Multi-source data fusion air conditioning system low-carbon operation optimization method and system

Abstract

The invention relates to the field of building energy management, in particular to a multi-source data fusion air conditioning system low-carbon operation optimization method and system, and the method comprises the steps: constructing an energy data asset library to form a commercial operation data flow; injecting the commercial operation data flow into an air conditioner asset evaluation model to generate an operation performance evaluation report; meanwhile, the business prediction engine generates a prospective insight report by using historical and market data; constructing a linear programming problem through the two reports, and outputting an operation strategy combination according to an operation target; selecting one item in the combination, analyzing the item into a scheduling plan, and issuing and executing the scheduling plan; and the execution performance is audited, and generated data is fed back to the evaluation model and the prediction engine for calibration, so that a closed-loop low-carbon operation optimization system is formed. And the system generates a specific production instruction through the operation performance and the insight report, and feeds back the model for calibration when a deviation occurs, so that rapid and accurate response to resource scheduling is realized, and the response efficiency of resource scheduling and the reliability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the field of building energy management, and in particular to a multi-source data fusion low-carbon operation optimization method and system for air-conditioning systems. Background Art

[0002] In the modern energy services industry, a business model centered on "energy cost management" and fulfilling clients' low-carbon commitments is becoming mainstream. Within this framework, professional energy service providers are required to meticulously manage the air conditioning system, the largest energy-consuming asset within a client's building. This is no longer a simple matter of equipment maintenance; it has evolved into a complex, data-driven business service delivery challenge: how to optimally operate the air conditioning system within a dynamic market and grid environment, while simultaneously meeting the diverse business objectives of "cost," "low carbon," and "comfort" in service level agreements.

[0003] Existing technologies for regulating such energy assets employ relatively simple scheduling logic. The fundamental principle is "on-demand conversion," whereby electricity is consumed immediately to meet end-user comfort requirements. This model results in the total value of the output service being completely subject to the real-time fluctuations in the input energy properties. Further logic is developed based on simplified predictions of energy supply properties, creating scheduling rules based on preset thresholds. This attempt to mitigate some known negative attributes by preemptively converting and storing electricity. However, a fundamental flaw of existing technologies lies in the fragmented and non-real-time capabilities of their business decision-making engines. They are unable to effectively integrate multiple real-time business information streams describing the "market environment," "energy asset status," and "customer demand." This "business information silo" makes it difficult for the pre-set scheduling logic to dynamically optimize in the ever-changing real-world business environment. Consequently, opportunities for low-carbon, low-cost "energy procurement" are missed, and accurate and efficient "on-demand service delivery" cannot be achieved. Consequently, the overall commercial benefits of these technologies are naturally capped.

[0004] In summary, this field requires a new intelligent business operation method, which must be able to manage building air conditioning systems as dynamic "thermal energy warehouses" through real-time multi-source data fusion to achieve globally optimal low-carbon operation and scheduling decisions throughout the entire energy "supply-storage-demand" chain.

[0005] Therefore, a low-carbon operation optimization method and system for air-conditioning systems based on multi-source data fusion are proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide a multi-source data fusion low-carbon operation optimization method and system for air-conditioning systems to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a multi-source data fusion low-carbon operation optimization method for an air conditioning system, comprising: Build an energy data asset library, integrate internal environmental data, air conditioning operating data, and market data, and form a commercial operation data flow through standardized processing; The commercial operation data stream is input into the air conditioning asset valuation model to calculate procurement costs and carbon footprints and generate an operational performance evaluation report; the business forecasting engine uses historical operation mode data and market forecast data to generate a forward-looking insight report; Construct linear programming problems through operational performance evaluation reports and forward-looking insight reports; perform linear programming based on pre-set business operation goals and output operational strategy combinations; Select one of the operational strategy combinations, interpret it as an energy resource scheduling plan, and issue it for execution through the asset management interface; Audit the execution performance of the energy resource scheduling plan, feed the generated financial cost and environmental data into the air conditioning asset valuation model and business forecasting engine, and perform adaptive calibration to form a low-carbon operation optimization system.

[0008] Preferably, the specific implementation process of constructing the energy data asset library and forming the commercial operation data flow includes: By deploying data pipelines, utilizing industry protocols and customized API interfaces, internal environmental data and air conditioning operating data are extracted from building management systems and IoT sensor networks, and market data from power markets and meteorological service providers are accessed in real time to form an energy data asset library. The raw data in the energy data asset library is deeply cleaned to identify and correct errors, fill missing values, and remove outliers. Data conversion and time series resampling are then performed as needed. The processed data is mapped to a pre-designed unified data model, defining a standard "language" and structure for the data source, clarifying the attributes and associations of data entities, completing standardization, and forming a commercial operation data flow.

[0009] Preferably, the specific generation process of generating the operational performance evaluation includes: The air-conditioning asset evaluation model first synchronizes the energy consumption, electricity price and carbon emission factor data in the commercial operation data stream to a unified time base through a forward filling algorithm, and performs point-by-point cost and carbon footprint calculations for each time step to obtain instantaneous cost and carbon emissions; performs multi-dimensional aggregation and attribution analysis, aggregates instantaneous data on demand, and attributes it to different dimensions; uses historical aggregated data and statistical methods to establish a relationship model between energy consumption and key driving factors, generates a dynamic baseline to complete the evaluation; and combines the instantaneous cost and carbon emissions to transmit them to a multi-dimensional dashboard to form an operation performance evaluation report.

[0010] Preferably, the specific process of generating forward-looking insights includes: The business forecasting engine obtains market forecast data from a third-party data service provider through an Internet API interface, as well as historical operating mode data from an energy data asset library, and processes them into a set of feature vectors through feature engineering. The feature vectors are input into a forecasting benchmark model that learns the inherent thermodynamic characteristics of buildings and equipment performance laws through historical operating mode data to generate a forecast sequence. The set of potential control strategies generated by the forecast sequence is used as input for multi-scenario simulation. Within a preset future time window, the cumulative costs and carbon emissions generated under the strategy are calculated in combination with the electricity price and carbon emission factors of the market forecast data to form a multi-strategy evaluation matrix. After analysis and processing, the multi-strategy evaluation matrix is ​​packaged into a forward-looking insight report.

[0011] Preferably, the specific implementation process of outputting the operation strategy combination includes: A linear programming problem is constructed using the operational performance evaluation report and the forward-looking insight report, and is converted into a mathematical cost function through the weighted sum of multiple objectives based on preset business operation objectives; constraints are set through the physical performance limits of the equipment, the safety specifications of the system operation, and the user-defined comfort range; a mathematical programming solver is called to solve the linear programming problem and calculate a control sequence that satisfies the constraints; and an operational strategy combination is generated by adjusting the weight coefficients for calculation.

[0012] Preferably, the specific implementation process of parsing into an energy resource scheduling plan includes: The preset meta-strategy will select the optimal strategy from the combination based on the real-time situation and risk preference; decompose the optimal strategy into specific device-level control instructions, and calculate the time series set points required to achieve the goal, output the energy resource scheduling plan, define the operating parameters and schedule of each sub-unit in the instruction format, and send it to the underlying equipment for execution.

[0013] Preferably, the specific implementation process of the closed-loop low-carbon operation optimization system includes: During the execution of the energy resource scheduling plan, the generated performance is independently audited and measured; the execution result data is obtained; data comparison and error calculation are performed on the forward-looking insight report; error attribution analysis is performed, and after determining the source of the error, the parameters of the air-conditioning asset evaluation model are corrected; at the same time, the dynamic baseline in the air-conditioning asset evaluation model is updated; and a closed-loop low-carbon operation optimization system is formed.

[0014] Preferably, a multi-source data fusion low-carbon operation optimization system for air conditioning systems includes: Energy data module: Builds an energy data asset library, integrates internal environmental data, air conditioning operating data, and market data, and forms a commercial operation data flow through standardized processing; Asset evaluation module: The commercial operation data flow is input into the air-conditioning asset evaluation model to calculate the procurement cost and carbon footprint and generate an operation performance evaluation report; Business Forecasting Module: The business forecasting engine uses historical operating model data and market forecast data to generate forward-looking insight reports; Operation Strategy Module: Constructs linear programming problems through operation performance evaluation reports and forward-looking insight reports; performs linear programming based on preset business operation goals and outputs operation strategy combinations; Instruction parsing module: selects one of the operation strategy combinations, parses it into an energy resource scheduling plan, and issues it for execution through the asset management interface; Feedback calibration module: Audits the execution performance of the energy resource scheduling plan, feeds the generated financial cost and environmental data back to the air conditioning asset valuation model and business forecasting engine, and performs adaptive calibration to form a low-carbon operation optimization system.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Proactive predictive action: Using a business forecasting engine and multi-scenario simulation, operational conditions can be previewed for the next several hours or even days. This allows for proactive and strategic adjustments to buildings, leveraging their thermal inertia as an invisible "cooling battery." This allows for maintaining a comfortable environment by operating at very low power levels, even with intermittent downtime, effectively mitigating operational risks and financial losses.

[0016] 2. Its adaptive learning mechanism, through precise parameter adjustments and dynamic baseline updates, can "learn" changes in buildings and equipment over time. This means that, rather than becoming inert over time, the system will gain a deeper understanding of the characteristics of the managed assets as it accumulates more data and experience. This will lead to increasingly precise optimization decisions and ensure a long-term return on investment.

[0017] 3. For decision optimization, real-time access to electricity market and carbon emission factor data is required. When the grid's green energy share is high and the carbon factor is low, the system may opt to increase electricity consumption, pre-cooling, or energy storage, even when electricity prices are not at their lowest. This aligns a company's energy consumption behavior with society's overall low-carbon transition goals, transforming macroeconomic energy policies into profitable business practices. This transforms operating costs from passive expenditures into proactively manageable profit drivers. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1A flow chart of a method for optimizing low-carbon operation of an air-conditioning system by fusion of multi-source data proposed in an embodiment of the present invention; Figure 2 A flowchart of obtaining an operating strategy combination proposed in an embodiment of the present invention; Figure 3 This is a structural diagram of a low-carbon operation optimization system for an air-conditioning system using multi-source data fusion proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] See also Figure 1-Figure 2 The present invention relates to a low-carbon operation optimization method for an air-conditioning system based on multi-source data fusion. The specific implementation process is as follows: Build an energy data asset library, integrate internal environmental data, air conditioning operating data, and market data, and form a commercial operation data flow through standardized processing; The commercial operation data stream is input into the air conditioning asset valuation model to calculate procurement costs and carbon footprints and generate an operational performance evaluation report; the business forecasting engine uses historical operation mode data and market forecast data to generate a forward-looking insight report; Construct linear programming problems through operational performance evaluation reports and forward-looking insight reports; perform linear programming based on pre-set business operation goals and output operational strategy combinations; Select one of the operational strategy combinations, interpret it as an energy resource scheduling plan, and issue it for execution through the asset management interface; Audit the execution performance of the energy resource scheduling plan, feed the generated financial cost and environmental data into the air conditioning asset valuation model and business forecasting engine, and perform adaptive calibration to form a low-carbon operation optimization system.

[0021] The technical solution of the present invention is further described in detail below with reference to specific embodiments.

[0022] Example 1 The present application embodiment discloses a method for optimizing the low-carbon operation of an air-conditioning system by fusion of multi-source data, which performs energy cost management and low-carbon optimization operation on the heating, ventilation and air-conditioning (HVAC) system in a public building A. Figure 1The specific steps of the method proposed in the present invention include: S1, building an energy data asset library, integrating data, and forming a commercial operation data stream through standardized processing; S2, inputting the commercial operation data stream into the air-conditioning asset evaluation model to generate an operation performance evaluation report; S3, the business forecasting engine uses historical operation mode data and market forecast data to generate a forward-looking insight report; S4, the operation performance evaluation report and the forward-looking insight report construct a linear programming problem, perform linear programming based on preset business operation goals, and output an operation strategy combination; S5, select one of the operation strategy combinations, parse it into an energy resource scheduling plan, and issue it for execution through the asset management interface; S6, audit the execution performance of the energy resource scheduling plan, and feed back the generated financial cost and environmental data to the air-conditioning asset evaluation model and the business forecasting engine for adaptive calibration to form a low-carbon operation optimization system.

[0023] Furthermore, an energy data asset library is constructed to integrate data and form a commercial operation data flow through standardized processing. This corresponds to the above step S1. The specific implementation process includes: An energy data asset library was constructed in Public Building A, connecting to the building management system (BMS) by deploying a BACnet / IP-based data pipeline. The system extracts key data points from the BMS on a one-minute polling cycle, such as the zone temperature reported by the variable air volume (VAV) terminal units on each floor; the start / stop status of the chiller (values ​​of 0 or 1); the chilled water outlet / return temperature; and the operating frequency of the cooling water pump inverter. The system also connects to the State Grid in real time via a RESTful API, providing a 24-hour time-of-use electricity price list clearly defining peak, flat, and off-peak periods; as well as the real-time carbon emission factor for the power grid published by the local environmental protection department.

[0024] The raw data stream from the energy data asset library enters the cleaning module. Error correction and missing value filling are performed. For example, if a temperature sensor returns an abnormal value such as -99°C at a certain moment, the system uses adjacent value interpolation to correct the value using the average of the two preceding and subsequent moments. Outliers are also removed, for example, using the 3σ criterion to identify and remove spikes in power consumption data.

[0025] All raw data with different frequencies, such as 1-minute temperature data and 15-minute electricity price data, are uniformly resampled to a standard 15-minute time interval to facilitate aligned analysis. Brick Schema is used as a unified data model. The cleaned and transformed data is mapped to this model. For example, a data point named ZN-L3-RM301-T from a BMS is standardized and labeled as: {Entity Type: Temperature_Sensor, Relationship: isLocationOf Room-301, Label: {Unit: °C, Source: BMS}}. Through this step, all data is given machine-readable, standardized semantics, ultimately forming a structured business operation data stream.

[0026] Through specific technical means such as BACnet, API, Brick Schema, and processing algorithms, the "data island" and semantic inconsistency problems caused by heterogeneous data sources are resolved, and chaotic, raw data is transformed into high-quality, standardized, machine-understandable data assets, providing a reliable data foundation for upper-level evaluation, prediction, and optimization models.

[0027] Furthermore, the commercial operation data stream is injected into the air-conditioning asset evaluation model to generate an operation performance evaluation report; corresponding to the above step S2; the specific implementation process includes: The air conditioning asset valuation model receives a standardized stream of commercial operating data at 15-minute intervals. First, the model uses a forward-fill algorithm to ensure that energy consumption, electricity price, and carbon factor data are fully aligned at each time step, such as 2:00:00 PM on July 15, 2023. It then performs a point-by-point calculation, calculating the instantaneous cost using the metered data showing the total power consumption of the air conditioning system and the corresponding time-of-use electricity price. The instantaneous carbon footprint is calculated using the carbon intensity displayed by the grid API.

[0028] This instantaneous data is aggregated across different dimensions. For example, instantaneous cost data for the entire day is accumulated across peak, flat, and off-peak periods to obtain the total electricity bill for each period. Furthermore, total energy consumption is attributed to equipment types such as chillers, pumps, and fans, analyzing the energy consumption contribution of each equipment type.

[0029] Using historical data from the past year, a set of characteristic variables influencing air conditioning system energy consumption is determined. This set includes at least the current and past N time periods (e.g., N = 24) of outdoor dry-bulb temperature, outdoor relative humidity, global solar radiation intensity, average indoor CO2 concentration (indicating building occupancy), temporal characteristics, and the total energy consumption of the air conditioning system over the past N time periods. A dynamic baseline is generated using a gradient boosting decision tree. Temperature, humidity, air velocity, and CO2 concentration, obtained through the sensor network, are input into a predictive mean voting (PMV) model to generate a comfort index. Operating parameters are collected from device controllers, and motor current characteristics are analyzed to determine device health. The comfort index and device health are integrated into the dynamic baseline to create a multidimensional dynamic baseline. This establishes a dynamic, multidimensional "ideal state" reference system that can assess the building's comprehensive operational performance in real time and reveal the inherent connections and underlying issues across dimensions. This significantly improves operational efficiency, reliability, and asset value preservation. For example, at the current time, with an outdoor temperature of 34°C and an occupancy rate of 80%, the dynamic baseline model predicts a reasonable energy consumption of 500kW. The actual measured energy consumption is 520kW, which is 4% higher. This deviation will be highlighted.

[0030] Ultimately, all instantaneous data, aggregated attribution analysis results, and performance deviations based on dynamic baselines are integrated and delivered to a multi-dimensional interactive dashboard, forming a graphically illustrated operational performance evaluation report for management review.

[0031] By establishing a dynamic baseline, a scientific and real-time performance measurement scale is provided, which can accurately quantify the costs and carbon emissions at each moment, and conduct attribution analysis from multiple dimensions, providing a data-driven and quantifiable basis for discovering energy-saving potential and operational problems.

[0032] Furthermore, the business forecasting engine uses historical operating model data and market forecast data to generate a forward-looking insight report, corresponding to the above S3 step. The specific implementation process includes: The business forecasting engine obtains market forecast data from third-party data service providers through an Internet API interface, including weather forecasts for the next 24 hours, such as temperature, humidity, solar radiation intensity, and electricity price schedules, and combines this with hourly historical operating mode data from the past few weeks in the energy data asset library. Through feature engineering, this data is converted into a set of feature vectors, for example, a vector containing [load for the past hour, current hour (0-23), whether it is a weekday (0 / 1), predicted temperature for the next hour, predicted electricity price for the next hour]. An unsupervised learning algorithm is used to analyze historical operating mode data, automatically learning and marking "event signatures" with typical characteristics as important contextual features. By giving semantics to real-time data, predictions and decisions are based on a deeper understanding of the internal operating status of the building, making optimization strategies more targeted and predictive.

[0033] The feature vectors are input into a long short-term memory (LSTM) model pre-trained using historical operating pattern data. This LSTM model consists of an input layer, two stacked LSTM layers, and a fully connected output layer. A dropout layer with a dropout rate of 0.2 is placed between the two LSTM layers to prevent model overfitting. The model is trained using the Adam optimizer, employing the mean squared error (MSE) as the loss function. The model learns the inherent thermodynamic properties of public building A, such as thermal inertia, and equipment performance patterns. The model outputs a 24-hour forecast sequence for the building's cooling load, with a time step of 15 minutes. For example, the model predicts that Building A's cooling load will peak at approximately 600 kW between 3:00 and 5:00 p.m. the following day due to high outdoor temperatures (36°C) and increased solar radiation.

[0034] Based on the predicted load, the system generates a set of potential control strategies for simulation. Strategy A (baseline strategy): Adheres to a standard operating schedule and provides cooling on demand. Strategy B (pre-cooling strategy): During the off-peak hours of 2:00 AM to 5:00 AM, the building temperature setpoint is lowered by 1.5°C in advance, utilizing the building structure to "store" cooling capacity. The system simulates the execution of these two strategies in a virtual environment. Combined with the predicted electricity price and carbon emission factors, the system calculates the cumulative cost and carbon emissions of each strategy over the next 24 hours, forming a multi-strategy evaluation matrix. The simulation results show that Strategy B reduces total operating costs by approximately 12% and total carbon emissions by approximately 8% compared to Strategy A, but it also increases energy consumption during the early morning hours.

[0035] The multi-strategy assessment matrix is ​​analyzed and visualized into an intuitive forward-looking insight report that clearly shows the cost, carbon emissions and potential risk trade-offs under different strategies.

[0036] By leveraging advanced machine learning models and multi-scenario simulations, it transforms "post-analysis" into "pre-planning," providing managers with quantitative insights into future operational risks and opportunities based on data predictions. This is a key step in achieving a shift from passive control to active optimization decisions.

[0037] Furthermore, the operational performance evaluation report and the forward-looking insight report construct a linear programming problem. Based on the preset business operation goals, linear programming is performed to output an operational strategy combination. Corresponding to the above S4 step, refer to Figure 2 ; The specific implementation process includes: Based on the real-time status in the operation performance evaluation report and the future prediction in the business forward-looking insight report, the system constructs the air conditioning control problem as a mixed integer linear programming (MILP) problem.

[0038] Pre-set business objectives, such as "synergistically optimizing energy costs and carbon emissions while meeting comfort standards," are converted into a mathematical cost function using a weighted sum of energy costs and carbon emissions. Furthermore, strict boundaries must be set for the linear programming problem: Regarding physical performance limits, Public Building A has two chillers, with Chiller 1's minimum part load rate (PLR) being 40%; Chiller 2's minimum operating time being 30 minutes to prevent frequent starts and stops; safety regulations require a minimum chilled water outlet temperature of 5°C to prevent pipe freezing; and the comfort range is that the indoor temperature in office areas must be maintained between 22°C and 26°C during working hours. When the solver finds an optimal solution that closely adheres to a constraint boundary, the system triggers a special "boundary exploration" mode to simulate "minor, temporary violations of the constraint" and utilizes the building's thermodynamic model and historical operating data for risk assessment. Finally, this "violation" strategy is added as a special option to the operational strategy portfolio, and its risk level and potential benefits are clearly marked. The constraint is transformed from a rigid number to a dynamic variable that can be evaluated and associated with physical risks, allowing more refined optimization decisions to be made.

[0039] Call a commercial mathematical programming solver such as Gurobi or CPLEX to solve the MILP problem. The system performs multiple iterative operations by adjusting the weight coefficients: Operation 1 is set to cost priority to obtain a control sequence that minimizes the total cost. Operation 2 is set to carbon emission priority to obtain a control sequence that minimizes total carbon emissions. Operation 3 is set to a balance strategy to obtain a solution with a relatively balanced cost and carbon emission. The results of these three operations, as well as the solutions under other weight combinations, together constitute an operating strategy combination that includes multiple optimization options. By building a rigorous mathematical optimization model with multiple constraints and quantifying business goals into adjustable cost functions, it is possible to intelligently explore the optimal balance between cost, carbon emissions and comfort, transforming complex operational decision-making problems into a solvable mathematical problem, and providing scientific, flexible and quantitative decision-making support.

[0040] Furthermore, one of the operational strategy combinations is selected, parsed into an energy resource scheduling plan, and issued for execution through the asset management interface; corresponding to the above-mentioned step S5; the specific implementation process includes: The system has a pre-set meta-strategy. For example, the meta-strategy for public building A on weekdays is "Balanced," and the meta-strategy for holidays is "Lowest Cost." Since it's a weekday, the system automatically selects the "Balanced Strategy" from the operational strategy portfolio as the optimal strategy for execution. The system breaks down the selected "Balanced Strategy," a macro-plan for the next 24 hours, into specific, executable device-level control instructions. For example, at 2:00 AM, the chilled water outlet temperature setpoint for chiller No. 1 is adjusted from 7°C to 6°C; at 2:15 AM, the inverter frequency command for the cooling tower fan is increased from 35Hz to 45Hz; and at 8:00 AM, the office area temperature setpoint is restored from 26°C to 24°C.

[0041] These time-series setpoints are integrated into a structured energy resource schedule. Defined in JSON format, this schedule clearly outlines the target values ​​for each BACnet object instance at 15-minute time steps over the next 24 hours. This schedule is then distributed via the system's integrated BACnet / IP gateway via WriteProperty commands to the corresponding controllers in the underlying BMS for execution at precisely predetermined times.

[0042] Through automated strategy decomposition and instruction generation, complex optimization results are converted into precise instruction sequences that the BMS can understand and execute, ensuring that the optimization intentions can be implemented without deviation. This is a key link in achieving automated and precise control.

[0043] Furthermore, the execution performance of the energy resource scheduling plan is audited, and the generated financial cost and environmental data are fed back to the air conditioning asset valuation model and business forecasting engine, and adaptive calibration is performed to form a low-carbon operation optimization system. This corresponds to the above step S6. The specific implementation process includes: During and after the execution of the energy resource scheduling plan, the system audits the actual performance generated using independent metering instruments, comparing the actual audit results with the predicted values ​​in the original multi-strategy evaluation matrix to determine the error. The system then initiates an attribution analysis algorithm to determine the source of the error and pinpoint the cause. The algorithm analyzes the correlation between relevant variables such as outdoor temperature, indoor CO2 concentration, and equipment operating data and the errors in the prediction model. The analysis found that the actual outdoor temperature was highly consistent with the prediction, but the CO2 concentration in the office area remained consistently higher than the historical level during the afternoon hours. The algorithm determined that the main source of the error was "higher-than-predicted building occupancy," meaning that the actual number of users exceeded the model's prediction, resulting in an increase in internal heat sources.

[0044] Based on the attribution conclusions, the system revised the upstream model and used the "occupancy-heat load" data from this incident as a new training sample to update the business forecasting engine, improving its sensitivity to future occupancy changes and predictive accuracy. Simultaneously, the actual energy consumption data from this incident was used to update the dynamic baseline in the air conditioning asset valuation model, ensuring it more accurately reflects the building's current operating characteristics. This "audit-comparison-attribution-correction" process is conducted continuously and automatically on a daily or weekly basis, forming a closed-loop low-carbon operation optimization system that continuously learns and evolves.

[0045] By establishing an automated, data-driven feedback loop, the system is equipped with the ability to self-calibrate and self-improve, ensuring that optimization decisions are always based on the most accurate building model, thereby ensuring long-term and sustainable energy savings and maximizing the life cycle value of the entire system.

[0046] The holistic approach defined in the present invention solves the technical problems of traditional air-conditioning system control strategies that are static, passive and have a single goal. By constructing a complete closed loop from data fusion, evaluation, prediction, decision-making to execution and feedback, it achieves dual optimization of costs and carbon emissions, giving the air-conditioning system unprecedented business intelligence and environmental adaptability, and significantly improving energy utilization efficiency and economic benefits.

[0047] Example 2 Through the low-carbon operation optimization system of air conditioning system based on multi-source data fusion, energy cost management and low-carbon optimization operation of heating, ventilation and air conditioning (HVAC) system in public building A are carried out. Figure 3 , the specific implementation of the energy data module is as follows: The system captures the following raw data streams in real time from all corners of the building and the external market: internal environmental data from the IoT sensor network: {timestamp: 14:01:10, sensor ID: T-L3-RM301, value: 25.8, unit: °C} {timestamp: 14:01:15, sensor ID: CO2-L3-RM301, value: 1250, unit: ppm}; air conditioning operating status data from the building automation system BACnet / IP: {timestamp: 14:01:05, object ID: CH-1_Power, value: 521.5, unit: kW} {timestamp: 14:01:0 8, Object ID: CHW-1_SupplyTemp, Value: 7.1, Unit: °C} {Timestamp: 14:01:20, Object ID: PUMP-1_Freq, Value: 45.2, Unit: Hz} {Timestamp: 14:02:00, Object ID: T-L3-RM302, Value: -99.9, Unit: °C}; Market data from the external API interface: {Timestamp: 14:00:00, Source: StateGrid_API, Information: {Time-of-use electricity price: 1.2, Price unit: ¥ / kWh, Carbon intensity: 485, Carbon intensity unit: g / kWh}}.

[0048] Upon receiving the raw data, the module immediately initiates a multi-stage processing pipeline. First, data cleansing is performed. The system automatically identifies the T-L3-RM302 value of -99.9 as a significant anomaly. Using adjacent value interpolation, the system takes the average of the two valid readings preceding and following it, 25.6°C and 25.7°C, correcting the erroneous data point to 25.65°C. Next, time series resampling is performed. To unify the analysis, all data at different frequencies, including seconds and minutes, are aggregated into standard 15-minute intervals. For example, the arithmetic mean of the 15 minute readings of CH-1_Power within the 2:00 PM to 2:15 PM time window (521.5, 519.8, ..., 520.5 kW) is calculated, resulting in a stable value of 520.0 kW representing that period. The cleaned and aligned data is then mapped into the Brick Schema data model, giving it a unified business meaning. This results in a clearly structured and semantically consistent business operation data stream record. The following is a single record generated at 14:15:00: {Timestamp: 2023-07-15 14:15:00, Total building power (kW): 520.0, Chiller 1 power (kW): 310.5, Pump unit power (kW): 115.2, Fan unit power (kW): 94.3, Average temperature (°C): 25.9, Maximum temperature zone: 26.4, Average CO2 concentration (ppm): 1220, Outdoor temperature (°C): 34.2, Electricity price (¥ / kWh): 1.2, Carbon intensity (g / kWh): 485}.

[0049] Example 3 In public building A, the specific implementation of the asset assessment module and the business forecast module is as follows: The asset assessment module receives the JSON-formatted business operation data stream record at 14:15:00 from the energy data module. The algorithm engine within the asset assessment module immediately performs a series of calculations on the input data: the instantaneous cost is calculated using the total building power (kW), the consumed time, and the electricity price (¥ / kWh): cost = 520.0kW * 0.25h * 1.2¥ / kWh = 156.0¥; the instantaneous carbon footprint is calculated using the total building power (kW), the consumed time, and the carbon intensity (g / kWh): carbon = 520.0kW * 0.25h * 485g / kWh = 63050g = 63.05kg.

[0050] To determine whether current energy consumption is reasonable, the module uses a built-in multivariate regression model trained on historical data. It assumes a base load of 25.5 kW, a coefficient for quantified outdoor temperature (°C) of 10.1, and a coefficient for quantified average CO2 concentration of 0.05. The model then calculates the theoretical "standard power consumption" under current conditions. Substituting the current data: predicted power = 25.5 + 10.1 * 34.2 + 0.05 * 1220 = 431.92 kW. Finally, the actual power consumption is compared with the baseline, revealing a deviation of approximately 20.4% in quantified operational efficiency. All calculation results are packaged into a concise and clear operational performance evaluation report: {Timestamp: 2023-07-15 14:15:00, Instantaneous Cost (¥): 156.0, Instantaneous Carbon (kg): 63.05, Baseline Power (kW): 431.92, Performance Deviation Percentage: 20.4, Status Message: Warning: Energy Consumption Significantly Higher Than Baseline}.

[0051] Meanwhile, the business forecasting module continuously digests the accumulated 15-minute business operation data streams from the past few weeks, as well as 24-hour weather and time-of-use electricity price forecasts from external APIs. By feeding both historical data and market forecast data into a pre-trained LSTM (Long Short-Term Memory) model, it learns the building's thermodynamic characteristics and operating patterns, and based on this, produces a rolling, accurate forecast of the building's cooling load for the next 24 hours (96 time points). The model's forecast results are packaged into a forward-looking insight report: {Forecast generated time: 2023-07-15 14:16:00, Forecast time horizon (hours): 24, Forecast load (kW): [{Timestamp: 14:30:00, Value: 525.0}, {Timestamp: 14:45:00, Value: 530.0}, {Timestamp: 15:00:00, Value: 540.0}, ... {Timestamp: 02:00:00, Value: 150.0}]}.

[0052] Example 4 In public building A, the specific implementation of the operation strategy module and the instruction parsing module is as follows: The operational performance evaluation report from the asset valuation module and the forward-looking insight report from the business forecast module are input into the operational strategy module. The air conditioning control problem is constructed as a linear programming problem and solved by a mixed integer linear programming model. First, a binary variable is generated to represent the discrete start-stop decision: δ ch，i (t)∈{0,1}: indicates whether the i-th chiller is turned on at time step t (1 for on, 0 for off). δ pump，j (t)∈{0,1}: Indicates whether the j-th pump is turned on at time step t. This is then used to generate a continuous variable representing the continuously adjustable operating set point: T chw，sp(t): chilled water outlet temperature set point (°C) at time step t; L ch，i (t): Load of the i-th chiller at time step t (kW); T zone，k (t): predicted indoor temperature (°C) of the kth zone at time step t.

[0053] The objective function quantifies the business goals (low cost, low carbon, and high comfort) into a single value J to be minimized. First, the predicted total electricity cost C at a single time step t is calculated. This is obtained by multiplying the total power consumption of all operating chillers by the electricity price for that period, and adding it to the demand response fee, obtained by multiplying the maximum load value by a specific demand charge unit price. The predicted total carbon emissions E at a single time step t are then calculated by multiplying the total power consumption of all operating chillers by the carbon emission factor of the grid for that period. Finally, at a single time step t, the severity of the deviation of the indoor temperature in all areas from the preset comfort zone is quantified. If the temperature is within the comfort zone, the penalty is zero. If it is outside the comfort zone, the deviation is calculated and then squared to increase the penalty for severe deviations. Finally, the penalty values ​​for all areas are added together to obtain the total discomfort penalty P. The cost C, carbon emissions E, and discomfort P are calculated for each future time step and then multiplied by their respective weight coefficients (w cost , w carbon , w comfort ), and finally add up the weighted results of all time steps to get the business target value J.

[0054] Calculations for business objectives must be performed within certain constraints. First, they must meet building thermodynamic constraints. Second, they must ensure that the load on chillers, whether running or not, remains within the permitted range, and that the total load provided by all operating chillers equals the predicted total building load. Furthermore, at all times, the predicted indoor temperature in any area must remain strictly within the lower and upper comfort limits, with no room for deviation.

[0055] The operation strategy module inputs all decision variables, objective functions and constraint equations, along with the next 24 hours of data obtained from the forecast module, into a commercial-grade MILP solver such as Gurobi or CPLEX. In order to resolve multi-objective conflicts, the system does not solve the problem only once, but adjusts the weight coefficient w cost , w carbon , w comfort Perform multiple iterative operations.

[0056] Operation 1 (cost optimization): Set w cost =0.9, w carbon =0.05, w comfort= 0.05. The solver computes the strategy with the lowest cost.

[0057] Operation 2 (optimal carbon emissions): Set w cost =0.05, w carbon =0.9, w comfort = 0.05. The solver calculates the strategy with the lowest carbon emissions.

[0058] Operation 3 (optimal comfort): Set w cost =0.05, w carbon =0.05, w comfort = 0.9. The solver calculates the strategy with the highest comfort.

[0059] The results of this series of calculations form a portfolio of operational strategies containing multiple optimal options. Each option represents a different trade-off between cost, carbon, and comfort, and is accompanied by estimated cost savings and carbon reduction data for final decision-making.

[0060] An optimal strategy selected from the Operational Strategy module, such as the JSON object with strategy ID: PRECOOL-BAL-20230715, is represented by a control sequence array, which defines the control plan for the next 24 hours. For example, a command in the sequence might look like: {Time: 02:00:00, Device: CH-1, Parameter: CHW_SupplyTemp_SP, Value: 6.0}. It then searches the semantic database for CH-1 and CHW_SupplyTemp_SP and retrieves their corresponding physical address information: {Protocol: 'BACnet / IP', Device Instance: 2, Object Type: 'analogOutput', Object Instance: 101}. After retrieving the physical address, the module assembles all this information into a structured command object that fully complies with the underlying BACnet communication protocol: {Execution Time: 2023-07-16 02:00:00, Protocol: BACnet / IP, Target Address: 192.168.1.10, Write Attributes: {Object Identifier: [analogOutput, 101], Attribute Identifier: presentValue, Value: 6.0, Priority: 8}}. Finally, all formatted command objects are placed in a timed queue sorted by execution time. A scheduling service continuously monitors this queue and, when each command's scheduled execution time arrives, accurately delivers it to the target device via the corresponding BACnet / IP gateway.

[0061] Example 5 In public building A, the specific implementation of the feedback calibration module is as follows: At 3:00 PM the following day, after executing the command, data collection was performed, recording an actual total power of 555.0 kW. Looking back at the original forecast, the business forecast module had predicted that the load at 3:00 PM under this strategy would be 540.0 kW. This calculation error was 15.0 kW. Deeper data mining using the Deviation Contribution Analysis algorithm revealed that the actual CO2 concentration during that period was 1550 ppm, significantly higher than the historical average of 1300 ppm used in the model's predictions. The error was therefore accurately attributed to "unexpected high-density crowd activity." This error information was captured and utilized by the system. The data from this event (outdoor temperature, CO2 concentration, ..., actual power) was used as a new, highly weighted sample to feed back and update the dynamic baseline regression model in the asset valuation module.

[0062] Through the precise division of labor among the six modules and the seamless flow of data, the system forms a complete closed loop from perception, analysis, decision-making, execution to final feedback learning, continuously improving the accuracy and effectiveness of its energy optimization for public building A.

[0063] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A low-carbon operation optimization method for air-conditioning systems based on multi-source data fusion, characterized in that: include: Build an energy data asset library, integrate internal environmental data, air conditioning operating data, and market data, and form a commercial operation data flow through standardized processing; The commercial operation data flow is input into the air-conditioning asset evaluation model to calculate the procurement cost and carbon footprint and generate an operation performance evaluation report; The business forecasting engine uses historical operating model data and market forecast data to generate forward-looking insight reports; Construct linear programming problems using operational performance evaluation reports and forward-looking insight reports; Conduct linear planning based on preset business operation goals and output operational strategy combinations; Select one of the operational strategy combinations, interpret it as an energy resource scheduling plan, and issue it for execution through the asset management interface; Audit the execution performance of the energy resource scheduling plan, feed the generated financial cost and environmental data into the air conditioning asset valuation model and business forecasting engine, and perform adaptive calibration to form a low-carbon operation optimization system.

2. The low-carbon operation optimization method of an air-conditioning system based on multi-source data fusion according to claim 1 is characterized in that: The specific implementation process of building an energy data asset library and forming a commercial operation data flow includes: by deploying data pipelines, utilizing industry protocols and customized API interfaces, extracting internal environmental data and air conditioning operating data from building management systems and Internet of Things sensor networks, and accessing market data from power markets and meteorological service providers in real time to form an energy data asset library; deeply cleaning the raw data of the energy data asset library to identify and correct errors, fill missing values, and eliminate outliers; then performing data conversion and time series resampling as needed; mapping the processed data to a pre-designed unified data model, defining a standard "language" and structure for the data source, and clarifying the attributes and associations of data entities to complete standardization and form a commercial operation data flow.

3. The low-carbon operation optimization method of an air-conditioning system based on multi-source data fusion according to claim 1 is characterized in that: The specific generation process of the operational performance evaluation includes: the air-conditioning asset evaluation model first synchronizes the energy consumption, electricity price and carbon emission factor data in the commercial operation data stream to a unified time base through a forward filling algorithm, and performs point-by-point cost and carbon footprint calculations for each time step to obtain instantaneous cost and carbon emissions; performs multi-dimensional aggregation and attribution analysis, aggregates instantaneous data on demand, and attributes it to different dimensions; uses historical aggregated data and statistical methods to establish a relationship model between energy consumption and key driving factors, generates a dynamic baseline to complete the evaluation; and combines the instantaneous cost and carbon emissions to transmit them to a multi-dimensional dashboard to form an operational performance evaluation report.

4. The low-carbon operation optimization method for air-conditioning system based on multi-source data fusion according to claim 1 is characterized in that: The specific generation process of the forward-looking insight report includes: the business forecast engine obtains market forecast data from a third-party data service provider through an Internet API interface, and obtains historical operating mode data from an energy data asset library, and processes them into a set of feature vectors through feature engineering; the feature vectors are input into a prediction benchmark model that learns the inherent thermodynamic characteristics of the building and the performance laws of the equipment through the historical operating mode data to generate a prediction sequence; the set of potential control strategies generated by the prediction sequence is used as input for multi-scenario simulation; within a preset future time window, the electricity price and carbon emission factor of the market forecast data are combined to calculate the cumulative cost and carbon emissions generated under the strategy to form a multi-strategy evaluation matrix; the multi-strategy evaluation matrix is ​​analyzed and processed and packaged into a forward-looking insight report.

5. The low-carbon operation optimization method of an air-conditioning system based on multi-source data fusion according to claim 1 is characterized in that: The specific implementation process of outputting the operation strategy combination includes: constructing a linear programming problem through the operation performance evaluation report and the forward-looking insight report, and converting it into a mathematical cost function through the weighted sum of multiple objectives based on the preset business operation objectives; setting constraints through the physical performance limits of the equipment, the safety specifications of the system operation, and the user-defined comfort range; calling a mathematical programming solver to solve the linear programming problem and calculate the control sequence that meets the constraints; and generating an operation strategy combination by adjusting the weight coefficients for calculation.

6. The low-carbon operation optimization method of an air-conditioning system based on multi-source data fusion according to claim 1 is characterized in that: The specific implementation process of parsing into an energy resource scheduling plan includes: the preset meta-strategy will screen the optimal strategy from the combination based on the real-time situation and risk preference; the optimal strategy will be decomposed into specific device-level control instructions, and the time series set points required to achieve the goal will be calculated to produce an energy resource scheduling plan. The operating parameters and schedule of each sub-unit are defined in an instruction format and sent to the underlying equipment for execution.

7. The low-carbon operation optimization method of an air-conditioning system based on multi-source data fusion according to claim 1 is characterized in that: The specific implementation process of the closed-loop low-carbon operation optimization system includes: independently auditing and measuring the performance generated during the execution of the energy resource scheduling plan; obtaining execution result data; performing data comparison and error calculation on the forward-looking insight report; performing error attribution analysis, determining the source of the error, and then correcting the parameters of the air-conditioning asset evaluation model; at the same time, updating the dynamic baseline in the air-conditioning asset evaluation model; forming a closed-loop low-carbon operation optimization system.

8. A multi-source data fusion low-carbon operation optimization system for air conditioning systems, characterized by: include: Energy data module: Builds an energy data asset library, integrates internal environmental data, air conditioning operating data, and market data, and forms a commercial operation data flow through standardized processing; Asset evaluation module: The commercial operation data flow is input into the air-conditioning asset evaluation model to calculate the procurement cost and carbon footprint and generate an operation performance evaluation report; Business Forecasting Module: The business forecasting engine uses historical operating model data and market forecast data to generate forward-looking insight reports; Operation Strategy Module: Constructs linear programming problems through operation performance evaluation reports and forward-looking insight reports; performs linear programming based on preset business operation goals and outputs operation strategy combinations; Instruction parsing module: selects one of the operation strategy combinations, parses it into an energy resource scheduling plan, and issues it for execution through the asset management interface; Feedback calibration module: Audits the execution performance of the energy resource scheduling plan, feeds the generated financial cost and environmental data back to the air conditioning asset valuation model and business forecasting engine, and performs adaptive calibration to form a low-carbon operation optimization system.

Images