A central air conditioning system control strategy generation and verification method

Abstract

The application provides a central air conditioning system control strategy generation and verification method, introduces digital twin virtual verification, strategy risk grading, closed-loop monitoring switching and model dynamic correction, forms a closed-loop intelligent control system of virtual and real fusion, and significantly improves the safety, stability, energy saving and self-adaptive ability of system operation.

Description

Technical Field

[0001] This invention relates to the field of central air conditioning system operation control technology, and in particular to a method for generating and verifying control strategies for central air conditioning systems. Background Technology

[0002] Central air conditioning systems are core energy-consuming devices in large commercial buildings, office buildings, industrial parks, and other similar settings. The rationality of their operation and control directly affects system energy efficiency, equipment lifespan, and indoor comfort. Traditional central air conditioning system operation and control methods are based on the current load, relying solely on real-time load feedback for adjustment. They trigger preset adjustments only by comparing real-time collected load-related parameters (such as temperature, pressure, and flow rate) with set thresholds, failing to predict future loads. This results in energy waste and delayed response to sudden load changes, impacting comfort.

[0003] Existing technologies incorporate load forecasting to achieve forward control of central air conditioning systems. However, existing predictive control schemes have significant technical drawbacks: First, they often use point forecasting to output single load values, failing to characterize the uncertainty of the forecast results and lacking a quality assessment mechanism. Blindly using erroneous forecast results can easily lead to system fluctuations, or even insufficient or excessive cooling. Second, control decisions rely solely on load fluctuation information without integrating key dimensions such as load trends and system capacity margins, which can easily result in aggressive or lagging control, leading to increased energy consumption or frequent equipment operation. Third, they do not fully consider the actual engineering constraints of central air conditioning systems, such as cooling capacity transfer delays, system thermal inertia, and high losses during chiller start-up and shutdown. This disconnect between theoretical design and engineering implementation can easily lead to control oscillations during control command execution, significantly reducing equipment lifespan. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method for generating and verifying control strategies for central air conditioning systems. This method introduces digital twin virtual verification, strategy risk classification, closed-loop monitoring switching, and model dynamic correction to form a closed-loop intelligent control system that integrates virtual and real elements, thereby significantly improving the system's safety, stability, energy efficiency, and adaptability.

[0005] To solve the above-mentioned technical problems, the technical solution used in this invention is as follows:

[0006] The method for generating and verifying a control strategy for a central air conditioning system according to the present invention includes the following steps:

[0007] S1. Data Acquisition: Acquire operational and environmental data of the central air conditioning system.

[0008] S2. Multi-timescale load forecasting: Construct a multi-timescale load forecasting model to predict load changes at different time scales in the future based on operational and environmental data, and generate load forecasting intervals at different time scales.

[0009] S3. Forecast Reliability Assessment: Calculate the comprehensive forecast reliability score for the load forecast interval and classify the forecast reliability level.

[0010] S4. Candidate control strategy generation: The reinforcement learning module generates several candidate control strategies based on the comprehensive prediction credibility score.

[0011] S5. Digital twin model simulation verification: Several candidate control strategies are input into the digital twin model for virtual testing. The digital twin model simulates the operating state of the central air conditioning system and outputs the simulation indicators and verification results after the control strategies are executed.

[0012] S6. Risk assessment and security gating of control strategies: Risk assessment is performed on several candidate control strategies. The risk levels of several candidate control strategies are divided by a preset risk threshold. It is determined whether the risk level of a candidate control strategy is lower than the preset security threshold. If so, the current control strategy is executed. If not, another control strategy with a risk level lower than the security threshold is selected for execution.

[0013] S7. Control strategy execution monitoring: Monitor the operating status of the central air conditioning system, analyze the current operating parameters in real time, and determine whether the deviation between the current operating parameters and the predicted operating parameters is greater than the preset deviation threshold. If not, continue to execute the current control strategy; if so, execute another control strategy to restore the central air conditioning system to stable operation.

[0014] S8; Digital twin model update; Correct the digital twin model based on the deviation between the current operating parameters and the predicted operating parameters.

[0015] Preferably, the operation and environment data includes load data, environmental data, and time data; the load forecast intervals at different time scales include short-term load forecast intervals and medium-term load forecast intervals.

[0016] Preferably, S3 specifically includes the following steps:

[0017] S3.1 Calculate the reliability of each load forecast interval.

[0018] S3.2 Determine the consistency of different load forecast intervals.

[0019] S3.3. The reliability of each load forecast interval and the consistency results of different load forecast intervals are integrated to obtain a comprehensive forecast reliability score.

[0020] Preferably, S3.1 includes assessing whether each load forecast interval covers historical actual load values; assessing whether each load forecast interval has effective information; assessing whether each load forecast interval changes smoothly over time; and assessing whether there are abnormal jumps in the load forecast interval.

[0021] Preferably, S3.2 includes determining whether the short-term load forecast range falls within the medium-term load forecast range; and determining the degree of consistency between the short-term load forecast range and the medium-term load forecast range in terms of load change risk.

[0022] Preferably, in S7, when the deviation between the current operating parameters and the predicted operating parameters is greater than a preset deviation threshold, the control strategy of the previous moment is executed or the preset safe operating mode is used to drive the operation of the central air conditioning system.

[0023] Preferably, S4 specifically includes the following steps:

[0024] S4.1 Obtain the current operation and environmental data of the central air conditioning system.

[0025] S4.2 Obtain the forecast results and forecast confidence level for the short-term load forecast interval, and obtain the forecast results and forecast confidence level for the medium-term load forecast interval.

[0026] S4.3 The candidate control policy network outputs suggestions for the original continuous and discrete actions.

[0027] S4.4 Apply credibility intensity scaling to continuous actions; apply credibility permission constraints to discrete actions; apply physical boundary restrictions and basic engineering constraints to both continuous and discrete actions.

[0028] S4.5. Form candidate control strategies for the current control cycle.

[0029] Preferably, S5 specifically includes the following steps:

[0030] S5.1. Use the current real state of the system as the initial state of the digital twin.

[0031] S5.2 Input external disturbance information, including the forecast results of the short-term load forecast range and the medium-term load forecast range, as well as outdoor environmental parameters.

[0032] S5.3. Use candidate control strategies as control inputs.

[0033] S5.4. Progressively advance the system state within the set simulation time domain.

[0034] S5.5 Output simulation metrics and verification results under candidate control strategies.

[0035] The simulation metrics include chilled water supply and return temperatures, cooling water supply and return temperatures, chilled water pump flow rate and cooling water pump flow rate, real-time power of the chiller, real-time cooling capacity of the chiller, total power of the central air conditioning system, and the deviation between the total power of the central air conditioning system and the predicted load.

[0036] The verification results include whether the candidate control strategy passes the feasibility verification, whether the candidate control strategy passes the stability verification, the total power prediction value under the candidate control strategy, the cooling matching result under the candidate control strategy, the changing trend of key state variables under the candidate control strategy, and the change of the number of devices under the candidate control strategy.

[0037] Preferably, S6 further includes a dual-condition judgment combining the comprehensive risk score and the comprehensive prediction credibility score. When the comprehensive prediction credibility score is greater than or equal to the minimum release threshold and the comprehensive prediction credibility score is less than or equal to the maximum allowable threshold, the current candidate control strategy is executed.

[0038] Compared with existing technologies, the beneficial effects of the central air conditioning system control strategy generation and verification method described in this invention are mainly reflected in the following aspects: This method introduces digital twin virtual verification, strategy risk classification, closed-loop monitoring switching and model dynamic correction to form a virtual-real integrated closed-loop intelligent control system, which significantly improves the system's safety, stability, energy efficiency and adaptability.

[0039] Candidate control strategies are input into a digital twin model for virtual simulation, predicting changes in temperature, flow rate, and energy consumption, as well as potential anomalies, to ensure the safe and stable operation of equipment. A quantitative risk assessment is conducted on candidate strategies, and risk levels are assigned, allowing only low-risk strategies to be executed. Through triple safety filtering of prediction reliability, virtual testing, and risk grading, high-risk aggressive actions are effectively intercepted, reducing frequent chiller start-ups and shutdowns and equipment overload losses. Stable operation is maintained even under complex conditions such as sudden load changes and extreme weather, significantly improving safety redundancy. A real-time deviation monitoring mechanism between actual operating parameters and digital twin predicted parameters is established. When the deviation exceeds a threshold, it automatically switches to a backup control strategy for rapid correction and return to stability. This effectively addresses sudden changes in outdoor temperature and humidity, significantly reduces indoor temperature fluctuations, continuously and stably ensures environmental comfort, and reduces reliance on manual intervention. Attached Figure Description

[0040] The above and other objects, features, and advantages of the invention will become clearer through a more detailed description of the preferred embodiments illustrated in the accompanying drawings. The same reference numerals denote the same parts throughout the drawings, and the drawings are not intentionally drawn to scale with actual dimensions; the focus is on illustrating the gist of the invention.

[0041] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0042] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention. In this embodiment, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the present invention.

[0043] It should be noted that when one element is considered to be "connected" to another element, it can be directly connected to and integrated with the other element, or there may be an intervening element present. The terms "mounted," "one end," "the other end," and similar expressions used in this invention are for illustrative purposes only.

[0044] like Figure 1 As shown, a method for generating and verifying a control strategy for a central air conditioning system includes the following steps:

[0045] S1. Data Acquisition: Acquire operational and environmental data for the central air conditioning system. This data includes load data, environmental data, equipment status data, system thermal data, and time data. The load data includes historical load time series data; the environmental data includes outdoor dry-bulb temperature, outdoor wet-bulb temperature, and outdoor air enthalpy; these reflect the impact of external meteorological conditions on the sensible and latent heat loads of the central air conditioning system; the time data includes current date, time period, day of the week, and other time-related features. All types of input data are combined chronologically to form a prediction input sample, which is used for subsequent multi-timescale load prediction model forecasting and digital twin model virtual testing.

[0046] The equipment includes three chillers, three cooling towers, three chilled water pumps, and three cooling pumps; the equipment status data includes the start / stop status and real-time power of the chillers, as well as their real-time cooling capacity, the adjustment frequency and power of the chilled water pumps, the adjustment frequency and power of the cooling pumps, and the adjustment frequency and power of the cooling towers.

[0047] The system thermal data includes chilled water supply temperature, chilled water return temperature, cooling water supply temperature, cooling water return temperature, and chilled water pump flow rate.

[0048] In one embodiment, data preprocessing includes time alignment, missing value handling, outlier identification and correction, and normalization. Time alignment maps all input data to a time granularity consistent with the control cycle, preferably 5 minutes. Missing value handling uses linear interpolation or forward imputation for short-term missing data, while discarding large, continuous missing samples. Outlier identification and correction identifies data exceeding the physical boundaries of the equipment, such as abnormal fluctuations in water supply temperature, sudden changes in pump flow, and significant inconsistencies between equipment start / stop status and operating parameters. Normalization normalizes or standardizes continuous variables to reduce the impact of different dimensions on model training; the chiller start / stop status is encoded using 0 / 1 codes.

[0049] S2. Multi-timescale load forecasting: Construct a multi-timescale load forecasting model to predict load changes at different time scales in the future based on operational and environmental data, and generate load forecasting intervals at different time scales, including short-term load forecasting intervals and medium-term load forecasting intervals.

[0050] The multi-timescale load forecasting model can be a recurrent neural network, a gated recurrent unit network, a long short-term memory network, or a combination of the three. In a preferred embodiment, the multi-timescale load forecasting model is a recurrent neural network.

[0051] The short-term load forecast interval corresponds to the load change in the next 15 to 30 minutes, and is used to support the rapid adjustment of continuous variables such as chilled water supply and return temperatures, chilled water pump frequency, and cooling water pump frequency; the medium-term load forecast interval corresponds to the load change in the next 60 to 120 minutes, and is used to assist in the determination of start-up and shutdown constraints of the three chillers.

[0052] S3. Forecast Reliability Assessment; Calculate the comprehensive forecast reliability score for the load forecast interval and classify the forecast reliability level. Specifically, this includes the following steps:

[0053] S3.1 Single-scale reliability calculation; calculate the reliability of each load forecast interval. This includes assessing whether each load forecast interval covers historical actual load; assessing whether each load forecast interval has effective information; assessing whether each load forecast interval has stability over time; and assessing whether there are abnormal jumps in the load forecast interval.

[0054] S3.2 Multi-scale consistency assessment; assessing the consistency of different load forecast intervals. This includes determining whether the short-term load forecast interval falls within the medium-term load forecast interval; and assessing the degree of consistency between the short-term and medium-term load forecast intervals in terms of load change risk. In this embodiment, the degree of overlap between the short-term and medium-term load forecast intervals is used to determine whether the load forecast interval falls within the medium-term load forecast interval; and the degree of consistency between the short-term and medium-term load forecast intervals in terms of load change risk is determined by assessing whether the actual load simultaneously covers both the short-term and medium-term load forecast intervals.

[0055] Calculate the multi-scale consistency score based on the multi-scale consistency assessment. The multi-scale consistency score is calculated by weighted summation of the overlap score and the risk synchronization score; the overlap score is obtained by the overlap ratio between the short-term load forecast interval and the medium-term load forecast interval; the risk synchronization score is calculated by comparing whether the actual load is simultaneously covered or exceeded by the short-term load forecast interval and the medium-term load forecast interval; the overlap score and the risk synchronization score are both existing technologies and will not be elaborated here.

[0056] S3.3. The reliability of each load forecast interval and the consistency results of different load forecast intervals are integrated to obtain a comprehensive forecast reliability score.

[0057] S3.4 The forecast reliability level includes high reliability level, medium reliability level and low reliability level; it is divided according to the score of the comprehensive forecast reliability score and the conflict relationship between the short-term load forecast interval and the medium-term load forecast interval.

[0058] Specifically, coverage assessment: Based on the actual load data within a preset historical time window, calculate the coverage rate of the load forecast interval to the historical actual load; the historical time window is the continuous 7 to 30 days of operating data before the current moment; the ratio of the number of times the actual load value falls into the load forecast interval within the historical time window to the total number of statistical counts is used as the coverage rate; when the coverage rate is greater than the preset coverage rate threshold, the load forecast interval is determined to have coverage effectiveness.

[0059] Effective information content assessment: Calculate the relative ratio between the width of the load forecast interval and the center value of the forecast interval; where the interval width is the difference between the upper and lower boundaries of the forecast interval, and the interval center value is the average of the upper and lower boundaries of the forecast interval; when the ratio of the load forecast interval width to the interval center value is less than a preset width threshold, the load forecast interval is determined to have effective information content.

[0060] Stationarity assessment: Based on the changes in the upper and lower boundaries of the prediction interval at consecutive time points, the rate of change of the interval boundary at adjacent time points is calculated; the rate of change is the ratio of the change in the upper boundary of the interval at adjacent time points to the upper boundary of the interval at the previous time point, and the ratio of the change in the lower boundary of the interval at adjacent time points to the lower boundary of the interval at the previous time point; when the rate of change is less than a preset rate of change threshold, the change process of the prediction interval is determined to be stationary.

[0061] Abnormal jump assessment: When the rate of change of the boundary between adjacent time intervals is greater than the preset abnormal jump threshold, an abnormal jump is determined to exist.

[0062] The overall credibility score is expressed as follows:

[0063]

[0064] To assess the reliability of short-term load forecast intervals, To assess the reliability of the medium-term load forecast range, For multi-scale consistency scores, For the corresponding weights. The preset weights.

[0065] The high reliability level indicates that there is no significant conflict between the short-term load forecast interval and the medium-term load forecast interval (the trends of the forecast results of the short-term load forecast interval and the medium-term load forecast interval are consistent), the current and future load trends are clear, and the forecast results can be used as a reliable basis for control.

[0066] The aforementioned medium confidence level indicates a certain degree of conflict between the short-term and medium-term load forecast intervals (the trends of the forecast results for the short-term and medium-term load forecast intervals deviate). The forecast results can be used as a reference, but should not be used to support overly aggressive control actions.

[0067] The low confidence level indicates a significant conflict between the short-term and medium-term load forecast intervals (the trends of the forecast results for the short-term and medium-term load forecast intervals are completely opposite), making the current forecast results unsuitable as a basis for aggressive control.

[0068] S4. Candidate Control Strategy Generation: The reinforcement learning module generates several candidate control strategies based on the comprehensive prediction confidence score. This includes the following steps:

[0069] S4.1 Obtain the current operation and environmental data of the central air conditioning system.

[0070] S4.2 Obtain the forecast results and forecast confidence level for the short-term load forecast interval, and obtain the forecast results and forecast confidence level for the medium-term load forecast interval.

[0071] S4.3 The candidate control policy network outputs suggestions for the original continuous and discrete actions.

[0072] S4.4 Apply credibility intensity scaling to continuous actions; apply credibility permission constraints to discrete actions; apply physical boundary restrictions and basic engineering constraints to both continuous and discrete actions.

[0073] S4.5. Form candidate control strategies for the current control cycle.

[0074] The candidate control strategies include: target setpoint for chilled water supply temperature, target setpoint for chilled water pump frequency, target setpoint for cooling tower fan frequency, adjustment suggestions for the number of chiller units in operation, adjustment suggestions for the number of cooling towers in operation, and adjustment suggestions for the number of chilled water pumps in operation.

[0075] The target setpoints for chilled water supply temperature, chilled water pump frequency, and cooling tower fan frequency are continuous control variables.

[0076] The recommendations for adjusting the number of chiller units, cooling towers, chilled water pumps, and cooling pumps in operation are discrete combination recommendations.

[0077] S5. Digital twin model simulation verification: Several candidate control strategies are input into the digital twin model for virtual testing. The digital twin model simulates the operating state of the central air conditioning system and outputs the simulation indicators and verification results after the control strategies are executed.

[0078] The digital twin model includes a thermodynamic sub-model, a device performance sub-model, an environmental interaction sub-model, current system state variables, external disturbance variables, and candidate control strategy variables; the thermodynamic sub-model, device performance sub-model, and environmental interaction model are existing technologies and will not be described in detail here.

[0079] The current system state variables reflect the current operating conditions of the system; external disturbance variables mainly include environmental parameters and predicted load information; the thermodynamic sub-model describes the temperature, flow rate, and heat transfer relationship between the chiller and cooling sides, reflecting the system's heat and cold balance and cooling response characteristics; the equipment performance sub-model describes the power, cooling capacity, flow rate, and efficiency characteristics of chillers, chilled pumps, cooling pumps, and cooling towers under different operating conditions; the environmental interaction sub-model describes the impact of environmental factors such as outdoor dry-bulb temperature, wet-bulb temperature, and enthalpy on the cooling side's heat dissipation capacity and system load changes.

[0080] The state evolution process of a digital twin model can be expressed as follows:

[0081]

[0082] This represents the system state vector at time t. Represents the candidate control strategy at time t; Let represent the external disturbance vector at time t; θ represent the set of parameters of the digital twin model; and F represent the digital twin state transition function. In this embodiment, the state vector... Ideally, this should include system thermal data and equipment status data. Control vector. Composed of candidate control strategies; external disturbance vector, This includes outdoor dry-bulb temperature, outdoor wet-bulb temperature, outdoor air enthalpy, forecast results for the short-term load forecast range, and forecast results for the medium-term load forecast range.

[0083] S5 specifically includes the following steps:

[0084] S5.1. Use the current real state of the system as the initial state of the digital twin.

[0085] S5.2 Input external disturbance information, including the forecast results of the short-term load forecast range and the medium-term load forecast range, as well as outdoor environmental parameters.

[0086] S5.3. Use candidate control strategies as control inputs.

[0087] S5.4. Progressively advance the system state within the set simulation time domain.

[0088] S5.5 Output simulation metrics and verification results under candidate control strategies.

[0089] The simulation metrics include chilled water supply and return temperatures, cooling water supply and return temperatures, chilled water pump flow rate and cooling water pump flow rate for several future steps, real-time power of the three chillers, real-time cooling capacity of the three chillers, total power of the central air conditioning system, and the deviation between the total power of the central air conditioning system and the predicted load.

[0090] The simulation time domain optimization is matched with the candidate control strategy. Specifically, when verifying continuous control variables, the next 3 to 6 control cycles can be optimized. When verifying the unit number switching strategy, the simulation window can be appropriately extended to a longer simulation window to observe the response trend after the switching of chiller, chilled pump, cooling pump and cooling tower.

[0091] During implementation, candidate control strategies must be constrained according to the actual control boundaries. Specifically, the chilled water supply temperature should always be maintained within the range of 7–13 ℃; the chilled water pump frequency should always be maintained within the range of 30–50 Hz; the cooling tower fan frequency should always be maintained within the range of 30–50 Hz; and the changes in the number of chillers, chilled water pumps, cooling pumps, and cooling towers should not exceed the actual equipment quantity boundary.

[0092] The verification results include whether the candidate control strategy passes the feasibility verification, whether the candidate control strategy passes the stability verification, the total power prediction value under the candidate control strategy, the cooling matching result under the candidate control strategy, the changing trend of key state variables under the candidate control strategy, and the change of the number of devices under the candidate control strategy.

[0093] S6. Risk assessment and security gating of control strategies: Perform a comprehensive risk score on several candidate control strategies, and determine whether the comprehensive risk score of the candidate control strategy is greater than or equal to the preset security threshold. If so, execute the current control strategy; otherwise, do not execute the current candidate control strategy.

[0094] The comprehensive risk score includes four types of risks, specifically: boundary exceedance risk, stability risk, cooling supply mismatch risk, and switching risk.

[0095] Boundary exceedance risk reflects whether water supply temperature, frequency, number of units, and equipment load rate are close to or exceed allowable ranges. Stability risk reflects whether key state variables exhibit significant oscillations or unstable trends within the simulation window. Cooling supply matching risk reflects the degree of deviation between the system's total cooling capacity and short-term predicted load, with a focus on preventing insufficient cooling. Switching risk reflects the frequency and intensity of changes in the number of chillers, cooling pumps, chilled water pumps, and cooling towers, constraining high-cost discrete actions.

[0096] The comprehensive risk score is expressed as follows:

[0097]

[0098] in, This indicates the risk of exceeding the boundary limit; Indicates stability risk; This indicates a risk of mismatch between cooling supply and demand; This indicates a risk of switching. The corresponding weights are determined by the system's operational stability. In this embodiment, if greater emphasis is placed on system stability, the weights of stability risk and switching risk are increased; if greater emphasis is placed on load assurance, the weight of cooling supply matching risk is increased.

[0099] Risk of exceeding boundaries The calculation method is as follows: Set upper and lower safety limits for each monitoring parameter, such as water supply temperature, frequency, number of operating units, and equipment load rate; assess each real-time parameter individually; if a parameter falls within the safe range, the individual risk is 0; if it falls below the lower limit or exceeds the upper limit, calculate the individual over-limit loss based on the relative deviation, with larger deviations resulting in higher losses; assign corresponding weights to each type of monitoring parameter, sum the over-limit losses of all parameters using a weighted average, and then normalize the results by combining them with the historical maximum over-limit loss to obtain the boundary over-limit risk value for the 0~1 range, with larger values ​​indicating higher over-limit risk.

[0100] Stability risk The calculation method is as follows: Extract time-series data of key states within the current simulation window, calculate the standard deviation of the time-series data to characterize the amplitude of fluctuations and oscillations; perform linear fitting on the sequence, and take the absolute value of the slope to characterize the long-term drift trend. Set weights for fluctuation and trend to fuse and obtain the univariate stability risk, then sum the weighted values ​​of all key variables, and normalize using the historical maximum instability index to output a stability risk score of 0 to 1. A higher score indicates more prominent oscillations and offsets in the equipment's operation within the window.

[0101] Cooling supply matching risks The calculation method is as follows: Extract the maximum total cooling capacity currently available to the system and the predicted peak load for the short term; subtract the total cooling capacity from the predicted peak load. If the result is less than or equal to 0, it indicates sufficient cooling capacity and a cooling supply mismatch risk of 0; if the difference is greater than 0, it indicates a cooling supply gap. Ratio the gap value to the predicted peak load and normalize the result to obtain a cooling supply mismatch risk of 0 to 1. The closer the value is to 1, the larger the cooling supply gap and the higher the risk of insufficient load guarantee.

[0102] Switching risks The calculation method involves: counting the total number of times the number of chillers, water pumps, and cooling towers are adjusted, started, or stopped within the sliding window; simultaneously, counting the number of units changed in each adjustment and calculating the average value; setting calculation weights for switching frequency and switching intensity; and normalizing and merging the number of switching times and the scale of a single change. The final result is a switching risk value in the range of 0 to 1. A higher score indicates more frequent equipment start-up and shutdown adjustments, larger single change magnitudes, and higher switching risk from discrete operations.

[0103] In a preferred embodiment, a dual-condition determination is made by combining a comprehensive risk score and a comprehensive prediction confidence score.

[0104]

[0105] in, This indicates that the candidate control policy is allowed to go live; This indicates that the candidate control strategy is not allowed to go live; The minimum release threshold is determined by the overall prediction credibility score. This represents the highest permissible threshold for the comprehensive risk score.

[0106] In the overall prediction credibility score Greater than or equal to the minimum release threshold, comprehensive risk score When the risk is less than or equal to the maximum allowable threshold, the current candidate control strategy is executed. This decision mechanism only executes the candidate control strategy when the prediction result itself is sufficiently reliable and the digital twin simulation results show that the strategy risk is within an acceptable range.

[0107] In this embodiment, candidate control strategies that meet the following criteria are directly executed: prediction credibility meets the release requirements, digital twin feasibility verification is passed, digital twin stability verification is passed, comprehensive risk score is lower than the set upper limit, and actions meet on-site execution constraints.

[0108] In this embodiment, candidate control strategies will not be executed if any of the following conditions exist: insufficient prediction confidence, obvious limit exceedance in simulation, insufficient cooling capacity prediction, failure to pass stability verification, excessive switching risk, or violation of on-site execution logic.

[0109] In a preferred embodiment, the method further includes a differential determination of continuous actions and discrete actions.

[0110] Continuous Action Strategy: For candidate control strategies that only involve adjustments to chilled water supply temperature, chilled water pump frequency, and cooling tower fan frequency, relatively lenient gating rules can be adopted. As long as the overall prediction reliability meets the standard, the overall risk score is controllable, and no obvious limit exceedances are triggered, the corresponding candidate control strategy can be executed.

[0111] Discrete Action Strategy: For candidate control strategies involving changes in the number of chillers, cooling towers, chilled water pumps, or cooling pumps, stricter gating conditions are adopted. Even when the overall prediction reliability meets the standard, the overall risk score is controllable, and no significant limit violations are triggered, it is still necessary to check the switching risk, cooling supply matching risk, and minimum start-stop interval constraints for the equipment.

[0112] Specifically, if the continuous action part of the candidate control strategy performs well, but the discrete action part has a high switching risk or insufficient start / stop conditions, the system will downgrade and allow the action to proceed; the system will freeze the discrete action and only execute the candidate control strategy corresponding to the continuous action part.

[0113] Downgraded release is applicable in the following situations: insufficient suggested conditions for changes in the number of units but beneficial fine-tuning of water supply temperature and pump frequency; high risk of changes in the number of cooling towers or pumps but clear benefits of current continuous action optimization; and a comprehensive risk score slightly above the safety threshold, but with the main risks concentrated in switching actions. The downgraded release mechanism can retain some optimization effects while ensuring system safety.

[0114] S7. Control strategy execution monitoring: Monitor the operating status of the central air conditioning system, analyze the current operating parameters in real time, and determine whether the deviation between the current operating parameters and the predicted operating parameters is greater than the preset deviation threshold. If not, continue to execute the current control strategy; if so, execute another control strategy to restore the central air conditioning system to stable operation.

[0115] Preferably, when the deviation between the current operating parameters and the predicted operating parameters exceeds a preset deviation threshold, the control strategy from the previous moment is executed or a preset safe operating mode is used to drive the central air conditioning system. This prioritizes restoring the stability of the central air conditioning system and freezes high-risk discrete actions. It reverts to a verified stable operating state, ensuring that cooling capacity is not lost.

[0116] The control strategy from the previous moment is executed, including rollback strategies for different control variables and rollback strategies for different anomaly levels.

[0117] Rollback strategies for different control variables: different rollback approaches are adopted for continuous and discrete variables; for continuous variables, such as chilled water supply temperature setpoint, chilled water pump frequency, and cooling tower fan frequency, the values ​​are returned to the vicinity of the most recent stable value or the preset safe value, and a gradual recovery is adopted to avoid the formation of new disturbances.

[0118] For discrete variables, such as the number of chillers, cooling towers, chilled water pumps, and cooling pumps in operation, if the current combination of units can meet the basic load requirements, the number of units to be frozen will not be changed further; if the current combination of equipment itself is an abnormal source, it will be restored to the most recent stable combination of equipment.

[0119] After entering the rollback process, the system prioritizes freezing high-risk actions, including suggestions for changes in the number of chillers, cooling towers, chilled pumps, and cooling pumps. This can first block discrete switching actions that cause the greatest disturbance to the system, preventing the anomaly from continuing to expand at the equipment combination level.

[0120] After freezing high-risk discrete actions, the system begins to gradually restore continuous control variables to the setpoints required by the safety rollback strategy. Specifically, this includes gradually restoring the chilled water supply temperature to the rollback target value, the chilled water pump frequency to the rollback target value, and the cooling tower fan frequency to the rollback target value. During the recovery process, the maximum adjustment step size limit per cycle should still be observed to avoid causing the system to experience another shock.

[0121] Rollback strategies for different levels of anomaly: Anomaly levels include mild anomaly, moderate anomaly, and severe anomaly.

[0122] Mild anomalies, such as slight deviations or short-term fluctuations. Rollback method: Freeze discrete actions and restore continuous variables to near the nearest stable control point.

[0123] Moderate anomalies, such as increased deviations over multiple consecutive cycles, or continuous deterioration of flow rate and temperature difference. Rollback method: Switch to the current conservative baseline strategy and pause reinforcement learning action output.

[0124] Severe anomalies: such as boundary violations, chiller malfunctions, and continuous switching anomalies. Rollback method: Enter a preset safety mode, freezing all high-risk actions and retaining only the basic operating combinations necessary to ensure cooling supply and system safety.

[0125] S8; Digital twin model update; Correct the digital twin model based on the deviation between the current operating parameters and the predicted operating parameters.

[0126] The following are examples:

[0127] At time t, the system first collects operational and environmental data for the current control cycle, including outdoor dry-bulb temperature, outdoor wet-bulb temperature, outdoor air enthalpy, chilled water supply and return temperatures, cooling water supply and return temperatures, chilled water pump and cooling water pump flow rates, real-time power and cooling capacity of the three chillers, chilled water pump frequency, cooling tower fan frequency, and the start / stop status of various equipment. This data, along with historical load sequences, is input into the load forecasting module. The forecasting module outputs short-term and medium-term load forecasts, along with corresponding forecast intervals.

[0128] Subsequently, the system performs a comprehensive reliability assessment of the current prediction results. If the comprehensive prediction reliability score is high, it indicates that the short-term load change trend is relatively reliable, and the system allows the candidate control strategy to output the control strategy with a higher action intensity; if the comprehensive prediction reliability score is at a medium level, the candidate control strategy is restricted to the vicinity of the current operating point; if the comprehensive prediction reliability score is low, the candidate control strategy is only allowed to make conservative fine-tuning, and discrete action suggestions are frozen if necessary.

[0129] After obtaining the overall confidence level of the prediction, the candidate control strategy reads the current operating and environmental data and outputs the original control actions, including suggestions for adjusting the chilled water supply temperature, adjusting the chilled water pump frequency, adjusting the cooling tower fan frequency, and suggestions for changing the number of chillers, pumps, and cooling towers.

[0130] Subsequently, the system scales the intensity of continuous actions based on the comprehensive prediction confidence score and imposes permission constraints on discrete actions. For example, in a low confidence scenario, the system suggests that the number of chillers can be directly frozen, while the chilled water supply temperature is only allowed to be slightly adjusted around the current value.

[0131] The constrained candidate control strategies are not directly implemented in the field; instead, they are first input into a digital twin simulation platform. The digital twin platform uses the current system state as initial conditions and the forecast results for the short-term and medium-term load forecast intervals, along with environmental variables, as disturbance inputs to virtually simulate the execution results of the candidate strategies over several future control cycles. It then outputs simulation metrics and verification results.

[0132] Simulation indicators should include at least: changes in chilled water supply and return temperatures, changes in cooling water supply and return temperatures, changes in flow rates of chilled water pumps and cooling water pumps, changes in chiller power and cooling capacity, and predicted total system power.

[0133] Next, the system verifies the feasibility and stability of the candidate strategies based on the verification results. If simulation indicators and verification results show that the candidate strategy will cause the chilled water supply temperature to exceed the limit, the chilled pump frequency to exceed the limit, the cooling capacity to be insufficient, or the key state variables to oscillate significantly, then the subsequent control strategy is determined to be unsuitable for online execution. If the candidate control strategy is generally effective, but the discrete action part has a high risk, the system can adopt a downgrade release method, retaining only the continuous variable adjustment part and freezing the high-risk unit switching action.

[0134] The system enters the actual system execution module through a safety-gated control strategy. During execution, the system does not need to issue the target value all at once, but rather prefers a gradual execution approach: for example, the chilled water supply temperature is adjusted by no more than 0.5 ℃ per control cycle, and the chilled water pump frequency and cooling tower fan frequency are adjusted by no more than 1–2 Hz per control cycle. If the number of chillers, pumps, or cooling towers is to be switched, the minimum start-stop interval and the current operating condition adaptation conditions must also be met.

[0135] During execution, the system continuously monitors the actual operating status and compares the current operating parameters with the predicted operating parameters. If any critical parameter exceeds its limit, system power abnormally increases, supply and return water temperatures significantly deviate from expectations, or operation deteriorates after equipment switching, the anomaly identification module is immediately triggered. If the anomaly is confirmed, the system suspends the current control strategy, freezes high-risk discrete actions, and invokes a preset safety rollback strategy to gradually restore continuous control variables to safe operating values.

[0136] After the current control cycle ends, the system will perform a deviation analysis between the actual operating results and the digital twin simulation results, and dynamically correct the thermodynamic sub-model, equipment performance sub-model, and environmental interaction sub-model of the digital twin model. The updated digital twin model will participate in the next round of candidate strategy verification, thus forming a continuous closed-loop optimization.

[0137] In this specification, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0138] In the description of this specification, the references to terms such as "preferred embodiment," "another embodiment," "other embodiment," or "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0139] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for generating and verifying control strategies for a central air conditioning system, characterized in that: Includes the following steps: S1. Data Acquisition; Acquiring operational and environmental data of the central air conditioning system; S2. Multi-timescale load forecasting: Construct a multi-timescale load forecasting model to predict load changes at different time scales in the future based on operational and environmental data, and generate load forecasting intervals at different time scales. S3. Forecast Reliability Assessment: Calculate the comprehensive forecast reliability score for the load forecast interval and classify the forecast reliability level; S4. Candidate Control Strategy Generation: The reinforcement learning module generates several candidate control strategies based on the comprehensive prediction confidence score. S5. Digital twin model simulation verification: Input several candidate control strategies into the digital twin model for virtual testing. The digital twin model simulates the operating state of the central air conditioning system and outputs the simulation indicators and verification results after executing the control strategies. S6. Risk assessment and security gating of control strategies: A comprehensive risk score is calculated for several candidate control strategies. It is determined whether the comprehensive risk score of the candidate control strategy is greater than or equal to the preset security threshold. If so, the current control strategy is executed; otherwise, the current candidate control strategy is not executed. S7. Control strategy execution monitoring: Monitor the operating status of the central air conditioning system, analyze the current operating parameters in real time, and determine whether the deviation between the current operating parameters and the predicted operating parameters is greater than the preset deviation threshold. If not, continue to execute the current control strategy; if so, execute another control strategy to restore the central air conditioning system to stable operation. S8; Digital twin model update; Correct the digital twin model based on the deviation between the current operating parameters and the predicted operating parameters.

2. The method for generating and verifying control strategies for a central air conditioning system according to claim 1, characterized in that: The operational and environmental data include load data, environmental data, and time data; the load forecast intervals at different time scales include short-term load forecast intervals and medium-term load forecast intervals.

3. The method for generating and verifying control strategies for a central air conditioning system according to claim 2, characterized in that: S3 specifically includes the following steps: S3.1 Calculate the confidence level of each load forecast interval; S3.2 Determine the consistency of different load forecast intervals; S3.

3. The reliability of each load forecast interval and the consistency results of different load forecast intervals are integrated to obtain a comprehensive forecast reliability score.

4. The method for generating and verifying control strategies for a central air conditioning system according to claim 3, characterized in that: S3.1 includes assessing whether each load forecast interval covers historical actual load values; assessing whether each load forecast interval has effective information; assessing whether each load forecast interval changes smoothly over time; and assessing whether there are abnormal jumps in the load forecast interval.

5. The method for generating and verifying control strategies for a central air conditioning system according to claim 3, characterized in that: S3.2 includes determining whether the short-term load forecast range falls within the medium-term load forecast range; and determining the degree of consistency between the short-term and medium-term load forecast ranges in terms of load change risk.

6. The method for generating and verifying control strategies for a central air conditioning system according to claim 1, characterized in that: In S7, when the deviation between the current operating parameters and the predicted operating parameters is greater than the preset deviation threshold, the control strategy of the previous moment is executed or the preset safe operation mode is used to drive the operation of the central air conditioning system.

7. The method for generating and verifying control strategies for a central air conditioning system according to claim 2, characterized in that: S4 specifically includes the following steps: S4.1 Obtain the current operation and environmental data of the central air conditioning system; S4.2 Obtain the forecast results and forecast confidence level for the short-term load forecast interval, and obtain the forecast results and forecast confidence level for the medium-term load forecast interval. S4.3 The candidate control strategy network outputs suggestions for the original continuous and discrete actions; S4.4 Apply credibility intensity scaling to continuous actions; apply credibility permission constraints to discrete actions; apply physical boundary restrictions and basic engineering constraints to both continuous and discrete actions. S4.

5. Form candidate control strategies for the current control cycle.

8. The method for generating and verifying control strategies for a central air conditioning system according to claim 1, characterized in that: S5 specifically includes the following steps: S5.

1. Use the current real state of the system as the initial state of the digital twin; S5.2 Input external disturbance information, including the forecast results of the short-term load forecast range and the medium-term load forecast range, as well as outdoor environmental parameters; S5.3, Use candidate control strategies as control inputs; S5.

4. Progressively advance the system state within the set simulation time domain; S5.5 Output simulation metrics and verification results under candidate control strategies; The simulation metrics include chilled water supply and return temperatures, cooling water supply and return temperatures, chilled water pump flow rate and cooling water pump flow rate, real-time power of the chiller, real-time cooling capacity of the chiller, total power of the central air conditioning system, and the deviation between the total power of the central air conditioning system and the predicted load. The verification results include whether the candidate control strategy passes the feasibility verification, whether the candidate control strategy passes the stability verification, the total power prediction value under the candidate control strategy, the cooling matching result under the candidate control strategy, the changing trend of key state variables under the candidate control strategy, and the change of the number of devices under the candidate control strategy.

9. The method for generating and verifying control strategies for a central air conditioning system according to claim 1, characterized in that: S6 also includes a dual-condition judgment that combines the comprehensive risk score and the comprehensive prediction credibility score. When the comprehensive prediction credibility score is greater than or equal to the minimum release threshold and the comprehensive prediction credibility score is less than or equal to the maximum allowable threshold, the current candidate control strategy is executed.

Images