Two-way collaborative regulation method for virtual power grid and refrigeration system
By collecting and preprocessing data, a cooling load model is established, the adjustable space is calculated using reinforcement learning algorithms, and an agent is trained under virtual power grid rules. This solves the problem of bidirectional coordinated regulation between the virtual power grid and the cooling system, realizes the coordinated optimization of power grid peak shaving and the cooling system, and ensures comfort and energy efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUXI CHAOS ENERGY TECH CO LTD
- Filing Date
- 2026-03-19
- Publication Date
- 2026-07-14
AI Technical Summary
Existing technologies struggle to accurately depict adjustable space over future timeframes while meeting comfort and equipment constraints. Furthermore, the virtual power grid and cooling system are controlled independently, making it difficult to achieve bidirectional closed-loop coordination.
By collecting parameters related to power generation, load, cooling system, and rules, data preprocessing and prediction are performed to establish a cooling load model. The adjustable space is calculated by combining reinforcement learning algorithms, and an intelligent agent is trained under virtual power grid rules to achieve bidirectional collaborative regulation.
It achieves the goal of meeting the peak-shaving needs of the power grid while ensuring indoor environmental comfort, reducing the energy consumption of the cooling system, and improving the stability of power grid operation and the energy utilization efficiency of the user side.
Smart Images
Figure CN122393995A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a bidirectional collaborative control method for virtual power grids and refrigeration systems, belonging to the fields of energy internet, demand response, building / industrial refrigeration control, and artificial intelligence collaborative control technology. Background Technology
[0002] With the increasing penetration of distributed renewable energy and the advancement of electricity market transactions, virtual power grids need to aggregate adjustable resources to participate in peak shaving, demand management, demand response, and ancillary services. However, building cooling systems (chimneys, chilled water pumps, cooling water pumps, cooling tower fans, etc.) have significant thermal inertia and comfort constraints, and their adjustability changes dynamically with external weather, building heat load, and equipment status.
[0003] Existing technologies typically employ fixed strategies or rule-based start / stop and setpoint control, which makes it difficult to accurately characterize the adjustable space over future time periods while meeting comfort and equipment constraints. Furthermore, the regulation of the virtual power grid and the optimization of the building's cooling system are carried out independently, making it difficult to achieve bidirectional closed-loop coordination between the virtual power grid and the cooling system.
[0004] Therefore, it is necessary to propose a new regulation method, with the peak-shaving task of the virtual power grid as the core upper-level objective, requiring the cooling system to adjust the power load according to the power grid instructions; at the same time, with the minimum energy consumption of the cooling system under the premise of meeting indoor comfort as the lower-level objective, a closed-loop coordination between the virtual power grid and the cooling system is formed. Summary of the Invention
[0005] This invention provides a two-way coordinated control method for virtual power grids and cooling systems. While meeting the peak-shaving needs of the power grid, it ensures indoor environmental comfort and reduces the energy consumption of the cooling system, effectively improving the stability of power grid operation and the energy utilization efficiency on the user side.
[0006] The technical solution adopted by this invention to solve its technical problem is: A bidirectional coordinated control method for virtual power grids and cooling systems includes the following steps: Step S1: Collect data, including parameters related to power generation, load, cooling system, and rules and prices; Step S2: Preprocess the collected data. The preprocessing methods include first removing noise from the data, then using interpolation or machine learning models to fill in missing data, and finally normalizing the data. Step S3: Based on the data preprocessed in step S2, the virtual grid side predicts the uncontrollable power generation and actual load in the future time domain to obtain the predicted power generation sequence and the predicted base load / actual load sequence; wherein, the uncontrollable power generation includes wind and solar power output and external power reception. Step S4: Determine the range of comfortable temperature inside the building, and establish a cooling load prediction model by taking into account external factors and thermal inertia to obtain the upper limit and lower limit of cooling load. Then, use the intelligent agent on the cooling system side to convert the upper limit and lower limit of cooling load into the upper limit and lower limit of cooling system power, respectively. Step S5: The intelligent agent on the cooling system side calculates the adjustable space under set constraints, using the baseline power under normal operating conditions as a reference, and reports the adjustable space to the virtual grid side; wherein, the set constraints include comfort constraints, feasible range constraints for cooling load, and equipment operation constraints. Step S6: Train the virtual grid-side agent under the virtual grid rules. Based on the predicted power generation sequence and predicted base load / actual load sequence obtained in step S3, the trained virtual grid-side agent is optimized based on the adjustable space reported in step S5 to generate the target adjustment amount or target power trajectory. Step S7: The intelligent agent on the refrigeration system side determines whether the generated target adjustment amount meets the adjustable space obtained in step S6. If it does, a control strategy is generated to adjust the equipment; if it does not, the maximum achievable adjustment amount is output to adjust the equipment. Furthermore, in step S1, the data includes parameters related to power generation such as historical power, real-time power, and weather information; parameters related to load such as historical actual load, real-time load, and energy consumption event information; parameters related to the cooling system such as indoor temperature, outdoor temperature and humidity, supply and return water temperature, pump / fan operating frequency, number of chillers in operation / load rate, and valve opening; and parameters related to rules and prices such as real-time and predicted electricity prices, demand response event windows, capacity application limits, baseline calculation rules, assessment standards, and penalty parameters. Furthermore, in step S2, data denoising refers to removing abnormally fluctuating data; Furthermore, in step S3, the future time domain... The power generation and actual load are used to predict the predicted power generation sequence. Predicted base load / actual load sequence ,in, ; Furthermore, in step S4, the specific steps for obtaining the upper limit and lower limit of the refrigeration system's electrical power are as follows: Step S41: Set the upper and lower limits of the comfortable body temperature. ; Step S42, construct a cooling load prediction model, including: ,in, This is the lower limit of the cooling load. This is the upper limit of the cooling load. External disturbances include external factors and thermal inertia; Step S43: Establish the mapping from cooling load to electrical power, and obtain... ,in As an intelligent agent on the refrigeration system side, This is the virtual environment state space for the refrigeration system. This is the lower limit of the electrical power of the refrigeration system. This is the upper limit of the electrical power of the refrigeration system; In step S5, the steps for obtaining the adjustable space are as follows: Step S51, at time Set the baseline power under normal operating conditions. ; Step S52, set the comfort constraints as follows The feasible range of cooling load is constrained as follows: Equipment operation constraints include the upper and lower limits of pump / fan operating frequency, minimum load rate of chiller operation, number of start-stop cycles, minimum start-stop time, and ramp rate; Step S53: Construct a virtual environment using a reinforcement learning algorithm, minimize power consumption within the feasible region, adjust the action space based on the given virtual environment state space, execute actions, and calculate and select actions that maximize the long-term cumulative reward of the cooling system by combining a preset reward function; wherein, the reward function used is a reward function that combines thermal balance and power minimization. Step S54, the adjustable capacity is obtained as follows: The capacity can be increased to ; Step S55: Attach virtual grid rules, including response latency, duration, ramp rate cap, assessment rules, and revenue model. The adjustable space reported to the virtual grid side is... ; Furthermore, in step S6, the predicted power generation sequence will be... Forecast base load / actual load sequence , In addition, the parameters related to rules and prices are input as inputs to the virtual environment state space of the virtual grid-side intelligent agent to obtain the target adjustment amount. or target power trajectory ; Furthermore, in step S7, the specific steps for adjusting the equipment are as follows: Step S71: Input the target adjustment amount of the virtual power grid side into the intelligent body on the cooling system side. ; Step S72, if According to the upper limit of cooling load Lower limit of cooling load Using a binary search method, an intermediate cooling load is obtained. The pump / chimney / fan is then adjusted by an intelligent agent on the refrigeration system side to approximate this intermediate value until the actual power change is achieved. satisfy ; Step S73, if Not in Within the specified range, the maximum adjustment range for the pump / chill / fan is: .
[0007] By employing the above technical solutions, the present invention has the following beneficial effects compared to the prior art: 1. The bidirectional coordinated control method for virtual power grid and cooling system provided by the present invention predicts the adjustable upper and lower limits of cooling load in the future time domain based on comfort constraints, external factors and thermal inertia, and can dynamically characterize the adjustable space. 2. The bidirectional collaborative control method for virtual power grid and cooling system provided by the present invention performs unified optimization and scheduling on the virtual power grid side, outputs equipment control quantities on the cooling system side, verifies the feasibility of adjustment and provides feedback on the execution status, and realizes a bidirectional closed loop between the virtual power grid and cooling system, which "reports adjustable space upwards and sends adjustment requirements downwards and verifies feasibility"; 3. The bidirectional coordinated control method for virtual power grid and cooling system provided by the present invention trains the agent under the constraints of virtual power grid rules and revenue model, which improves the performance rate and economic benefits, while reducing the risk of comfort breach. 4. The bidirectional collaborative control method for virtual power grid and refrigeration system provided by the present invention enables the refrigeration system to collaboratively output control quantities to multiple devices (pumps / chillers / fans), avoiding efficiency reduction or inoperability caused by single-point adjustment. Attached Figure Description
[0008] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0009] Figure 1 This is a schematic diagram illustrating the principle of the bidirectional coordinated control method for virtual power grids and refrigeration systems provided by the present invention. Figure 2 This is a schematic diagram of the operating principle of the refrigeration system side in the bidirectional coordinated control method for virtual power grids and refrigeration systems provided by the present invention; Figure 3 This is a schematic diagram of the operating principle of the virtual power grid side in the bidirectional coordinated control method for virtual power grid and refrigeration system provided by the present invention. Detailed Implementation
[0010] The present invention will now be described in further detail with reference to the accompanying drawings. In the description of this application, it should be understood that the terms "left side," "right side," "upper part," "lower part," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, 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. "First," "second," etc., do not indicate the importance of the components, and therefore should not be construed as a limitation of the present invention. The specific dimensions used in this embodiment are only for illustrating the technical solution and do not limit the scope of protection of the present invention.
[0011] As described in the background section, given the current situation where virtual power grids need to aggregate adjustable resources to participate in services such as peak shaving, demand management, and demand response, and considering that building cooling load is one of the main electricity loads of the urban power grid, existing technologies separate the regulation of the virtual power grid from the optimization of the building cooling system. This makes it difficult to respond efficiently to the power grid's peak shaving commands and to achieve optimal energy-saving operation of the cooling system while ensuring building comfort, thus failing to maximize the synergistic benefits between the power grid side and the user side.
[0012] To address the aforementioned challenges, this application provides a bidirectional collaborative control method for virtual power grids and cooling systems. Its innovations mainly include the following: the ability to predict power generation and actual load; the ability to predict the upper and lower limits of cooling load based on considerations of comfort, thermal inertia, and external factors; training and executing an agent under the rules of the virtual power grid; and achieving bidirectional collaborative control between the virtual power grid and the cooling system, which involves "positively informing the adjustable space and negatively issuing and verifying the adjustment requirements."
[0013] The design concept of the entire regulation method is as follows: Figure 1 As shown, the specific steps include: Step S1: Collect data, which includes parameters related to power generation, load, cooling system, and rules and prices. Among the data, the parameters related to power generation include historical power, real-time power, and weather information of distributed generation; the parameters related to load include historical actual load, real-time load, and energy consumption event information (weekdays / holidays / events); the parameters related to cooling system include indoor temperature, outdoor temperature and humidity, supply and return water temperature, pump / fan operating frequency, number of chillers in operation / load rate, and valve opening; the parameters related to rules and prices include real-time and forecasted electricity prices, demand response event windows, capacity application limits, baseline calculation rules, assessment standards, and penalty parameters.
[0014] Step S2 involves preprocessing the collected data. This preprocessing includes denoising the data, using interpolation or machine learning models to fill in missing data, and finally normalizing the data. Due to equipment malfunctions, personnel records, and other issues, some data anomalies may occur. Anomalies are initially screened based on the process engineers' understanding of the actual operating conditions. Then, box plots are used to filter outouts for individual features, and LOF (List of Elements) is used to filter out anomalies involving multiple feature combinations.
[0015] Among the several methods for standardizing the collected data in this step, noise reduction removes abnormal fluctuations, such as extreme values caused by equipment failure; interpolation or machine learning models complete missing data; and normalization maps data of different dimensions to the same interval, eliminating the influence of dimensions for subsequent prediction and model training.
[0016] Step S3: Based on the preprocessed data from step S2, the virtual power grid side targets the future time domain. The uncontrollable power generation and actual load are predicted to obtain the predicted power generation sequence. Predicted base load / actual load sequence ,in, Uncontrollable power generation refers to the portion that is significantly affected by the external environment and is relatively difficult to control, including solar power output and external power reception. Therefore, power generation prediction is divided into three parts: solar power output prediction, wind power output prediction, and external power reception prediction. The prediction model can be a time series model, a machine learning model, or a fusion model, and can output point predictions and confidence intervals to provide a basis for subsequent risk constraints. Depending on the data dimensions obtained, wind and solar power output prediction can obtain environmental data (wind force, wind speed, light intensity, cloud thickness) and use a machine learning model, while external power reception prediction can only obtain historical data and use a time series model.
[0017] Step S4: Determine the range of comfortable temperature inside the building, and establish a cooling load prediction model by taking into account external factors and thermal inertia to obtain the upper limit and lower limit of cooling load. Then, use the intelligent agent on the cooling system side to convert the upper limit and lower limit of cooling load into the upper limit and lower limit of cooling system power, respectively. Step S41: Set the upper and lower limits of the comfortable body temperature. , serving as the core benchmark for cooling load constraints; Step S42, taking into account external factors (outdoor temperature and humidity, solar radiation, heat dissipation from people or equipment, etc.) and thermal inertia (one or more of the following: building envelope heat capacity, indoor air heat capacity, and cold storage / water system heat capacity), construct a cooling load prediction model, including: ,in, This is the lower limit of the cooling load, meeting the minimum cooling demand under the constraint of the lower limit of comfort, and avoiding overcooling or ineffective energy consumption; The maximum cooling load limit must not exceed the maximum cooling demand constrained by the comfort limit, to prevent room temperature from exceeding the standard. External disturbances include external factors and thermal inertia; Step S43: Establish the mapping from cooling load to electrical power, and obtain... ,in As an intelligent agent on the refrigeration system side, This is the virtual environment state space for the refrigeration system. This is the lower limit of the electrical power of the refrigeration system. This represents the upper limit of the electrical power of the refrigeration system.
[0018] Step S5 is another function of the cooling system. The intelligent agent on the cooling system side uses the baseline power under normal operating conditions as a reference to calculate the adjustable space under set constraints and reports the adjustable space to the virtual grid side. The set constraints include comfort constraints, feasible range constraints for cooling load, and equipment operation constraints.
[0019] The specific steps are as follows: Step S51, at time Set the baseline power under normal operating conditions. ; Step S52, set the comfort constraints as follows The feasible range of cooling load is constrained as follows: Equipment operation constraints include the upper and lower limits of pump / fan operating frequency, minimum load rate of chiller operation, number of start-stop cycles, minimum start-stop time, and ramp rate; that is, the intelligent agent on the refrigeration system side needs to perform optimization and boundary calculations under the above constraints. The design of comfort constraints avoids room temperature exceeding the standard, and the equipment operation constraints strictly follow the threshold values of the operating parameters of each piece of equipment in the refrigeration system to prevent equipment damage. Step S53: Perform forward derivation of the objective, construct a virtual environment using reinforcement learning algorithms, minimize power consumption (or energy consumption) within the feasible region, adjust the action space based on the given virtual environment state space conditions, execute actions, and calculate and select actions that maximize the long-term cumulative reward of the cooling system by combining a preset reward function; the reward function used is a reward function that combines thermal balance and power minimization; it should be noted that the minimum power consumption or minimum energy consumption is obtained through the cooling system-side agent, which will be explained in detail later.
[0020] Step S54: Based on the above optimization results and power boundary, the time is calculated. At that time, the capacity (peak shaving capability) can be reduced to [a lower value]. The capacity (Pinggu capacity) can be increased to ; Step S55: Add virtual grid rules, including response latency, duration, ramp rate cap, assessment rules, and revenue model. Therefore, the adjustable space reported to the virtual grid side is... .
[0021] In the above steps, regarding the intelligent agent on the cooling system side, this application uses the reinforcement learning SAC algorithm to build the intelligent agent and simulate the real cooling system. The system is divided into a cooling side and a freezing side. An energy-saving adaptive model of the cooling system is constructed, which includes a virtual environment on the cooling side and a virtual environment on the freezing side. Based on the equipment parameters and physical rules of the real computer room, a "virtual environment on the cooling side + virtual environment on the freezing side" is constructed to form a computer room energy-saving adaptive model that is mapped 1:1 to the real computer room. At the same time, a reward function (minimum total power consumption, meeting thermal balance conditions, etc.) and adjustable actions (cooling water pump frequency, chilled water pump frequency, chiller power, cooling tower fan frequency) are established.
[0022] Specifically, regarding the cooling side, the state space settings include: cooling tower fan operating frequency, inlet water temperature, number of pumps, pump operating frequency, number of chiller units in operation, outlet water temperature, cooling load, and outdoor wet-bulb temperature. The action space settings include: cooling tower fan operating frequency range ±5Hz, and pump operating frequency range ±5Hz. The reward function settings include: setting the value of cooling tower heat dissipation / (heat generated by cooling-side equipment + cooling load on the chiller side) within the range of 0.9~1.1; if the condition is met... If the conditions are not met, then , The reward function for the normalized total cooling power is as follows: To minimize the total power on the cooling side while satisfying the constraints.
[0023] Regarding the chilled water side, the state space settings include: chiller outlet water temperature, number of pumps, pump operating frequency, and chilled water return temperature (which can be replaced with environmental comfort indicators in other project scenarios). The action space settings include: pump operating frequency range ±5Hz, and chiller outlet water temperature range ±1℃. The reward function settings include: if the chilled water return temperature is less than a threshold, then... If the conditions are not met, then , The reward function for the normalized total power on the cryogenic side is: Minimize the total power on the freezing side while satisfying the constraints.
[0024] Step S6: Train the virtual grid-side agent under the virtual grid rules, using the predicted power generation sequence obtained in step S3. and the predicted base load / actual load sequence Based on this, with the goal of maximizing the economic benefits of the power grid while minimizing losses, In addition, parameters related to rules and prices are used as inputs to the virtual environment state space of the virtual grid-side intelligent agent to generate the target adjustment quantity. or target power trajectory .
[0025] This step focuses on virtual grid-side decision-making and training to generate adjustment requirements. Regarding the subject to be trained: the virtual grid-side intelligent agent, its state space settings include: (1) grid load, including current load and predicted load (short-term prediction), used to reflect the current load situation of the grid. (2) grid power supply status, including available generator capacity, energy storage system status (SOC, battery energy storage capacity, etc.), and grid health status (e.g., fault conditions). (3) current electricity price, including real-time price fluctuations based on electricity market dynamics. (4) grid loss, the energy lost during power transmission, calculated by methods including line resistance, transformer efficiency, etc. (5) energy storage system status, if an energy storage system exists, the battery charging and discharging status, battery life, energy storage capacity, etc. should be considered.
[0026] The action space settings include: (1) Generator output regulation, adjusting the output power of different generator sets according to the current load demand and grid conditions. It can be further refined into adjustable generator sets (e.g., conventional power generation, renewable energy generation, distributed energy, etc.), such as adjusting the output of conventional power plants, adjusting the power generation of renewable energy (e.g., wind power, solar power), and adjusting the charging and discharging power of energy storage devices. (2) Grid load regulation, balancing the grid load by adjusting the consumption of controllable loads (e.g., industrial load, commercial load), and sending it to the cooling system. (3) Energy storage system regulation, adjusting the charging and discharging strategy of the energy storage system to ensure that the energy storage system can provide sufficient power during peak grid load and charge the energy storage during off-peak hours.
[0027] The optimized reward function settings include: (1) Economic benefits, electricity price and electricity consumption: calculate the economic cost of electricity consumption based on grid load and real-time electricity price. (2) Minimize grid losses. Grid losses are directly related to grid load fluctuations. Losses can be reduced by reducing load fluctuations and improving the stability of grid operation. If the steady-state and dynamic losses of the power grid need to be considered, the smooth operation of the power grid should be rewarded and large fluctuations should be punished. (3) Battery / energy storage system efficiency: If the energy storage system can charge and discharge reasonably, ensure the stable operation of the power grid during peak hours and reduce power grid losses, a reward will be given. (4) Comprehensive reward: Combining all rewards, a final total reward function is formed: This ensures that the reward function guides the agent to balance economic benefits, stability, grid losses, and the efficiency of energy storage systems.
[0028] Step S7: The intelligent agent on the refrigeration system side determines whether the generated target adjustment amount satisfies the adjustable space obtained in step S6. If it does, a control strategy is generated to adjust the equipment; if not, the maximum achievable adjustment amount is output to adjust the equipment. The specific steps are as follows: Step S71: Input the target adjustment amount of the virtual power grid side into the intelligent body on the cooling system side. ; Step S72, if According to the upper limit of cooling load Lower limit of cooling load Using a binary search method, an intermediate cooling load is obtained. The pump / chimney / fan is then adjusted by an intelligent agent on the refrigeration system side to approximate this intermediate value until the actual power change is achieved. satisfy This scenario represents a scenario where the intelligent agent on the refrigeration system side aims to meet the target adjustment amount, minimize power consumption, and ensure comfort and safety. It adjusts the operating frequency of pumps / fans and the number / load rate of chillers to change the actual power output. fit as closely as possible .
[0029] Step S73, if Not in Within this range, representing an infeasible scenario, the maximum adjustment range for the pump / chimney / fan is calculated as follows: Identify the reasons for infeasibility, such as comfort temperature boundary limitations, equipment operating upper limit constraints, or insufficient ramp rate, and feed back the maximum achievable adjustment amount, reasons for infeasibility, and feasible control strategies / adjustable editing information to the virtual grid side to support its modification of the scheduling scheme or coordinated response with other adjustable resources.
[0030] In summary, the bidirectional coordinated control method for virtual power grids and cooling systems provided in this application, such as... Figure 2 and Figure 3As shown, the virtual power grid side knows the "adjustable space" for unified clearing and strategy formulation, while the cooling system side knows the "target adjustment amount" to generate executable control actions and verify whether the adjustment requirements can be met. Through system load forecasting and peak-shaving task generation on the virtual power grid side, combined with cooling load forecasting and equipment optimization scheduling on the cooling system side, peak shaving is achieved and economic benefits are obtained. Simultaneously, the feasibility of adjustment and feedback on execution status are verified, realizing the synergistic optimization of power grid peak-shaving requirements and building cooling energy conservation. While meeting power grid peak-shaving requirements, indoor environmental comfort is ensured, and cooling system energy consumption is reduced, effectively improving power grid operation stability and user-side energy utilization efficiency.
[0031] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.
[0032] The meaning of "and / or" as used in this application includes both situations where each exists alone or both exist simultaneously.
[0033] The term "connection" as used in this application can mean a direct connection between components or an indirect connection between components through other components.
[0034] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.
Claims
1. A bidirectional coordinated control method for virtual power grids and refrigeration systems, characterized in that: Includes the following steps: Step S1: Collect data, including parameters related to power generation, load, cooling system, and rules and prices; Step S2: Preprocess the collected data. The preprocessing methods include first removing noise from the data, then using interpolation or machine learning models to fill in missing data, and finally normalizing the data. Step S3: Based on the data preprocessed in step S2, the virtual grid side predicts the uncontrollable power generation and actual load in the future time domain to obtain the predicted power generation sequence and the predicted base load / actual load sequence; wherein, the uncontrollable power generation includes wind and solar power output and external power reception. Step S4: Determine the range of comfortable temperature inside the building, and establish a cooling load prediction model by taking into account external factors and thermal inertia to obtain the upper limit and lower limit of cooling load. Then, use the intelligent agent on the cooling system side to convert the upper limit and lower limit of cooling load into the upper limit and lower limit of cooling system power, respectively. Step S5: The intelligent agent on the cooling system side calculates the adjustable space under set constraints, using the baseline power under normal operating conditions as a reference, and reports the adjustable space to the virtual grid side; wherein, the set constraints include comfort constraints, feasible range constraints for cooling load, and equipment operation constraints. Step S6: Train the virtual grid-side agent under the virtual grid rules. Based on the predicted power generation sequence and predicted base load / actual load sequence obtained in step S3, the trained virtual grid-side agent is optimized based on the adjustable space reported in step S5 to generate the target adjustment amount or target power trajectory. In step S7, the intelligent agent on the refrigeration system side determines whether the generated target adjustment amount meets the adjustable space obtained in step S6. If it does, a control strategy is generated to adjust the equipment; if it does not, the maximum achievable adjustment amount is output to adjust the equipment.
2. The bidirectional coordinated control method for virtual power grids and cooling systems according to claim 1, characterized in that: In step S1, the data includes parameters related to power generation such as historical power, real-time power, and weather information; parameters related to load such as historical actual load, real-time load, and energy consumption event information; parameters related to the cooling system such as indoor temperature, outdoor temperature and humidity, supply and return water temperature, pump / fan operating frequency, number of chillers in operation / load rate, and valve opening; and parameters related to rules and prices such as real-time and predicted electricity prices, demand response event windows, capacity application limits, baseline calculation rules, assessment standards, and penalty parameters.
3. The bidirectional coordinated control method for virtual power grids and cooling systems according to claim 1, characterized in that: In step S2, data denoising refers to removing abnormally fluctuating data.
4. The bidirectional coordinated control method for virtual power grids and cooling systems according to claim 1, characterized in that: In step S3, the future time domain The uncontrollable power generation and actual load are predicted to obtain the predicted power generation sequence. Predicted base load / actual load sequence ,in, .
5. The bidirectional coordinated control method for virtual power grids and cooling systems according to claim 1, characterized in that: In step S4, the specific steps for obtaining the upper limit and lower limit of the refrigeration system's electrical power are as follows: Step S41: Set the upper and lower limits of the comfortable body temperature. ; Step S42, construct a cooling load prediction model, including: ,in, This is the lower limit of the cooling load. This is the upper limit of the cooling load. External disturbances include external factors and thermal inertia; Step S43: Establish the mapping from cooling load to electrical power, and obtain... ,in As an intelligent agent on the refrigeration system side, This is the virtual environment state space for the refrigeration system. This is the lower limit of the electrical power of the refrigeration system. This represents the upper limit of the electrical power of the refrigeration system.
6. The bidirectional coordinated control method for virtual power grids and cooling systems according to claim 1, characterized in that: In step S5, the steps for obtaining the adjustable space are as follows: Step S51, at time Set the baseline power under normal operating conditions. ; Step S52, set the comfort constraints as follows The feasible range of cooling load is constrained as follows: Equipment operation constraints include the upper and lower limits of pump / fan operating frequency, minimum load rate of chiller operation, number of start-stop cycles, minimum start-stop time, and ramp rate; Step S53: Construct a virtual environment using a reinforcement learning algorithm, minimize power consumption within the feasible region, adjust the action space based on the given virtual environment state space, execute actions, and calculate and select actions that maximize the long-term cumulative reward of the cooling system by combining a preset reward function; wherein, the reward function used is a reward function that combines thermal balance and power minimization. Step S54, the adjustable capacity is obtained as follows: The capacity can be increased to ; Step S55: Attach virtual grid rules, including response latency, duration, ramp rate cap, assessment rules, and revenue model. The adjustable space reported to the virtual grid side is... .
7. The bidirectional coordinated control method for virtual power grids and cooling systems according to claim 1, characterized in that: In step S6, the predicted power generation sequence is... Forecast base load / actual load sequence , In addition, parameters related to rules and prices are used as inputs to the virtual environment state space of the virtual grid-side intelligent agent to obtain the target adjustment amount. or target power trajectory .
8. The bidirectional coordinated control method for virtual power grids and cooling systems according to claim 1, characterized in that: In step S7, the specific steps for adjusting the equipment are as follows: Step S71: Input the target adjustment amount of the virtual power grid side into the intelligent body on the cooling system side. ; Step S72, if According to the upper limit of cooling load Lower limit of cooling load Using a binary search method, an intermediate cooling load is obtained. The pump / chimney / fan is then adjusted by an intelligent agent on the refrigeration system side to approximate this intermediate value until the actual power change is achieved. satisfy ; Step S73, if Not in Within the specified range, the maximum adjustment range for the pump / chill / fan is: .