An ai optimization method and system for building automation load adding and subtracting strategy-oriented cold water unit group control threshold

Abstract

This invention discloses an AI optimization method and system for chiller group control thresholds based on building automation (BA) load increase / decrease strategies. The method acquires chiller room operating data, BA system control data, and external environmental data. It maps the load increase / decrease thresholds, supply water temperature settings, differential pressure settings, dead zones, hold times, rate of change limits, and backoff values ​​from different BA systems into a unified parameter vector. Based on state characteristics and candidate future parameter sequences, it predicts future water temperature, load, and power trajectories and performs rolling optimization using BA writable parameters as decision variables. The optimized parameters are then written into the BA system after verification against upper and lower limits, rate of change, hold time, faults, communication, and manual locking. The original BA logic then executes device control, restoring safety thresholds in case of anomalies. This scheme does not directly control equipment start / stop, reducing the risk of temperature hysteresis oscillations and frequent start / stop cycles.

Description

Technical Field

[0001] This invention relates to the fields of heating, ventilation and air conditioning, industrial refrigeration, building automation and edge intelligent control technology, and in particular to an AI optimization method and system for chiller group control thresholds for building automation load increase and decrease strategies. Background Technology

[0002] Large commercial buildings, data centers, hospitals, rail transit stations, and industrial plants typically have multiple chiller units, multiple chilled water pumps, cooling water pumps, and cooling towers. Existing chiller rooms are usually controlled by building automation systems, PLCs, or DDCs. The mainstream control method is to set fixed thresholds for adding or removing chillers based on indicators such as the temperature difference between chilled water supply and return, flow rate, supply water temperature, pressure difference, and chiller load rate, and to use PID control to adjust the supply water temperature or pressure difference.

[0003] Different building automation system manufacturers have differences in load increase / decrease logic, point naming, writable permissions, threshold meanings, interlocking methods, and fault rollback methods. Common logics include adding units based on load rate, adding units based on water supply temperature deviation, adding units based on duration confirmation, deleting units based on low load rate, protecting units based on minimum start / stop time, and selecting units based on a shift queue.

[0004] While the above methods offer high engineering safety, fixed thresholds are difficult to adapt to future weather, personnel load, equipment efficiency, and electricity price changes, which can easily lead to problems such as lagging water supply temperature, frequent start-ups and shutdowns under critical loads, and excessively high peak total power in the computer room.

[0005] If the artificial intelligence control algorithm is directly sent to the equipment start / stop point or the frequency setpoint of the frequency converter, it may bypass the interlocking, protection, manual intervention and fault rollback logic that has been verified in the on-site building automation system or PLC, which poses an engineering safety risk.

[0006] Therefore, a chiller group control optimization technology solution is needed that can utilize predictive optimization capabilities without compromising the safety boundaries of on-site building automatic load increase and decrease. Summary of the Invention

[0007] The purpose of this invention is to provide an AI optimization method and system for chiller group control thresholds for building automation load increase / decrease strategies, which solves the problems of fixed threshold response lag, difficulty in adapting to different building automation systems, high risk of direct AI control of equipment start-up and shutdown, and lack of future load forecasting.

[0008] To achieve the above objectives, the present invention does not directly output start / stop commands to chillers, water pumps, or cooling towers, but instead limits the prediction and optimization results to control thresholds or set parameters that the building automation system is allowed to write.

[0009] The on-site building automation system or PLC continues to execute the original interlocking, shift, minimum start / stop time, fault shielding, and safety backoff logic.

[0010] This invention abstracts the load-reduction strategies of different building automation manufacturers into a unified parameter vector through a strategy adaptation template, and periodically optimizes the threshold of the building automation system after safety filtering by injecting a multi-step prediction model and a rolling time-domain model predictive control optimizer with future control parameters.

[0011] The method of the invention includes: acquiring chiller room operation data, building automation system control data, and external environment data; establishing a strategy adaptation template and mapping on-site load increase / decrease points to a unified parameter vector; aligning, cleaning, and normalizing the data to form a state feature vector; inputting the candidate future parameter vector sequence into a multi-step state prediction model to predict future water temperature, load, and power trajectories; establishing an objective function using the writable parameters of the building automation system as decision variables and solving for the optimized parameter sequence; performing safety filtering on the current period parameters; writing the parameters into the building automation system after passing the safety filtering, otherwise maintaining or restoring the safety threshold.

[0012] The unified parameter vector may include one or more of the following: load factor threshold for adding a chiller, load factor threshold for reducing a chiller, chilled water supply temperature setpoint, manifold differential pressure setpoint, allowable temperature dead zone, confirmation holding time for adding a chiller, confirmation holding time for reducing a chiller, maximum change range per cycle, and safety backoff value.

[0013] The multi-step state prediction model can be a sequence-to-sequence model, a recurrent neural network model, a gated recurrent unit model, a long short-term memory network model, a temporal convolutional network model, a Transformer temporal model, or a combination of the above models. The candidate future parameter vector sequence serves as the input to the future control features during the decoding stage, ensuring that the predicted trajectory corresponds to the candidate control parameters.

[0014] The model predictive control objective function may include energy consumption terms, peak power terms, temperature exceedance penalty terms, equipment start-up / shutdown penalty terms, parameter switching penalty terms, and safety boundary penalty terms. The optimization algorithm may be a genetic algorithm, particle swarm optimization algorithm, simulated annealing algorithm, mixed integer programming algorithm, Bayesian optimization algorithm, or enumeration search algorithm.

[0015] The safety filtering may include parameter upper and lower limit verification, parameter change rate verification, minimum hold time verification, equipment fault status verification, communication status verification, manual lock status verification, local / remote control rights verification, water flow switch status verification, and maintenance shielding status verification.

[0016] Compared with AI-powered group control solutions that directly control the start and stop of equipment, this invention limits the control output to the writable thresholds and set parameters of the building automation system, without bypassing the on-site building automation system or PLC safety interlocks, thus reducing the risks of project implementation.

[0017] Compared with fixed threshold building automation group control schemes, this invention can adjust the threshold for adding and subtracting machines and the set parameters on a rolling basis according to future weather, load, equipment status and electricity price weights, reduce water supply temperature lag oscillation, reduce short cycle start-stop under critical load, and improve cross-project reuse capability.

[0018] Compared to schemes that rely solely on historical data for prediction, this invention inputs candidate future control parameters into the prediction model, thereby coupling the prediction results with control decisions and improving the accuracy of predicting the system response after active threshold adjustment.

[0019] Figure 1 is a diagram of the overall structure of the system of the present invention.

[0020] Figure 2 is a schematic diagram of the closed loop of building automation parameter adaptation and threshold writing in this invention.

[0021] Figure 3 is a flowchart of the prediction and model predictive control optimization of the injected future control parameters in this invention.

[0022] Figure 4 is a schematic diagram of the safety filtering and rollback state machine of the present invention.

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0024] As shown in Figure 1, the system of this invention includes an edge computing device, a building automation system, a PLC or DDC, sensors and meters, an external environment data interface, and controlled equipment such as chillers, water pumps, and cooling towers. The edge computing device communicates with the building automation system, PLC, sensor gateway, and smart meters via industrial Ethernet or fieldbus, and performs data acquisition, cleaning, prediction, optimization, security filtering, and parameter writing locally.

[0025] Edge computing devices can employ ARM, x86, or other embedded computing platforms, running Linux, Windows, real-time operating systems, or other operating systems capable of implementing the functions of this invention. The hardware platform, operating system, and specific processor model are merely examples and do not constitute a limitation on the scope of protection.

[0026] During the project commissioning phase, the building automation system's point table is read to determine readable and writable points. Readable points include chilled water supply and return temperatures, cooling water supply and return temperatures, flow rate, differential pressure, unit status, fault status, main unit power, pump frequency, and current building automation thresholds. Writable points include load factor thresholds for adding and removing units, supply water temperature settings, differential pressure settings, control dead zones, confirmation hold time, and backoff trigger flags. Each point is configured with a unit, upper and lower limits, default values, write cycle, and communication address.

[0027] As shown in Figure 2, the strategy adaptation module establishes a strategy adaptation template based on the load addition and reduction logic of the on-site building automation system. For building automation systems that add units based on load rate, the load rate threshold and the confirmation and hold time for adding units are mapped to a unified parameter vector; for building automation systems that add units based on water supply temperature deviation, the water supply temperature setpoint, temperature dead zone, and confirmation and hold time are mapped to a unified parameter vector; for building automation systems that control based on differential pressure or flow rate, the differential pressure setpoint or flow rate level is mapped to a unified parameter vector.

[0028] The data cleaning module timestamps and aligns operational data, control data, and external environmental data according to a unified time base. Outlier removal is performed on data exceeding the physical rate of change, interpolation or hold-up processing is performed on short-term communication packet loss data, and temperature, flow rate, power, thresholds, and setpoints are normalized to form a state feature vector for the current control cycle.

[0029] The predictive model takes into account historical state sequences, environmental prediction data, current building automation parameters, and candidate future parameter sequences as inputs. The model outputs the supply water temperature, return water temperature, cooling load, and power trajectory for the next K time steps. The candidate future parameter sequences are input during the decoding phase, enabling the model to simulate the system's dynamic response after adopting a set of future building automation thresholds.

[0030] As shown in Figure 3, the model predictive control optimizer uses the building automation parameter sequence for the next K time steps as decision variables to calculate the objective function value. The objective function can be expressed as a weighted sum of energy consumption, peak power, temperature exceedance penalty, equipment start / stop penalty, parameter switching penalty, and safety boundary penalty. After obtaining the optimized parameter sequence in the future rolling time domain, the optimizer extracts only the parameters from the first control cycle for current writing, and re-rolls the optimization based on the latest data in the next control cycle.

[0031] As shown in Figure 4, the current cycle parameters output by the optimizer need to be filtered for safety. If the parameter exceeds the manually set boundary, the change range exceeds the single cycle limit, the minimum hold time has not been reached since the last write, the device is in a fault or maintenance state, the building automation system is in a local control or manual lock state, or the communication health is insufficient, then the write will be rejected and the original value will be retained.

[0032] Once the current cycle parameters pass the security filter, the communication writing module writes the optimized parameters to the AV, BV, MSV, or corresponding register points of the building automation system via BACnet / IP, OPC UA, ModbusTCP, Profinet, or vendor-open APIs. The building automation system then executes equipment addition, reduction, hold, or rotation control according to its original load increase / decrease logic.

[0033] If the actual water temperature, power, or equipment status becomes abnormal after the parameters are written, the rollback module restores the threshold set by the last manual confirmation, the original threshold of the building automation system, or the preset seasonal safety threshold, and records the event and alarm information. In rollback mode, the edge computing device can stop writing and only maintain monitoring.

[0034] This invention may also include an online learning step. After the building automation system executes the optimized parameters, it collects the actual temperature trajectory, actual power trajectory, and actual equipment status, and updates the prediction model or corrects the model parameters based on the deviation between the actual trajectory and the predicted trajectory, thereby improving the subsequent prediction and optimization effects.

[0035] This invention can be applied to energy-saving retrofits of large commercial buildings, public buildings, industrial refrigeration plants, data center refrigeration plants, and park energy stations. The system can be deployed without replacing the existing building automation system or PLC control logic, making it industrially applicable.

Claims

1. An AI optimization method and system for chiller group control thresholds based on building automation load increase / decrease strategies, characterized in that, The process includes the following steps: S1. Acquire the operating data of the chiller room and the control data of the building automation system. The operating data includes at least the chilled water supply and return temperatures, cooling water supply and return temperatures, flow rate, differential pressure, unit operating status, unit power, pump frequency, and fault status. The control data includes at least the readable and writable points, current thresholds, setpoints, and interlocking statuses related to chiller unit load increase / decrease in the building automation system. S2. Establish a strategy adaptation template based on the load increase / decrease logic of the on-site building automation system, mapping the load increase / decrease thresholds, supply water temperature settings, differential pressure settings, control dead zones, minimum hold times, rate of change limits, and rollback default values ​​from different building automation manufacturers or projects to a unified parameter vector. S3. Perform timestamp alignment, outlier removal, missing value compensation, and feature normalization on the operating data, control data, and external environment data to form a state feature vector for the current control cycle. S4. Based on the state feature vector and... The candidate future parameter vector sequence is used to predict the trajectory of chilled water supply temperature, cooling load, and total power of the machine room in the future rolling time domain using a multi-step state prediction model. The candidate future parameter vector sequence is input into the multi-step state prediction model as a future control feature. S5. Using the writable parameter vector of the building automation system as the optimization decision variable, a model predictive control objective function is established. The objective function includes at least an energy consumption term, a temperature limit violation penalty term, an equipment start / stop or parameter switching penalty term, and a safety boundary penalty term. The objective function is solved based on the prediction results of the multi-step state prediction model to obtain the optimized parameter sequence in the future rolling time domain. S6. The first set of parameters corresponding to the current control cycle in the optimized parameter sequence is extracted, and safety filtering is performed on the first set of parameters. The safety filtering includes at least parameter upper and lower limit verification, parameter change rate verification, minimum hold time verification, equipment fault status verification, communication status verification, and manual lockout status verification. S7. When the first set of parameters passes the safety filter, they are written to the corresponding writable points of the building automation system, and the building automation system executes the chiller unit's addition, reduction, maintenance, or rotation control according to its original load increase / decrease logic; when the first set of parameters fails to pass the safety filter or an abnormality occurs after being written, the original threshold or preset safety threshold of the building automation system is maintained or restored.

2. The method according to claim 1, characterized in that, The strategy adaptation template includes a location adaptation table, which includes at least the location name, communication protocol address, data type, unit, read / write permissions, upper limit of value, lower limit of value, default value, safety fallback value, and parameter semantic tags. The parameter vector includes at least three of the following: added load rate threshold, reduced load rate threshold, chilled water supply temperature setpoint, manifold differential pressure setpoint, allowable temperature dead zone, added load confirmation hold time, reduced load confirmation hold time, and maximum change range per single cycle.

3. The method according to claim 1, characterized in that, The multi-step state prediction model is a sequence-to-sequence model, a recurrent neural network model, a gated recurrent unit model, a long short-term memory network model, a temporal convolutional network model, a Transformer temporal model, or a combination of the above models; the future control features are input together with the prediction output of the previous time step during the model decoding stage to obtain the future state trajectory corresponding to the candidate parameter sequence; the external environment data includes at least one of outdoor temperature, outdoor humidity, solar irradiance, weather forecast data, building schedule, population density estimate, and time-of-use electricity price.

4. The method according to claim 1, characterized in that, The model predictive control objective function also includes a peak power penalty term and an electricity price weight term, so that the optimizer outputs a parameter vector for peak shaving and valley filling under the premise of satisfying the chilled water supply temperature constraint; the optimization algorithm for solving the objective function in step S5 is any one or a combination of genetic algorithm, particle swarm optimization algorithm, simulated annealing algorithm, mixed integer programming algorithm, Bayesian optimization algorithm or enumeration search algorithm.

5. The method according to claim 1, characterized in that, The security filtering also includes verifying at least one of the following: local control status, remote control status, water flow switch status, chilled water differential pressure status, cooling water differential pressure status, unit alarm status, and maintenance shielding status of the building automation system; the writing to the building automation system is to write the optimization parameters to AV, BV, MSV, or corresponding register points via BACnet / IP, OPC UA, Modbus TCP, Profinet, or vendor open APIs.

6. The method according to claim 1, characterized in that, The method also includes an online learning step: after the building automation system executes the optimized parameters, it collects the actual temperature trajectory, actual power trajectory and actual equipment status, and updates the prediction model or correction model parameters with the deviation between the actual trajectory and the predicted trajectory; the rollback includes at least one of the following: restoring the threshold of the last manual confirmation, restoring the original threshold of the building automation system, restoring the preset seasonal safety threshold, stopping writing and only maintaining monitoring, and issuing an alarm to the operation and maintenance interface.

7. An AI optimization method and system for chiller group control thresholds based on building automation load increase / decrease strategies, characterized in that, include: The data acquisition module is used to acquire operating data of the chiller room, control data of the building automation system, and external environmental data; The strategy adaptation module maps load-reduction related points and parameters from different building automation systems to a unified parameter vector; the prediction module predicts the water temperature trajectory, load trajectory, and power trajectory in the future rolling time domain based on the state feature vector and candidate future parameter vector sequence; the optimization module solves the model predictive control objective function using the writable parameter vector of the building automation system as decision variables; and the safety filtering module verifies the optimization parameters for safety boundaries, change rates, hold times, equipment status, communication status, and manual lockout status. The communication writing module is used to write the optimized parameters into the corresponding points of the building automation system after the optimized parameters have passed the security filter. The rollback module is used to restore the original threshold or preset security threshold when verification fails, communication is abnormal, or operation is abnormal.

8. The system according to claim 7, characterized in that, The system is deployed on an edge computing device, which communicates locally with the chiller room PLC, sensors, meters, and building automation system, and performs prediction, optimization, security filtering, and parameter writing locally. The strategy adaptation module has multiple pre-set strategy templates from building automation manufacturers. Each strategy template includes conditions for adding or removing machines, rotation conditions, minimum start / stop constraints, a list of writable points, upper and lower limits of parameters, and default values ​​for rollback.

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

Images