the service life of the air conditioner;

[0020] In extreme weather scenarios, increase the weights of the LSTM sub-model;

[0021] In scenarios where the air conditioner's service life exceeds a preset threshold, the weight of the GraphSAGE sub-model is increased.

[0022] Furthermore, it also includes:

[0023] Multidimensional dynamic constraints are embedded as boundary conditions for model inference into the adjustable load assessment model.

[0024] The multi-dimensional dynamic constraints include equipment constraints, building constraints, time constraints, and comfort constraints;

[0025] The equipment constraints include a rule for narrowing the adjustment range based on the service life of the air conditioner: when the service life of the air conditioner exceeds 10 years, the adjustment range is narrowed to 40%-90% of the rated adjustment range;

[0026] The building constraints include adjustment rate correction rules based on building age and external insulation integrity rate: when the building age exceeds 15 years and the external insulation integrity rate is less than 70%, the adjustment rate is reduced by 20%;

[0027] The time constraints include adjustment rules for the lower limit of adjustment capacity based on weekday and time period characteristics: during weekday daytime hours, the lower limit of adjustment capacity is increased by 10%;

[0028] The comfort constraints include an indoor temperature fluctuation range of ±1℃ and a humidity range of 40%-60%. By directly embedding multi-dimensional constraints such as equipment physical limitations, building thermal performance, energy consumption time characteristics, and indoor comfort requirements as boundary conditions for model inference, rather than post-processing trimming, the feasibility of the evaluation results in the physical world is ensured. Furthermore, by introducing quantified specific constraint rules, the model output strictly conforms to actual operational boundaries, avoiding the generation of infeasible solutions and significantly improving the reliability and practicality of the evaluation results.

[0029] Furthermore, it also includes:

[0030] Based on real-time heat load forecasts, the entire control scheme of the air conditioning system is updated every 15 minutes using the latest heat load forecasts and real-time operating data through the rolling mixed-integer linear programming solver. This rolling update mechanism enables the control scheme to respond in real-time to changes in heat load and system state fluctuations, achieving continuous optimization under dynamic conditions. The rolling optimization strategy effectively addresses prediction errors and disturbances, ensuring the adaptability and optimality of the control scheme in all-weather operation.

[0031] A public building virtual power plant adjustable load assessment system includes:

[0032] The acquisition module is used to acquire the original impact factor data of public buildings participating in the virtual power plant within the target period. The original impact factor data includes at least the air conditioning system operating parameters and building thermal parameters.

[0033] The feature construction module is used to construct hysteresis features characterizing the thermal inertia of the air conditioning system based on the operating parameters of the air conditioning system, and to construct multi-parameter coupled features based on the coupling relationship between the operating parameters of the air conditioning system and the building thermal parameters, so as to obtain the first main influencing factor data.

[0034] The assessment and control module is used to load the pre-trained adjustable load assessment model, receive the first main influencing factor data output by the feature construction module and input it into the model to obtain the adjustable load assessment results within the target period, and simultaneously generate a full-chain control scheme for the air conditioning system.

[0035] The adjustable load assessment model is a multi-model fusion model and integrates a rolling mixed integer linear programming solver.

[0036] Furthermore, the evaluation and control module is also used to embed multi-dimensional dynamic constraints as boundary conditions for model inference into the adjustable load evaluation model;

[0037] The multi-dimensional dynamic constraints include equipment constraints, building constraints, time constraints, and comfort constraints.

[0038] Furthermore, the evaluation and control module adopts a combination architecture of FPGA and GPU acceleration card, and the generation time of a single evaluation and control scheme does not exceed 1 minute;

[0039] The data acquisition delay of the acquisition module shall not exceed 10 seconds.

[0040] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0041] The above technical solution has the following advantages or technical effects:

[0042] 1. This invention solves the technical problem of insufficient physical mechanism characterization in traditional assessments by constructing a refined data acquisition system for the aging degree of air conditioning system equipment and the thermophysical characteristics of buildings, combined with the quantitative expression of the hysteresis characteristics and multi-parameter coupling characteristics that characterize thermal inertia, and significantly improves the accuracy and reliability of adjustable load assessment.

[0043] 2. This invention achieves complementary advantages of XGBoost, LSTM, and GraphSAGE models through a multi-model fusion architecture and scenario-based adaptive weight adjustment. Furthermore, through embedded modeling with multi-dimensional dynamic constraints, it ensures that the evaluation results strictly conform to the actual operating boundaries, significantly improving the model's generalization ability and the physical feasibility of the evaluation results in different public building scenarios.

[0044] 3. This invention achieves the synchronous transformation of evaluation results into a full-chain control scheme for the air conditioning system's main unit, auxiliary units, and terminal units by integrating a rolling mixed-integer linear programming solver. Through a 15-minute rolling update mechanism, it forms an integrated closed loop of "prediction-evaluation-control", solving the pain point of traditional evaluation and control being separated.

[0045] 4. This invention, through hardware implementation using a heterogeneous acceleration architecture of FPGA and GPU, reduces the time required for generating a single evaluation and control scheme to less than 1 minute, thus meeting the real-time scheduling requirements of virtual power plants. Through modular system design and software implementation using readable storage media, it provides complete support for the engineering implementation and promotion of the technical solution, taking into account both the scheduling benefits of virtual power plants and the operational benefits of public facilities themselves. Attached Figure Description

[0046] Figure 1 This is a flowchart of the method of the present invention;

[0047] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0048] Figure 3 This is a schematic diagram of the structure of the electronic device of the present invention. Detailed Implementation

[0049] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Example 1

[0051] A method for assessing the adjustable load of a virtual power plant in a public building. This embodiment uses a commercial complex in a core urban business district (building type: commercial complex; building orientation: due south; building age: 8 years; external insulation: polyurethane material + 50mm thickness + 92% integrity rate; adjustable load includes central air conditioning system and 20 commercial elevators) participating in a regional virtual power plant peak-shaving task as an example to illustrate the method of this invention in detail. Figure 1 As shown, the method includes the following steps:

[0052] Step S1: Obtain raw impact factor data for the target period.

[0053] The target period is set at monthly (30 days), with daily evaluation during the grid peak load period from 17:00 to 21:00. The collected raw influencing factor data must cover multi-dimensional information related to adjustable loads in the scenario of public-construction participation in a virtual power plant, with a focus on strengthening the collection of specific data on air conditioning systems, including:

[0054] Basic building information: Building type (commercial complex), total building area of ​​80,000 square meters, building orientation due south, number of floors 12, building age of 8 years, building envelope parameters, building external insulation parameters (polyurethane material, thickness 50mm, integrity rate 92%, used to calculate building external insulation correction factor 0.98).

[0055] Load equipment parameters: Adjustable load types include central air conditioning systems and elevators. The central air conditioning system is a water-cooled screw chiller unit with a rated power of 2000kW, an adjustment threshold of 30%-100% of the rated power, a COP value of 4.2, a service life of 5 years, and a performance degradation coefficient of 0.9 (calculated based on 2% degradation per year). The chilled water pump has a rated frequency of 50Hz and a frequency conversion range of 30Hz-50Hz. The elevator is a permanent magnet synchronous traction type with a single unit rated power of 15kW, and an adjustment threshold of three levels: "off / low power standby / full load operation". Simultaneously, specific parameters of the air conditioning system are collected: chiller unit power-cooling capacity-cooling water inlet temperature correlation parameters; chilled water pump frequency-power-head correlation coefficient; air conditioning service life accurate to the year; heat exchange area of ​​terminal equipment; and valve adjustment characteristics.

[0056] Operating status data: Real-time power of central air conditioning, output cooling capacity, supply and return water temperature difference; terminal flow deviation (calculated by formula δQ=(Q meas Q design ) / Q design ×100%, the pass rate requirement is ≤±10%; pipeline heat loss rate (calculated using the formula LossRate=(Q plant ΣQ terminal ) / Q plant ×100%, the pass rate requirement is ≤3%; Orientation difference coefficient (ΔT)orient =T i T avg ); compressor operating frequency, chilled water / cooling water supply and return temperature difference; chilled water pump real-time frequency and operating power; indoor Concentration is used to calculate occupancy rate; equipment runtime, current operating mode (automatic / manual); historical adjustment response records (including number of successful responses, number of failures, average response delay, etc.). Among these, Q... meas To measure the terminal flow rate, Q design To design the end-point flow, Q plant The cooling capacity output by the air conditioning unit, ΣQ terminal T is the sum of the cooling capacity of each terminal unit. i T represents the indoor temperature of a room facing a certain direction. avg This represents the average indoor temperature of the building.

[0057] Environmental data: Outdoor dry-bulb temperature, relative humidity, direct / diffuse solar irradiance, wind speed, and temperature difference (the difference between outdoor temperature and the indoor set temperature of 26℃) from 17:00 to 21:00 daily during the target period. The data are from the meteorological monitoring station on the roof of the building, with a sampling interval of 15 minutes.

[0058] Energy demand constraints: During business hours (10:00-22:00), commercial complexes must ensure basic lighting and elevator services. The comfort threshold requires indoor temperature to be controlled at 24-26℃, humidity at 40%-60%, and temperature fluctuation within ±1℃. The air conditioning system must operate continuously.

[0059] Virtual power plant dispatch requirements: The response time for this peak shaving task is required to be ≤5 minutes, the adjustment capacity range is 300-600kW, the continuous adjustment duration is 2 hours, and the dispatch compensation standard is 0.5 yuan / kWh.

[0060] In the above data, the service life and performance degradation coefficient of the air conditioner are operating parameters of the air conditioning system that reflect the degree of equipment aging; the external insulation parameters of the building, the building's age, and the building's orientation are building thermal parameters that reflect the building's thermophysical characteristics.

[0061] Step S2: Data Preprocessing and Feature Extraction

[0062] The raw impact factor data obtained in step S1 is preprocessed and features are extracted, with a focus on enhancing the feature engineering of the air conditioning system, to obtain the first main impact factor data. The specific process is as follows:

[0063] S21: Data Preprocessing

[0064] Data cleaning: Remove duplicate and invalid data. For example, remove terminal flow data where the flow deviation exceeds ±10% of the design deviation during air conditioning heat load collection, and remove abnormal readings where the network heat loss rate exceeds 3%; remove power surge data when the air conditioner has been in use for more than 10 years, and correct for this using a performance degradation coefficient. Additionally, periods in "manual" mode are marked as unadjustable and excluded from subsequent evaluations.

[0065] Missing value handling: Mean imputation, interpolation, or fitting based on similar building data were used. For three days of missing solar radiation data in the meteorological data, linear interpolation was used for imputation. For one day of missing building external insulation integrity rate data, the 93% mean of commercial complexes of the same era and type was used for imputation. For missing performance parameters of the air conditioning system, the mean of the same model of equipment plus correction for service life was used for imputation.

[0066] Outlier handling: Outlier data is identified and corrected using the 3σ principle and box plot method. For example, outliers in the central air conditioning power consumption data that exceed the mean ± 3 times the standard deviation are removed (a total of 3 outlier data points caused by equipment failure are removed); constant values ​​of cold and hot temperature differences are verified and corrected in conjunction with the building's external insulation conditions.

[0067] Normalization: Continuous data such as temperature, humidity, and load rate are normalized to eliminate dimensional differences. Specifically, the heat load data is normalized using the formula Q = meter reading × COP × building insulation correction factor × air conditioning performance degradation factor. In this embodiment, COP = 4.2, building insulation correction factor = 0.98, and air conditioning performance degradation factor = 0.9, resulting in the electrical load data.

[0068] Data is categorized into buckets based on factors such as equipment runtime, building age, and air conditioner age. For example, air conditioner age is categorized into buckets for 0-3 years, 4-8 years, 9-12 years, and over 12 years. Time characteristics (hours, days of the week, holidays, weekdays / restdays, daytime / nighttime) are encoded, with hourly characteristics divided into 15-minute time intervals.

[0069] S22: Feature Extraction

[0070] Feature selection: Core features strongly correlated with adjustable loads were selected by random forest feature importance scoring, including outdoor temperature, solar irradiance, temperature difference between hot and cold, building external insulation correction coefficient, air conditioning performance degradation coefficient, average heat load of central air conditioning in the previous 2 hours, and hourly occupancy rate, while weakly correlated wind speed features were removed.

[0071] Lag Feature Construction: Lag features such as the heat load of the air conditioning system in the first 15 minutes, the average heat load in the first hour, and the heat load trend in the first 2 hours are added to characterize the system's thermal inertia. In this embodiment, the average heat load in the first 2 hours and the heat load trend in the first 15 minutes are constructed as lag features. One-hot encoding (N / E / S / W) is applied to the building orientation to amplify the impact of heat load differences between rooms with different orientations.

[0072] Coupling Feature Construction: Constructing multi-parameter coupling features to quantify the synergistic influence of multiple factors. The coupling features constructed in this embodiment include:

[0073] The product of the temperature difference between hot and cold and the building's external insulation correction factor is: temperature difference between hot and cold × 0.98;

[0074] The product of solar irradiance and air conditioning performance degradation coefficient: solar irradiance × 0.9;

[0075] The product of time period characteristics and building occupancy rate: Time period coefficient × Occupancy rate, where the time period coefficient for 17:00-18:00 is 1.2;

[0076] The product of solar irradiance, time-of-day coefficient, and building external insulation correction coefficient: solar irradiance × 1.2 × 0.98, quantifies the synergistic effect of sunlight and insulation capacity of south-facing buildings during the evening hours.

[0077] Spatiotemporal coupling feature construction: Combining the solar radiation characteristics of a building facing south, a combined feature of solar irradiance and orientation coefficient is constructed.

[0078] The final first main impact factor data includes the aforementioned lag characteristics, coupling characteristics, and core characteristics after screening.

[0079] Step S3: Adjustable load assessment and control scheme generation

[0080] The data of the first major influencing factor is input into the pre-trained adjustable load assessment model to obtain the adjustable load assessment results within the target period, and the full-chain control scheme of the air conditioning system is generated simultaneously.

[0081] S31: Constructing an Adjustable Load Assessment Model

[0082] This embodiment uses the XGBoost-LSTM-GraphSAGE multi-model fusion model, and the specific construction process is as follows:

[0083] Data Collection: Historical data for the commercial complex over the past six months was collected, including historical influencing factor data (including specific data for the air conditioning system) under different adjustment strategies, and corresponding historical evaluation indicators such as actual adjustable load, response speed, and adjustment accuracy. Historical heat load forecast data for the air conditioning system was also supplemented. The data was then combined with an empirical value method and an XGBoost-LSTM fusion algorithm. Input features included outdoor dry-bulb temperature, solar irradiance, temperature difference between hot and cold, building external insulation parameters, air conditioning service life, time characteristics, occupancy rate, and lagging heat load.

[0084] Data preprocessing and feature extraction: Perform the preprocessing and feature extraction process described in step S2 on the historical impact factor data to obtain the second main impact factor data corresponding to different adjustment strategies.

[0085] Model training:

[0086] Train the XGBoost model (to capture non-linear features), the LSTM model (to capture time series dependencies), and the GraphSAGE model (to capture multi-parameter coupling relationships) respectively.

[0087] An adaptive weight fusion method is employed to dynamically adjust the weights of each sub-model based on real-time scene characteristics (such as weather conditions and the age of the air conditioner). For example, in extreme weather scenarios, the thermal inertia effect of the air conditioner is significant, thus increasing the weight of the LSTM sub-model; in scenarios involving older air conditioners that are over 8 years old, the multi-parameter coupling relationships are complex, thus increasing the weight of the GraphSAGE sub-model. In this embodiment, a preset threshold of 8 years is used; when the air conditioner is over 8 years old, the GraphSAGE weight is increased by 10%-20%.

[0088] Multi-dimensional dynamic constraints are embedded as boundary conditions for model inference. Specific constraints include:

[0089] Equipment constraints: The chiller unit operates within the range of 30%-100% of its rated cooling capacity, and the chilled water pump frequency is 30Hz-50Hz; when the air conditioner has been in use for more than 10 years, the adjustment range is narrowed to 40%-90% of the rated adjustment range.

[0090] Building constraints: When a building is over 15 years old and the external insulation integrity rate is less than 70%, the adjustment rate is reduced by 20%. In this example, the building is 8 years old and the external insulation integrity rate is 92%, so the adjustment rate is normal.

[0091] Time constraints: During weekday daytime hours (17:00-18:00), the lower limit of the regulation capacity is increased by 10%, and during weekday evening hours (18:00-21:00), the lower limit of the regulation capacity returns to normal. In this embodiment, 17:00-18:00 is weekday daytime, and the lower limit of the regulation capacity is increased by 10%.

[0092] Comfort constraints: Indoor temperature fluctuations are controlled within ±1℃, and humidity is controlled within the range of 40%-60%.

[0093] Meanwhile, the operational status data input into the model includes features such as terminal flow deviation, pipeline heat loss rate, and orientation difference coefficient, which are used to evaluate the operational health status and load distribution characteristics of the air conditioning system in real time.

[0094] Hyperparameter optimization: With the optimization objectives of adjusting capacity error, response time deviation, and stability coefficient, the model hyperparameters (learning rate, tree depth, number of iterations, weights of each sub-model, etc.) are determined by grid search.

[0095] Energy-saving control model integration: A rolling mixed integer linear programming (MILP) solver is embedded in the evaluation model to construct an integrated "evaluation-control" logic and generate the optimal control scheme for the air conditioning system simultaneously.

[0096] S32: Evaluation Results Output

[0097] The first main influencing factor data obtained in step S2 is input into the trained adjustable load assessment model to obtain the adjustable load assessment results for the target period (17:00-21:00 daily). The assessment results output in this embodiment include:

[0098] Core evaluation indicators:

[0099] Overall adjustable load specifications: The total adjustable load capacity is 380-550kW, of which the adjustable capacity of the air conditioning system alone is 320-460kW (accounting for 84%-85%), the average response speed is 2.8 minutes / time, the adjustment accuracy deviation rate is ≤2.5%, and the continuous adjustment time is 2 hours.

[0100] Air conditioning specific indicators: percentage of individually adjustable capacity of air conditioning system, response delay time of approximately 30 seconds, adjustment step size of 10kW, and equipment loss coefficient dynamically calculated based on the number of adjustments.

[0101] Optimal control scheme for air conditioning system (generated based on MILP solver):

[0102] Main unit control: 17:00-18:00 (weekday daytime + peak passenger flow) reduce the air conditioner power to 75% of the rated power, adjust the compressor operating frequency to 45Hz, and set the chilled water outlet temperature to 7.5℃; 18:00-20:00 maintain the air conditioner power at 65%, the compressor frequency at 42Hz, and the chilled water outlet temperature at 8.0℃; 20:00-21:00 reduce the air conditioner power to 55%, the compressor frequency at 39Hz, and the chilled water outlet temperature at 8.5℃.

[0103] Auxiliary equipment control: The frequency of chilled water pumps is adjusted accordingly: 42Hz from 17:00 to 18:00, reduced to 39Hz from 18:00 to 20:00, and reduced to 36Hz from 20:00 to 21:00; the frequency of cooling water pumps is optimized synchronously.

[0104] Terminal control: Adjust the opening of terminal valves and air supply speed according to the actual load demand of each area to ensure that the indoor temperature fluctuation does not exceed ±1℃.

[0105] Dynamic adjustment mechanism: The control parameters are updated every 15 minutes based on the latest heat load forecast and real-time operating data via the MILP solver. In this embodiment, an FPGA and GPU accelerator card combination architecture is adopted, and the time for a single evaluation and control scheme generation is ≤45 seconds, meeting the real-time requirements.

[0106] Decision support information:

[0107] Load Priority Ranking: Based on adjustment potential, response speed, adjustment cost, and equipment loss, the air conditioning system and elevator system are prioritized. The air conditioning system accounts for more than 84% and is given priority for adjustment.

[0108] Recommended air conditioning adjustment strategy: Based on the weather conditions of the day (high temperature and high humidity), building condition (8-year-old building, 5-year-old air conditioning), and time scenario (weekday daytime), it is recommended to adopt a "small amplitude, high frequency" adjustment strategy.

[0109] Benefit-Risk Assessment: The net benefit under different adjustment strategies is calculated by combining the dispatch compensation standard (0.5 yuan / kWh) with electricity costs and air conditioning equipment depreciation costs. In this embodiment, the gradient downward adjustment strategy is expected to increase net benefit by 12% compared to the traditional strategy.

[0110] Parameter sensitivity analysis: The impact weight of key output parameters on the evaluation results is shown. Among them, the "cold and hot temperature difference" has the highest impact weight (35%). It is recommended to strengthen the precise control of indoor temperature to improve the flexibility of adjustment.

[0111] Alarm prompts: Real-time monitoring of peak heat load, issuing an alarm when it approaches the equipment's upper limit; issuing an alarm when abnormal operating status of the air conditioning equipment is detected (such as excessive terminal flow deviation); issuing a warning when indoor temperature or humidity exceeds the comfort threshold.

[0112] Example 2

[0113] A public building virtual power plant adjustable load assessment system. This embodiment corresponds to the commercial complex in Embodiment 1, providing a public building virtual power plant adjustable load assessment system. Figure 2 As shown, the system includes:

[0114] The acquisition module integrates multiple data interfaces (including building equipment monitoring system interface, meteorological monitoring station interface, virtual power plant dispatch platform interface, building file management system interface, and equipment operation and maintenance management system interface) to automatically collect raw influencing factor data within the target period. The acquisition module includes a dedicated data acquisition submodule for air conditioning systems, used to collect data on air conditioning service life, performance degradation, terminal flow deviation, and pipeline heat loss rate; a building parameter acquisition submodule, used to collect building age and external insulation data; and a time scene recognition submodule, used to automatically synchronize with the calendar to determine weekdays / restdays and combine sunrise and sunset times to determine day / night. Data acquisition latency does not exceed 10 seconds.

[0115] Feature Construction Module: This module preprocesses and extracts features from the raw influencing factor data obtained by the acquisition module. It includes a built-in data cleaning unit based on physical constraints (e.g., terminal flow deviation threshold ±10%, pipeline heat loss rate threshold 3%), a missing value imputation unit, a normalization unit, and a binning unit. Simultaneously, the feature construction module includes a dedicated feature construction unit for air conditioning systems, used to generate lag features (e.g., heat load in the previous 15 minutes, average value in the previous hour), multi-parameter coupled features (e.g., temperature difference × external insulation coefficient, solar irradiance × performance degradation coefficient), and spatiotemporal coupled features. The feature construction module outputs the first primary influencing factor data.

[0116] The assessment and control module loads a pre-trained adjustable load assessment model, receives the first main influencing factor data output from the feature construction module and inputs it into the model to obtain the adjustable load assessment results within the target period, and simultaneously generates a full-chain control scheme for the air conditioning system. The assessment and control module includes:

[0117] The multi-model fusion computing unit integrates XGBoost, LSTM, and GraphSAGE sub-models and adopts an adaptive weight fusion strategy to dynamically adjust weights based on real-time scene features.

[0118] The dynamic constraint embedding unit is used to embed multi-dimensional dynamic constraints (equipment, building, time, comfort constraints) as boundary conditions for model inference into the model.

[0119] A rolling mixed-integer linear programming solver is used to generate full-chain control schemes for air conditioning system main units, auxiliary units, and terminal units, and supports rolling updates based on the latest data every 15 minutes.

[0120] The auxiliary decision generation unit is used to output information such as load priority ranking, strategy recommendation, revenue and risk assessment, and sensitivity analysis.

[0121] The alarm unit is used for real-time alarms for peak heat load, equipment malfunctions, and excessive comfort levels.

[0122] The assessment and control module adopts a combination architecture of FPGA and GPU accelerator card, and the time for generating a single assessment and control scheme is no more than 1 minute, meeting the real-time scheduling requirements of the virtual power plant. Simultaneously, this module supports one-click uploading of assessment results and control commands to the virtual power plant scheduling platform and building management system (BMS).

[0123] Example 3

[0124] An electronic device and a computer-readable storage medium. This embodiment provides an electronic device, such as... Figure 3 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method described in Embodiment 1.

[0125] Memory: including ROM (stores basic programs), RAM (stores real-time acquired raw data and intermediate evaluation results, capacity ≥32GB), and solid-state drive (stores pre-trained models, historical evaluation data, building parameter correction tables, air conditioning performance degradation curves, time scene division rules, etc., capacity ≥1TB).

[0126] Processor: It adopts a combination architecture of FPGA dedicated integrated circuit and GPU accelerator card (such as NVIDIA A100), which supports multi-model parallel operation, efficient MILP solution and fast execution of special algorithms for air conditioning system, ensuring real-time response.

[0127] I / O interfaces: Includes Gigabit Ethernet interface, RS485 interface (8 channels), and USB 3.0 interface (4 channels), supporting high-speed communication with building equipment, weather stations, virtual power plant platforms, building archive systems, and equipment operation and maintenance systems.

[0128] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in Embodiment 1. The storage medium can be any medium capable of storing program code, such as a USB flash drive, external hard drive, read-only memory, random access memory, magnetic disk, or optical disk. The program includes specialized modules such as a heat load data preprocessing subroutine, a dynamic constraint embedding subroutine, a multi-parameter coupling feature construction subroutine, a multi-model fusion training subroutine, a MILP optimization solution subroutine, and an air conditioning control scheme generation subroutine.

[0129] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for assessing the adjustable load of a public virtual power plant, characterized in that, include: Obtain the original impact factor data of public buildings participating in the virtual power plant within the target period. The original impact factor data includes at least the air conditioning system operating parameters reflecting the degree of equipment aging and the building thermal parameters reflecting the building thermophysical properties. Based on the operating parameters of the air conditioning system, a hysteresis feature characterizing the thermal inertia of the air conditioning system is constructed. Based on the coupling relationship between the operating parameters of the air conditioning system and the building thermal parameters, a multi-parameter coupling feature is constructed to obtain the data of the first main influencing factor. Input the data of the first main influencing factor into the pre-trained adjustable load assessment model to obtain the adjustable load assessment results within the target period; The adjustable load assessment model is a multi-model fusion model, and the model integrates a rolling mixed integer linear programming solver to synchronously generate a full-chain control scheme for the air conditioning system.

2. The method for assessing the adjustable load of a public virtual power plant according to claim 1, characterized in that, The hysteresis characteristics include at least one of the heat load of the previous 15 minutes, the average heat load of the previous 1 hour, and the heat load trend of the previous 2 hours. The multi-parameter coupling characteristics include at least one of the following: the product of the cold and hot temperature difference and the building external insulation correction coefficient, the product of solar irradiance and the air conditioning performance attenuation coefficient, and the product of time period characteristics and building occupancy rate.

3. The method for assessing the adjustable load of a public virtual power plant according to claim 1, characterized in that, The adjustable load assessment model is constructed through the following steps: Train the XGBoost sub-model, LSTM sub-model, and GraphSAGE sub-model separately; The outputs of the XGBoost sub-model, LSTM sub-model, and GraphSAGE sub-model are fused by adaptive weights, which are dynamically adjusted based on real-time scene features.

4. The method for assessing the adjustable load of a public virtual power plant according to claim 3, characterized in that, The application scenarios include meteorological conditions and the service life of air conditioners; In extreme weather scenarios, increase the weights of the LSTM sub-model; In scenarios where the air conditioner's service life exceeds a preset threshold, the weight of the GraphSAGE sub-model is increased.

5. The method for assessing the adjustable load of a public virtual power plant according to claim 1, characterized in that, Also includes: Multidimensional dynamic constraints are embedded as boundary conditions for model inference into the adjustable load assessment model. The multi-dimensional dynamic constraints include equipment constraints, building constraints, time constraints, and comfort constraints; The equipment constraints include a rule for narrowing the adjustment range based on the service life of the air conditioner: when the service life of the air conditioner exceeds 10 years, the adjustment range is narrowed to 40%-90% of the rated adjustment range; The building constraints include adjustment rate correction rules based on building age and external insulation integrity rate: when the building age exceeds 15 years and the external insulation integrity rate is less than 70%, the adjustment rate is reduced by 20%; The time constraints include adjustment rules for the lower limit of adjustment capacity based on weekday and time period characteristics: during weekday daytime hours, the lower limit of adjustment capacity is increased by 10%; The comfort constraints include an indoor temperature fluctuation range of ±1℃ and a humidity range of 40%-60%.

6. The method for assessing the adjustable load of a public virtual power plant according to claim 1, characterized in that, Also includes: Based on real-time heat load forecasts, the entire control scheme of the air conditioning system is updated every 15 minutes using the latest heat load forecasts and real-time operating data through the rolling mixed integer linear programming solver.

7. A public building virtual power plant adjustable load assessment system, characterized in that, include: The acquisition module is used to acquire the original impact factor data of public buildings participating in the virtual power plant within the target period. The original impact factor data includes at least the air conditioning system operating parameters and building thermal parameters. The feature construction module is used to construct hysteresis features characterizing the thermal inertia of the air conditioning system based on the operating parameters of the air conditioning system, and to construct multi-parameter coupled features based on the coupling relationship between the operating parameters of the air conditioning system and the building thermal parameters, so as to obtain the first main influencing factor data. The assessment and control module is used to load the pre-trained adjustable load assessment model, receive the first main influencing factor data output by the feature construction module and input it into the model to obtain the adjustable load assessment results within the target period, and simultaneously generate a full-chain control scheme for the air conditioning system. The adjustable load assessment model is a multi-model fusion model and integrates a rolling mixed integer linear programming solver.

8. The method for assessing the adjustable load of a public building virtual power plant according to claim 7, characterized in that, The evaluation and control module is also used to embed multi-dimensional dynamic constraints as boundary conditions for model inference into the adjustable load evaluation model. The multi-dimensional dynamic constraints include equipment constraints, building constraints, time constraints, and comfort constraints.

9. The method for assessing the adjustable load of a public building virtual power plant according to claim 7, characterized in that, The evaluation and control module adopts a combination architecture of FPGA and GPU acceleration card, and the generation time of a single evaluation and control scheme is no more than 1 minute. The data acquisition delay of the acquisition module shall not exceed 10 seconds.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 6.

Images