An air-conditioning energy-saving control method and system based on a virtual power plant

Abstract

The present invention provides an air-conditioning energy-saving control method and system based on a virtual power plant. By obtaining the operating parameter sets of multiple air-conditioning devices and real-time power grid load data in a target area, an air-conditioning energy-saving control model corresponding to the multiple air-conditioning devices is determined. According to the real-time power grid load data and load prediction data, the power grid load fluctuation threshold of the target area is determined, and the operating parameter sets of the multiple air-conditioning devices are dynamically optimized through a parameter optimization model to generate an optimized subset of device operating parameters. The temperature set values and operating modes of the multiple air-conditioning devices in a future period are adjusted to generate an air-conditioning group control strategy corresponding to the future period. Operating feedback data is obtained, and according to the deviation value between the actual power grid load data and the load prediction data, the weight parameters of the load prediction model are updated, and the optimization rules of the parameter optimization model are calibrated to generate a calibrated model. The present invention can improve the refined regulation ability of the air-conditioning group and the energy coordination efficiency.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular, to an air-conditioning energy-saving control method and system based on a virtual power plant. Background Art

[0002] The air-conditioning energy-saving control of a virtual power plant aims to achieve dynamic balance of the power grid load by coordinating the operating parameters of air-conditioning equipment in the region. In the prior art, fixed-threshold control or a load response model based on a unified adjustment strategy is usually adopted to achieve peak shaving and valley filling of the power grid load by presetting a fixed power threshold or uniformly adjusting the air-conditioning temperature set value. However, the fixed-threshold control cannot adapt to the dynamic changes of the power grid fluctuations and user behaviors, resulting in insufficient adjustment accuracy and frequent triggering of equipment start and stop. Moreover, the uniform adjustment strategy ignores the differences in equipment priorities and user comfort requirements, easily causing abnormal air-conditioning functions or a decline in user experience in critical scenarios. At the same time, the load prediction model and the equipment control model in the existing methods are isolated from each other, and the prediction results and execution strategies lack real-time coordination, resulting in response lag and long-term operation energy efficiency decay, and it is difficult to meet the refined control requirements of the virtual power plant for the air-conditioning group. Summary of the Invention

[0003] The present invention provides an air-conditioning energy-saving control method and system based on a virtual power plant.

[0004] In a first aspect, an embodiment of the present invention provides an air-conditioning energy-saving control method based on a virtual power plant. The method includes: obtaining an operating parameter set of a plurality of air-conditioning devices in a target area and real-time power grid load data, and determining an air-conditioning energy-saving control model corresponding to the plurality of air-conditioning devices; the air-conditioning energy-saving control model includes a load prediction model and a parameter optimization model, wherein the load prediction model is used to generate load prediction data in a future period according to historical load data and real-time environmental parameters, and the parameter optimization model is used to generate a parameter adjustment strategy for each corresponding air-conditioning device according to the load prediction data; determining a power grid load fluctuation threshold of the target area according to the real-time power grid load data and the load prediction data, and dynamically optimizing the operating parameter set of the plurality of air-conditioning devices through the parameter optimization model to generate an optimized subset of device operating parameters; adjusting the temperature set value and operating mode of the plurality of air-conditioning devices in the future period according to the optimized subset of device operating parameters to generate an air-conditioning group control strategy corresponding to the future period; obtaining the actual power grid load data and air-conditioning group operation feedback data in the future period, and updating the weight parameters of the load prediction model according to the deviation value between the actual power grid load data and the load prediction data; calibrating the optimization rule of the parameter optimization model according to the updated weight parameters and the air-conditioning group operation feedback data to generate a calibrated air-conditioning energy-saving control model.

[0005] In a second aspect, an air conditioner energy-saving control system provided by an embodiment of the present invention includes: a memory in which a computer program is stored; a processor configured to load the computer program to implement the air conditioner energy-saving control method based on a virtual power plant as described above.

[0006] The air conditioner energy-saving control method based on a virtual power plant provided by the present invention forms a closed-loop control architecture by deeply coupling a load prediction model and a parameter optimization model, generates a dynamically optimized subset of device operation parameters in real time by combining grid load data and environmental parameters, and hierarchically adjusts the temperature set value and power distribution strategy based on device priority identification and user behavior feedback, which can significantly improve the response efficiency of the air conditioner group and the grid load regulation accuracy; by introducing an adaptive weight update mechanism for the load prediction model and a rule dynamic calibration mechanism for the parameter optimization model, the control strategy continuously adapts to the grid fluctuation trend and the actual needs of users, effectively solving the problems of cumulative prediction deviation and insufficient long-term operation stability in traditional methods; at the same time, based on the collaborative optimization design of device priority division and grid load fluctuation threshold, the power of low-priority devices is accurately adjusted on the premise of ensuring the comfort of users of key devices, achieving a balance between energy-saving goals and user experience; further, through end-to-end model training and application data consistency design, scene adaptation deviation is eliminated, ensuring the reliable execution of control instructions in a complex grid environment, thereby comprehensively improving the refined regulation ability of the virtual power plant for the air conditioner group and the energy coordination efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a flowchart of an air conditioner energy-saving control method based on a virtual power plant provided by an embodiment of the present invention.

[0008] Figure 2 is a schematic diagram of the composition of an air conditioner energy-saving control system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] Next, the technical solutions in the embodiments of the present invention will be described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0010] Please refer to Figure 1 , Figure 1 is a flowchart of an air conditioner energy-saving control method based on a virtual power plant provided by an embodiment of the present invention. The air conditioner energy-saving control method based on a virtual power plant can be executed by an air conditioner energy-saving control system. The air conditioner energy-saving control method based on a virtual power plant may include the following steps:

[0011] Step S100: Obtain the operating parameter sets and real-time grid load data of multiple air-conditioning devices in the target area, and determine the air-conditioning energy-saving control models corresponding to the multiple air-conditioning devices; the air-conditioning energy-saving control models include a load prediction model and a parameter optimization model. The load prediction model is used to generate load prediction data for a future period based on historical load data and real-time environmental parameters, and the parameter optimization model is used to generate parameter adjustment strategies for each corresponding air-conditioning device according to the load prediction data.

[0012] In the embodiments of the present invention, the target area refers to the geographical scope where air-conditioning energy-saving control needs to be carried out, such as a commercial park, an office building or a residential building, etc. The operating parameter sets include various parameters of multiple air-conditioning devices during operation, such as temperature set values, operating powers, wind speed gears, operating modes (cooling, heating, dehumidification, etc.). The real-time grid load data refers to the magnitude of the electric power load borne by the grid in the target area at the current moment, which reflects the degree of power consumption tension of the grid at this moment.

[0013] The load prediction model is a model for load prediction based on historical data and real-time environmental parameters. The historical load data is a record of the grid load situation in the target area over a past period of time, and these data contain the load change rules under different time periods, different seasons, and different weather conditions. The real-time environmental parameters include real-time temperature, humidity, light intensity, etc. These factors will all affect the operation of air-conditioning devices and the grid load. The load prediction model can adopt a neural network model, such as a long short-term memory network (LSTM), which can handle time series data well and capture the long-term dependence relationship of load data. The LSTM model includes an input layer, a hidden layer, and an output layer. The input layer receives historical load data and real-time environmental parameters, the LSTM units in the hidden layer process the data through memory and forgetting mechanisms, and the output layer generates load prediction data for a future period.

[0014] The parameter optimization model then formulates appropriate parameter adjustment strategies for each air-conditioning device according to the load prediction data generated by the load prediction model. For example, when the load prediction data shows that the grid load will reach a peak in a future period, the parameter optimization model can formulate strategies to reduce the operating power of some air-conditioning devices or adjust their operating modes to relieve the grid pressure. The parameter optimization model can be optimized using a genetic algorithm. The genetic algorithm simulates the biological evolution process and searches for the optimal parameter adjustment strategy in the solution space through operations such as selection, crossover, and mutation. In this process, each possible parameter adjustment strategy is regarded as an individual, and multiple individuals form a population. Through continuous iterative evolution, the optimal parameter adjustment strategy is finally found.

[0015] Step S200: Determine the grid load fluctuation threshold of the target area based on the real-time grid load data and the load prediction data, and dynamically optimize the operating parameter sets of multiple air conditioning devices through a parameter optimization model to generate an optimized subset of device operating parameters.

[0016] The real-time grid load data reflects the actual load situation of the grid at the current moment, while the load prediction data is an estimate of the grid load in future periods. The grid load fluctuation threshold refers to the fluctuation limit of the grid load within a preset range (the specific range is not limited). Determining it is crucial for reasonably adjusting the operating parameters of air conditioning devices. For example, when the grid load fluctuation exceeds the threshold, more aggressive energy-saving measures may be required; when the load fluctuation is within the threshold range, relatively mild parameter adjustments can be made.

[0017] Dynamically optimizing the operating parameter sets of multiple air conditioning devices through a parameter optimization model means adjusting the operating parameters of air conditioning devices in real time according to different load situations. For example, when the grid load is high, lower the temperature set values of air conditioning devices in some non-critical areas, or switch some air conditioning devices to the energy-saving mode. Through this dynamic optimization, the operation of air conditioning devices can better meet the load demand of the grid and achieve the purpose of energy conservation.

[0018] As an implementation method, in step S200, determine the grid load fluctuation threshold of the target area based on the real-time grid load data and the load prediction data, and dynamically optimize the operating parameter sets of multiple air conditioning devices through a parameter optimization model to generate an optimized subset of device operating parameters, which can specifically include the following steps:

[0019] Step S210: Obtain the current temperature set values, operating power thresholds, and device priority identifiers of each air conditioning device in the target area.

[0020] The current temperature set value refers to the temperature set for each air conditioning device currently, which directly affects the cooling or heating effect and energy consumption of the air conditioner. The operating power threshold refers to the maximum power allowed for an air conditioning device during normal operation. Exceeding this threshold may cause damage to the device or affect the stable operation of the grid. The device priority identifier is used to distinguish the importance of different air conditioning devices. The specific configuration method is not limited. For example, in a commercial office building, the air conditioning devices in areas such as meeting rooms and server rooms may have a higher priority, while the air conditioning devices in public corridors and stairwells may have a relatively lower priority.

[0021] Step S220: Divide multiple load adjustment intervals within the future period according to the peak load period and the valley load period in the load prediction data.

[0022] The peak load period in the load forecasting data refers to the time period when the grid load reaches the maximum value, and the valley load period is the time period when the grid load reaches the minimum value. Based on these two periods, the future time periods can be divided into multiple load adjustment intervals. For example, in a day, the morning rush hour and the evening rush hour may be the peak load periods of the grid, while the early morning period is the valley load period. A day can be divided into multiple time periods, such as the morning peak period, the morning off-peak period, the noon period, the afternoon off-peak period, the evening peak period, and the night valley period, etc., and each time period corresponds to a load adjustment interval.

[0023] Step S230: Set the maximum load value and the minimum load value corresponding to the grid load fluctuation threshold in the parameter optimization model, and generate the opening and closing strategies of the air-conditioning equipment corresponding to each load adjustment interval according to the maximum load value and the minimum load value.

[0024] In practical applications, the parameter optimization model can adopt a linear programming model. The linear programming model finds the optimal solution by defining the objective function and the constraint conditions and satisfying the set conditions. In this scenario, the objective function can be to minimize the grid load fluctuation or to maximize the energy-saving effect of the air-conditioning equipment, and the constraint conditions include the maximum and minimum values of the grid load, the operating power threshold of the air-conditioning equipment, etc. For example, assume that the objective function is to minimize the grid load fluctuation, the constraint conditions are that the grid load is between the maximum load value and the minimum load value, and the operating power of the air-conditioning equipment does not exceed its operating power threshold. By solving this model with the linear programming algorithm, the opening and closing strategies of the air-conditioning equipment corresponding to each load adjustment interval can be obtained, that is, which equipment should be turned on and which equipment should be turned off to achieve the optimal energy-saving and load adjustment effects.

[0025] Step S240: Based on the device priority identifier, group multiple air-conditioning devices, determine the high-priority device group and the low-priority device group, and according to the opening and closing strategies of the air-conditioning equipment, reduce the operating power threshold of the low-priority device group during the peak load period and increase the temperature setting value of the high-priority device group during the valley load period.

[0026] Grouping air-conditioning equipment based on device priority identification allows for more targeted parameter adjustment. High-priority device groups are, for example, equipment with high requirements for the indoor environment or crucial for business operations, such as air-conditioning equipment in data centers; low-priority device groups are equipment with relatively lower environmental requirements, such as air-conditioning equipment in public areas. During peak load periods, reducing the operating power threshold of low-priority device groups can effectively reduce the grid load; while during valley load periods, increasing the temperature setpoint of high-priority device groups can reduce the energy consumption of air-conditioning equipment while ensuring comfort. After determining high-priority and low-priority device groups, a distributed control system can be used to achieve parameter adjustment for different device groups. The distributed control system disperses control functions to each device or device group and realizes information interaction and coordination through a communication network. For example, during peak load periods, the distributed control system can send instructions to reduce the operating power threshold to each device in the low-priority device group according to the air-conditioning equipment on / off strategy. After receiving the instructions, the device automatically adjusts its operating power. During valley load periods, similarly, instructions to increase the temperature setpoint are sent to the equipment in the high-priority device group through the distributed control system to achieve precise control of different device groups.

[0027] Step S241: Obtain the historical usage frequency data, user-preset preferred temperature range, and equipment energy consumption level of each air-conditioning equipment.

[0028] The historical usage frequency data reflects the usage situation of each air-conditioning equipment over a past period. Equipment with a high usage frequency may have a greater impact on user comfort. The user-preset preferred temperature range refers to the temperature range set by the user according to their own needs and habits. Different users may have different feelings and requirements for temperature. The equipment energy consumption level indicates the energy consumption level of the air-conditioning equipment. The higher the energy consumption level, the more electrical energy the equipment consumes during operation. Obtaining this information helps to more accurately evaluate the importance and energy consumption situation of each air-conditioning equipment. To obtain this data, a data acquisition system can be established. The data acquisition system can collect the operating data of air-conditioning equipment in real time through sensors and communication modules. Specifically, it can include the following steps of historical usage frequency data. For the user-preset preferred temperature range, relevant information can be input by the user through the user interface or intelligent terminal device and stored in the database. The equipment energy consumption level can be obtained by querying the product manual or relevant certification documents of the equipment and entered into the data acquisition system. The data acquisition system can use cloud computing to upload the collected data to the cloud for storage and analysis for subsequent comprehensive priority score calculation.

[0029] Step S242: Determine the usage activity score of each air-conditioning equipment based on the historical usage frequency data, and generate the comprehensive priority score of each equipment by combining the user-preset preferred temperature range and the equipment energy consumption level.

[0030] The usage activity score can be calculated based on historical usage frequency data. For example, devices with a high usage frequency can be given a higher usage activity score. The comprehensive priority score is obtained by comprehensively considering factors such as the usage activity score, the user's preset preferred temperature range, and the device energy consumption level. For example, corresponding weights can be assigned to each factor, and then the comprehensive priority score can be calculated by weighted summation. Assume that the weight of the usage activity score is 0.4, the weight of the user's preset preferred temperature range is 0.3, and the weight of the device energy consumption level is 0.3. Then the comprehensive priority score = usage activity score × 0.4 + user's preset preferred temperature range score × 0.3 + device energy consumption level score × 0.3. When calculating the usage activity score, a time series analysis method can be used. The time series analysis method can model and predict historical usage frequency data, and calculate the usage activity score based on the data's change trend and periodic characteristics. For example, the moving average method can be used to smooth the historical usage frequency data, and then calculate the usage activity score based on the processed data. Or, other methods can also be used to calculate the usage activity score. For example, assume that the statistical period for calculating the usage activity score of air conditioning equipment is the past month. The daily startup duration of each air conditioner is counted. If an air conditioner has a total startup duration of 200 hours in 30 days of this month, with an average daily startup of about 6.7 hours. Set the basic score value to 50 points, and add 2 points for each additional hour of daily startup duration. This air conditioner exceeds the basic daily startup duration (assuming the basic is 3 hours) by 3.7 hours, and the additional score is 3.7 × 2 = 7.4 points. Then the usage activity score of this air conditioner is 50 + 7.4 = 57.4 points. When generating the comprehensive priority score, the fuzzy comprehensive evaluation method can also be used. The fuzzy comprehensive evaluation method can comprehensively consider multiple factors (such as the usage activity score, the user's preset preferred temperature range, and the device energy consumption level), and calculate the comprehensive priority score of each device by determining the weight and membership function of each factor.

[0031] Step S243: Classify the first device set with a comprehensive priority score greater than the preset threshold as the high-priority device group, and classify the remaining second device set as the low-priority device group.

[0032] The preset threshold is a scoring boundary set according to the actual situation, which is used to distinguish high-priority devices and low-priority devices. Devices with a comprehensive priority score greater than the preset threshold are classified into the high-priority device group, and these devices are usually more important for the indoor environment and business operations; devices with a comprehensive priority score less than or equal to the preset threshold are classified into the low-priority device group, and these devices can be adjusted more flexibly when the power grid load is tight. The determination of the preset threshold can adopt the clustering analysis method. The clustering analysis method can cluster the comprehensive priority scores of all air-conditioning devices, and determine a suitable threshold according to the clustering results to divide the devices into high-priority device groups and low-priority device groups. For example, the K-Means clustering algorithm can be used to divide the comprehensive priority score data into two categories, and then determine the preset threshold according to the clustering center and data distribution.

[0033] Step S244: Assign a first optimization weight to the high-priority device group and a second optimization weight to the low-priority device group in the parameter optimization model, where the first optimization weight is greater than the second optimization weight.

[0034] When assigning the optimization weights, the analytic hierarchy process can be adopted. The analytic hierarchy process decomposes the problem into multiple levels and factors by constructing a hierarchical structure model, and then determines the relative importance of each factor through pairwise comparison, so as to obtain the optimization weights of each device group. For example, in this scenario, the high-priority device group and the low-priority device group can be used as the target layer, and the factors affecting the device priority (such as usage activity, user needs, energy consumption, etc.) can be used as the criterion layer. By comparing the importance of each factor, the first optimization weight and the second optimization weight are determined.

[0035] Step S245: According to the first optimization weight and the second optimization weight, perform parameter adjustment operations on the high-priority device group and the low-priority device group within the load adjustment range respectively.

[0036] When performing the parameter adjustment operation, the model predictive control method can be adopted. The model predictive control method establishes a dynamic model of the device, predicts the operating state of the device in the future period of time, and adjusts the control parameters in real time according to the optimization goal and constraint conditions. For example, during the peak load period, for the low-priority device group, the model predictive control method can predict the operating state of the device after reducing the power according to the second optimization weight and the power grid load situation, and then adjust the power adjustment instruction in real time to ensure that the device minimizes the impact on the indoor environment while meeting the power grid load requirements. During the valley load period, for the high-priority device group, the model predictive control method can also adjust the temperature setting value according to the first optimization weight and the user's preset preferred temperature range to achieve a balance between energy saving and comfort.

[0037] As an implementation manner, in step S245, according to the first optimization weight and the second optimization weight, parameter adjustment operations are respectively performed on the high-priority device group and the low-priority device group within the load adjustment range, which may specifically include:

[0038] Step S2451: During the peak load period, according to the second optimization weight of the low-priority device group, determine the allowable maximum power reduction ratio thereof, and generate a corresponding power adjustment instruction set based on the maximum power reduction ratio.

[0039] During the peak load period, in order to relieve the power grid pressure, it is necessary to adjust the power of the low-priority device group. According to the second optimization weight, the allowable maximum power reduction ratio of the low-priority device group can be determined. For example, if the second optimization weight is small, it means that a relatively large power reduction operation can be performed on the low-priority device group. At this time, a relatively high maximum power reduction ratio can be set. Based on this ratio, a corresponding power adjustment instruction set is generated. Each instruction corresponds to a low-priority device, and the instruction content includes information such as the power value that needs to be reduced. When determining the maximum power reduction ratio, an adaptive control algorithm can be used. The adaptive control algorithm can automatically adjust the maximum power reduction ratio according to the real-time change of the power grid load and the operating state of the low-priority device group. For example, when the power grid load increases sharply, the adaptive control algorithm can increase the maximum power reduction ratio to quickly relieve the power grid pressure; when the power grid load tends to be stable, the maximum power reduction ratio can be appropriately reduced to ensure the basic comfort of the indoor environment. When generating the power adjustment instruction set based on the determined maximum power reduction ratio, a rule engine can be used to automatically generate power adjustment instructions according to the preset rules and conditions. For example, the rule can stipulate that when the operating power of a certain low-priority device exceeds the set threshold, its power is reduced according to the maximum power reduction ratio.

[0040] Step S2452: During the valley load period, according to the first optimization weight of the high-priority device group, determine the allowable temperature set value increase range thereof, and generate a corresponding temperature adjustment instruction set based on the user preset preferred temperature range.

[0041] When determining the temperature set value increase range, user feedback and environmental monitoring data can be combined. For example, by installing temperature sensors and humidity sensors indoors to monitor the indoor environmental parameters in real time, and at the same time collecting user feedback information on temperature changes. According to the first optimization weight, the user preset preferred temperature range and the environmental monitoring data, a reasonable temperature set value increase range is determined. When generating the temperature adjustment instruction set, intelligent algorithms such as genetic algorithms or particle swarm algorithms can be used. These algorithms can search for the optimal temperature adjustment scheme and generate the corresponding temperature adjustment instruction set on the premise of meeting user needs and environmental conditions.

[0042] Step S2453: Input the power adjustment instruction set and the temperature adjustment instruction set into the parameter optimization model for conflict detection. If a resource conflict is detected between the temperature adjustment instruction of the high-priority device group and the power adjustment instruction of the low-priority device group, the temperature adjustment instruction of the high-priority device group shall be preferentially executed.

[0043] In actual operation, there may be resource conflicts between the power adjustment instruction set and the temperature adjustment instruction set. For example, the power adjustment of a certain device may affect the temperature adjustment of another device. Input these two instruction sets into the parameter optimization model for conflict detection. When a conflict is detected, since the high-priority device group is more important for the indoor environment and business operation, the temperature adjustment instruction of the high-priority device group is preferentially executed to ensure the normal operation of high-priority devices and user comfort. When performing conflict detection, a constraint satisfaction problem-solving algorithm can be used. The constraint satisfaction problem-solving algorithm can regard the power adjustment instruction set and the temperature adjustment instruction set as a set of constraint conditions, and determine whether there is a conflict through search and reasoning. For example, if the power adjustment of a low-priority device affects the temperature adjustment of a high-priority device, it will be determined that there is a conflict. When a conflict is detected, the temperature adjustment instruction of the high-priority device group is preferentially executed, which is based on the principle of ensuring the normal operation of high-priority devices and user comfort. To implement this priority strategy, a priority queue can be set in the parameter optimization model, and the temperature adjustment instructions of the high-priority device group are arranged in the front to ensure that they are preferentially executed when a conflict occurs.

[0044] Step S2454: Generate an optimized subset of device operation parameters according to the adjustment instructions after conflict detection.

[0045] The subset of device operation parameters includes optimized temperature set values, operating powers, etc. of each air-conditioning device, and these parameters will be used for subsequent control of air-conditioning devices to achieve the goal of air-conditioning energy conservation. When generating the optimized subset of device operation parameters, data fusion technology can be used. Data fusion technology can fuse the adjustment instructions after conflict detection and the original device operation parameters to generate a new and optimized subset of device operation parameters. For example, the weighted average method can be used to fuse the parameters in the adjustment instructions and the original parameters, and different weights are assigned according to the importance and credibility of the instructions. The generated optimized subset of device operation parameters will be used as the basis for subsequent air-conditioning device control to ensure the best balance between energy conservation and comfort while meeting the grid load requirements.

[0046] Step S300: Adjust the temperature set values and operating modes of multiple air-conditioning devices in the future time period according to the optimized subset of device operation parameters, and generate an air-conditioning group control strategy corresponding to the future time period.

[0047] The subset of optimized device operation parameters includes parameters such as the temperature set values and operating powers of each air conditioner device after optimization. Based on these parameters, adjusting the temperature set values and operating modes of multiple air conditioner devices in the future time period is to make the operation of the air conditioner devices more in line with the power grid load demand and user comfort requirements. For example, if the optimized parameters show that a certain air conditioner device needs to increase the temperature set value and switch to the energy-saving mode in the future time period, then control the device according to this adjustment plan.

[0048] Generating an air conditioner group control strategy corresponding to the future time period is to integrate the adjustment plans for multiple air conditioner devices to form a unified control strategy. This strategy needs to consider the mutual influence between each air conditioner device and the overall situation of the power grid load. For example, within a certain load adjustment range, it may be necessary to simultaneously adjust the temperature set values and operating modes of multiple air conditioner devices to achieve the best energy-saving and load adjustment effects.

[0049] As an implementation manner, in step S300, according to the subset of optimized device operation parameters, adjusting the temperature set values and operating modes of multiple air conditioner devices in the future time period and generating an air conditioner group control strategy corresponding to the future time period may specifically include:

[0050] Step S310: Extract the temperature set value adjustment amount and operating mode adjustment amount corresponding to each of the multiple air conditioner devices from the subset of optimized device operation parameters, where the temperature set value adjustment amount includes the temperature difference for upward or downward adjustment, and the operating mode adjustment amount includes the switching identifier for the cooling mode, energy-saving mode, or standby mode.

[0051] The temperature set value adjustment amount refers to the temperature difference that needs to be increased or decreased relative to the current temperature set value, which reflects the change in the temperature of the air conditioner device required to achieve the energy-saving and load adjustment goals. The operating mode adjustment amount is the operating mode identifier that the air conditioner device needs to switch to, such as switching from the cooling mode to the energy-saving mode, or from the energy-saving mode to the standby mode, etc. Extracting this information from the subset of optimized device operation parameters provides a specific adjustment basis for subsequent control operations. In actual operation, data parsing algorithms can be used to extract the required information from the subset of optimized device operation parameters. For example, regular expression matching technology can be used to extract the temperature set value adjustment amount and operating mode adjustment amount from the parameter subset according to the preset format rules. For the temperature set value adjustment amount, it can be represented as a numerical value, with a positive number indicating an increase in temperature and a negative number indicating a decrease in temperature. For the operating mode adjustment amount, preset codes can be used to represent different operating mode switches, such as "1" indicating switching from the cooling mode to the energy-saving mode, "2" indicating switching from the energy-saving mode to the standby mode, etc.

[0052] Step S320: According to the temperature setpoint adjustment amount and the operating mode adjustment amount, allocate time window adjustment tasks for each air-conditioning device within the future time period, where the time window adjustment task includes an operation instruction to adjust the temperature setpoint or switch the operating mode at the start, middle, or end of the future time period.

[0053] Based on the extracted temperature setpoint adjustment amount and operating mode adjustment amount, allocate time window adjustment tasks for each air-conditioning device within the future time period. The time window adjustment task specifies at which moment within the future time period the temperature setpoint of the air-conditioning device needs to be adjusted or the operating mode needs to be switched. For example, for a certain air-conditioning device, the allocated time window adjustment task may be to increase the temperature setpoint by 2°C at the start of the future time period and switch the operating mode from the cooling mode to the energy-saving mode. To implement this allocation process, a task scheduling algorithm can be used. The task scheduling algorithm can generate specific time window adjustment tasks for each air-conditioning device based on the temperature setpoint adjustment amount and operating mode adjustment amount, combined with the time division of the future time period. For example, the earliest deadline first (EDF) scheduling algorithm can be used to sort according to the deadline of the adjustment task (i.e., the start, middle, or end of the future time period), and give priority to arranging tasks with earlier deadlines. This can ensure that the adjustment tasks are completed within the specified time.

[0054] Step S330: Obtain the device response delay parameters of multiple air-conditioning devices and the grid load fluctuation trend data of the target area. Determine the minimum interval duration for each air-conditioning device to execute the time window adjustment task according to the device response delay parameters, and dynamically sort the execution order of the time window adjustment tasks in combination with the grid load fluctuation trend data.

[0055] The device response delay parameter refers to the time required for an air-conditioning device to actually execute a control instruction after receiving it. Different air-conditioning devices may have different response delay parameters, which will affect the execution of the time window adjustment task. Determining the minimum interval duration for each air-conditioning device to execute the time window adjustment task according to the device response delay parameter can avoid sudden changes in the grid load caused by multiple devices executing adjustment tasks simultaneously.

[0056] The grid load fluctuation trend data of the target area reflects the change trend of the grid load in the future time period. Combining the grid load fluctuation trend data and dynamically sorting the execution order of the time window adjustment tasks can make the adjustment operations of air conditioning equipment more in line with the change of the grid load. For example, when the grid load is about to reach the peak, some adjustment tasks to reduce power can be preferentially executed. The device response delay parameter can be obtained by actually testing the air conditioning equipment or referring to the technical specification of the equipment. When determining the minimum interval duration, a safe interval time can be set according to the maximum value of the device response delay parameter to avoid sudden changes in the grid load caused by multiple devices executing adjustment tasks simultaneously. For the grid load fluctuation trend data of the target area, it can be obtained in real time through the grid monitoring system. When dynamically sorting the execution order of the time window adjustment tasks in combination with these data, heuristic algorithms such as genetic algorithms or simulated annealing algorithms can be used. These algorithms can search for the optimal task execution order considering the grid load fluctuation trend, making the adjustment operations of air conditioning equipment more in line with the change of the grid load.

[0057] Step S340: Based on the dynamically sorted execution order, bind the temperature setpoint adjustment amount, the operating mode adjustment amount to the corresponding time window adjustment tasks, and generate an individual control instruction set for each air conditioning equipment.

[0058] According to the dynamically sorted execution order, bind the temperature setpoint adjustment amount, the operating mode adjustment amount to the corresponding time window adjustment tasks to form an individual control instruction set for each air conditioning equipment. The individual control instruction set contains the specific adjustment tasks and execution times that each air conditioning equipment needs to execute in the future time period, providing detailed instructions for the precise control of the air conditioning equipment. During the binding process, a data structure can be used to store this information. For example, a dictionary or an object can be used to represent the individual control instruction set of each air conditioning equipment, where the keys can be the relevant information of the temperature setpoint adjustment amount, the operating mode adjustment amount and the time window adjustment tasks, and the values are the specific parameters or operation instructions. In this way, the temperature setpoint adjustment amount, the operating mode adjustment amount can be closely associated with the corresponding time window adjustment tasks, facilitating subsequent control operations.

[0059] Step S350: Detect conflicts among the instructions in the individual control instruction set. If it is detected that there are operation mode switching operations of multiple air conditioning equipments in the same time window, which may lead to the risk of sudden change in the grid load, re-prioritize the conflicting instructions according to the device priority identifier, and verify the matching degree between the re-sorted instruction set and the load prediction data in the future time period.

[0060] Conflict detection is performed on each instruction in the individual control instruction set to avoid sudden changes in the power grid load caused by the adjustment operations of multiple air conditioning devices within the same time window. For example, if multiple high-power air conditioning devices switch from the standby mode to the cooling mode at the same time, it may cause a significant instantaneous increase in the power grid load, posing a safety risk. When a conflict is detected, the conflicting instructions are re-prioritized according to the device priority identifier to ensure the smooth execution of the adjustment tasks of high-priority devices first.

[0061] The matching degree verification of the re-prioritized instruction set with the load prediction data for the future time period is to ensure that the adjusted tasks after re-prioritization will not deviate significantly from the predicted situation of the power grid load. If the matching degree verification fails, it may be necessary to further adjust the instruction set to ensure that the adjustment operations of the air conditioning devices meet the requirements of the power grid load. When performing conflict detection, graph theory algorithms can be used. The adjustment task of each air conditioning device is regarded as a node in the graph, and the edges between the nodes represent the time relationship and possible conflicts between the tasks. By traversing the nodes and edges in the graph, it can be detected whether there is a situation where the adjustment tasks of multiple devices within the same time window cause a sudden change in the power grid load. When a conflict is detected, the conflicting instructions are re-prioritized according to the device priority identifier, and a priority queue data structure can be used. The adjustment tasks of high-priority devices are arranged at the front of the queue and executed first. When performing the matching degree verification of the re-prioritized instruction set with the load prediction data for the future time period, statistical analysis methods can be used to calculate the similarity between the load changes in the instruction set and the load prediction data. If the similarity reaches the preset threshold, it is considered that the matching degree verification passes.

[0062] Step S360: According to the matching degree verification result, the individual control instruction sets that have no conflicts and conform to the load prediction data are merged into an air conditioning group control strategy.

[0063] According to the matching degree verification result, the individual control instruction sets that have no conflicts and conform to the load prediction data are merged into an air conditioning group control strategy. This strategy combines the adjustment tasks of all air conditioning devices and can achieve a balance between air conditioning energy conservation and indoor environmental comfort on the premise of ensuring the safe and stable operation of the power grid. During the merging process, data fusion technology can be used. The individual control instruction set of each air conditioning device is regarded as a data source, and these data sources are integrated into a unified air conditioning group control strategy through a data fusion algorithm. For example, the weighted average method or the voting method can be used to assign different weights according to the reliability and importance of the individual control instruction sets, and then they are merged into a final control strategy. The air conditioning group control strategy generated in this way can comprehensively consider the adjustment tasks of all air conditioning devices and achieve the goals of safe and stable operation of the power grid and air conditioning energy conservation.

[0064] Step S400: Obtain the actual power grid load data and the operation feedback data of the air conditioner group within the future time period, and update the weight parameters of the load forecasting model according to the deviation value between the actual power grid load data and the load forecasting data.

[0065] The actual power grid load data within the future time period refers to the magnitude of the power load actually borne by the power grid within the future time period, which reflects the real operation situation of the power grid. The operation feedback data of the air conditioner group is the actual operation status information of multiple air conditioner devices during the implementation of the control strategy, such as the actual temperature, actual power, etc.

[0066] Updating the weight parameters of the load forecasting model according to the deviation value between the actual power grid load data and the load forecasting data is to improve the accuracy of the load forecasting model. If the deviation value is large, it indicates that there is a large difference between the prediction result of the load forecasting model and the actual situation, and the weight parameters of the model need to be adjusted so that the model can better adapt to the actual situation.

[0067] As an implementation method, in step S400, updating the weight parameters of the load forecasting model according to the deviation value between the actual power grid load data and the load forecasting data may specifically include:

[0068] Step S410: Extract the predicted load change sequence corresponding to the future time period from the load forecasting data, and obtain the actual load change sequence of the same time period in the actual power grid load data.

[0069] The predicted load change sequence is the prediction result of the load forecasting model for the power grid load change within the future time period. It is a time series data that contains the predicted load values at different time points. The actual load change sequence is the real change situation of the actual power grid load within the same time period. Extracting these two sequences is for subsequent deviation calculation. When extracting the predicted load change sequence, the part corresponding to the future time period can be screened from the load forecasting data by timestamp matching. For example, assuming the future time period is from 9:00 am to 5:00 pm on January 1, 2024, then according to the timestamp information, the predicted load values within this time period in the load forecasting data can be extracted to form the predicted load change sequence. For the actual power grid load data, it can be collected in real time through the power grid monitoring system, and the actual load change sequence can be extracted according to the same time range. This can ensure the consistency of the two sequences in time and facilitate subsequent deviation calculation.

[0070] Step S420: Calculate the load difference between the predicted load change sequence and the actual load change sequence at each time point to generate a deviation value sequence.

[0071] By calculating the load difference between the predicted load change sequence and the actual load change sequence at each time point, a deviation value sequence can be obtained. The deviation value sequence reflects the magnitude of the prediction error of the load prediction model at each time point. For example, if at a certain time point, the predicted load value is 1000 kW and the actual load value is 1100 kW, then the load difference at this time point is 100 kW, and this difference will be used as an element in the deviation value sequence. When calculating the load difference, the method of point-by-point subtraction can be used. For example, for the i-th time point in the predicted load change sequence and the actual load change sequence, they are respectively denoted as the predicted load value Pi and the actual load value Ai, then the load difference Di at this time point is Di = Ai - Pi. Arranging the load differences at all time points in sequence forms the deviation value sequence. This sequence intuitively reflects the prediction error situation of the load prediction model at each time point.

[0072] Step S430: Determine the weight error amounts of the load prediction model for the temperature correlation parameter and the time correlation parameter according to the fluctuation amplitude and duration of the deviation value sequence.

[0073] The fluctuation amplitude and duration of the deviation value sequence reflect the severity and duration of the prediction error of the load prediction model. Based on this information, the weight error amounts of the load prediction model for the temperature correlation parameter and the time correlation parameter can be determined. The temperature correlation parameter refers to the model parameter related to temperature, because temperature will affect the operation of air conditioning equipment and the grid load; the time correlation parameter is the model parameter related to time, such as the load change law in different time periods, etc. The weight error amount indicates the magnitude of the adjustment required for the weights of the model for these parameters. The fluctuation amplitude of the deviation value sequence can be measured by calculating the difference between the maximum value and the minimum value in the sequence, and the duration can be determined by counting the number of consecutive time points when the deviation value is greater than the preset threshold. For the weight error amounts of the temperature correlation parameter and the time correlation parameter, the method of regression analysis can be used to determine. For example, establish a regression model with the fluctuation amplitude and duration of the deviation value sequence as independent variables and the weight error amounts of the temperature correlation parameter and the time correlation parameter as dependent variables, and calculate the weight error amounts by fitting the regression equation. In this way, according to the characteristics of the deviation value sequence, the adjustment direction and magnitude of the weights of the model for different parameters can be accurately determined.

[0074] Step S440: Use the gradient descent algorithm to perform backpropagation on the weight error amounts and adjust the connection weights of the neural network layer in the load prediction model.

[0075] The gradient descent algorithm is a general optimization algorithm used to find the minimum value of a function. In the load forecasting model, by backpropagating the weight error amount, the gradient descent algorithm is used to adjust the connection weights of the neural network layer. Backpropagation means transmitting the error from the output layer to the input layer, and adjusting the connection weights according to the magnitude and direction of the error to make the prediction result of the model more accurate. The core idea of the gradient descent algorithm is to update the connection weights along the negative gradient direction of the error function to gradually reduce the prediction error. During the backpropagation process, first calculate the partial derivative of the error function with respect to each connection weight to obtain the gradient of the weight. Then, according to the formula of the gradient descent algorithm, update the connection weights. For example, for the connection weight W, its update formula is W = W - η * ∇E, where η is the learning rate and ∇E is the gradient of the error function with respect to W. Through multiple iterative updates, the connection weights are continuously adjusted to make the prediction result of the model more accurate.

[0076] Step S450: Smooth the adjusted connection weights and the historical weight data to generate updated weight parameters.

[0077] Smoothing is to avoid drastic changes in the connection weights and make the adjustment of the model more stable. To smooth the adjusted connection weights and the historical weight data, methods such as the moving average method can be used. For example, calculate the weighted average of the adjusted connection weights and the historical weight data as the updated weight parameters. This can make the model maintain stability while adapting to the new actual situation.

[0078] As an implementation, in step S440, the gradient descent algorithm is used to backpropagate the weight error amount and adjust the connection weights of the neural network layer in the load forecasting model, which may specifically include:

[0079] Step S441: According to the distribution characteristics of historical load data, real-time environmental temperature, humidity, and user behavior characteristics in the input layer nodes, determine the initial weight allocation ratio corresponding to each neuron in the hidden layer nodes.

[0080] The input layer nodes contain information such as historical load data, real-time environmental temperature, humidity, and user behavior characteristics. The distribution characteristics of this information reflect the degree of their influence on load forecasting. Based on these distribution characteristics, determining the initial weight allocation ratios corresponding to each neuron in the hidden layer nodes is to enable the neural network to better learn the relationship between this information and load forecasting. For example, if the historical load data has a large fluctuation range, a larger initial weight can be assigned to the hidden layer neurons related to the historical load data. When determining the initial weight allocation ratios, the principal component analysis (PCA) method can be used. The PCA method can perform dimensionality reduction on multiple input variables (historical load data, real-time environmental temperature, humidity, and user behavior characteristics) and extract the main components. Then, based on the contribution degree of each component to load forecasting, the initial weight allocation ratios corresponding to each neuron in the hidden layer nodes are determined. For example, if a certain component has a greater contribution to load forecasting, a larger initial weight can be assigned to the hidden layer neurons related to that component.

[0081] As an implementation manner, in step S441, according to the distribution characteristics of the historical load data, real-time environmental temperature, humidity, and user behavior characteristics in the input layer nodes, determining the initial weight allocation ratios corresponding to each neuron in the hidden layer nodes may specifically include the following steps:

[0082] Step S4411: Obtain the fluctuation range and period characteristics of the historical load data, the change trend characteristics of the real-time environmental temperature and humidity, and the time distribution density characteristics of the user behavior characteristics, and generate a feature correlation degree set corresponding to each input layer node.

[0083] The fluctuation range and period characteristics of the historical load data reflect the change law of the load. The change trend characteristics of the real-time environmental temperature and humidity reflect the influence of environmental factors on the load. The time distribution density characteristics of the user behavior characteristics represent the time law of the user using the air conditioning equipment. By analyzing these characteristics, a feature correlation degree set corresponding to each input layer node is generated, and this set contains the information on the correlation degree between each input layer node and load forecasting. When obtaining the fluctuation range and period characteristics of the historical load data, the Fourier transform method can be used. The Fourier transform can convert time series data into the frequency domain to analyze the periodic components and fluctuation range of the data. For the change trend characteristics of the real-time environmental temperature and humidity, the moving average method or the linear regression method can be used to fit the change trend of the data. For the time distribution density characteristics of the user behavior characteristics, by counting the number of user behaviors in different time periods and drawing a time distribution histogram, the distribution density characteristics can be analyzed. Based on these characteristics, the correlation degree between each input layer node and load forecasting is calculated to generate the feature correlation degree set.

[0084] Step S4412: Calculate the first weight influence factor of the historical load data node and the real-time ambient temperature node, the second weight influence factor of the humidity node, and the third weight influence factor of the user behavior feature node according to the morphological correlation between the fluctuation amplitudes of the features in the feature correlation degree set and the predicted load curve.

[0085] Calculate the weight influence factors of different input layer nodes according to the morphological correlation between the fluctuation amplitudes of the features in the feature correlation degree set and the predicted load curve. The first weight influence factor reflects the influence degree of historical load data and real-time ambient temperature on load prediction, the second weight influence factor represents the influence degree of humidity on load prediction, and the third weight influence factor reflects the influence degree of user behavior features on load prediction. These weight influence factors are used to determine the initial weight allocation ratio. When calculating the weight influence factors, a correlation analysis method can be adopted. Correlation analysis can calculate the correlation coefficient between two variables, reflecting the degree of association between them. For the historical load data node and the real-time ambient temperature node, calculate their correlation coefficient with the predicted load curve as the first weight influence factor. Similarly, calculate the correlation coefficient between the humidity node and the predicted load curve as the second weight influence factor, and calculate the correlation coefficient between the user behavior feature node and the predicted load curve as the third weight influence factor. These weight influence factors reflect the importance of different input layer nodes to load prediction.

[0086] Step S4413: Determine the initial weight transfer ratio from the input layer nodes to the hidden layer nodes based on the first weight influence factor, the second weight influence factor, and the third weight influence factor, where the transfer ratio is the normalized allocation result of each weight influence factor on the hidden layer neurons.

[0087] Determine the initial weight transfer ratio from the input layer nodes to the hidden layer nodes based on the calculated first weight influence factor, the second weight influence factor, and the third weight influence factor. Normalized allocation means processing each weight influence factor so that their sum is 1, and then allocating the weights to the hidden layer neurons according to the normalized weight influence factors. This can ensure the rationality and balance of the initial weight allocation. Normalized allocation can be achieved by dividing each weight influence factor by their sum. For example, assume the first weight influence factor is W1, the second weight influence factor is W2, and the third weight influence factor is W3, then the normalized transfer ratios are P1 = W1 / (W1 + W2 + W3), P2 = W2 / (W1 + W2 + W3), P3 = W3 / (W1 + W2 + W3). According to these transfer ratios, allocate the weights of the input layer nodes to the hidden layer neurons to ensure the rationality and balance of the initial weight allocation.

[0088] Step S4414: According to the number of hidden layer nodes and the transfer ratio, allocate the historical load data nodes corresponding to the first weight influence factor to the pre-order neurons of the hidden layer nodes, allocate the humidity nodes corresponding to the second weight influence factor to the intermediate neurons, and allocate the user behavior feature nodes corresponding to the third weight influence factor to the post-order neurons.

[0089] According to the number of hidden layer nodes and the transfer ratio, allocate the weights corresponding to different input layer nodes to different positions of the hidden layer. Allocate the historical load data nodes to the pre-order neurons of the hidden layer nodes because historical load data usually has an important early influence on load prediction; allocate the humidity nodes to the intermediate neurons considering that the influence of humidity on load may be more obvious in the intermediate stage; allocate the user behavior feature nodes to the post-order neurons because user behavior features may have a greater influence in the later stage of load prediction. During the allocation process, the weights that each neuron should be allocated can be calculated according to the number of hidden layer nodes and the transfer ratio. For example, assume there are n neurons in the hidden layer and the transfer ratio corresponding to the first weight influence factor is P1, then P1 can be evenly distributed to the first k pre-order neurons (k is determined according to the actual situation). Similarly, distribute the transfer ratio corresponding to the second weight influence factor to the middle m neurons, and distribute the transfer ratio corresponding to the third weight influence factor to the post-order l neurons (m + l + k = n). In this way, the weights can be reasonably allocated to different positions of the hidden layer according to the importance of different input layer nodes.

[0090] Step S4415: Based on the allocation results of the pre-order neurons, intermediate neurons, and post-order neurons, generate the initial weight allocation ratios of the neurons in the hidden layer nodes, and verify the fitting degree between the initial weight allocation ratios and the load fluctuation pattern in the predicted load curve. If the fitting degree does not reach the preset threshold, re-adjust the calculation method of the feature correlation degree set.

[0091] Based on the allocation results of the preceding neurons, intermediate neurons and subsequent neurons, the initial weight allocation ratio of each neuron in the hidden layer node is generated. Then, the degree of fit between this initial weight allocation ratio and the load fluctuation pattern in the predicted load curve is verified. If the degree of fit does not reach the preset threshold, it means that the initial weight allocation may be unreasonable, and it is necessary to readjust the calculation method of the feature association set to obtain a more appropriate initial weight allocation ratio. After the initial weight allocation ratio is generated, the generated initial weight is used for load forecasting to obtain a predicted load curve. Then, the degree of fit between the predicted load curve and the load fluctuation pattern in the actual load curve is calculated. The degree of fit can be measured by calculating the correlation coefficient or mean square error between the two. If the degree of fit does not reach the preset threshold, it means that the initial weight allocation may be unreasonable, and it is necessary to readjust the calculation method of the feature association set. For example, the parameters of the Fourier transform, the window size of the moving average method or the correlation analysis method can be adjusted to obtain a more appropriate feature association set, thereby generating a more reasonable initial weight allocation ratio.

[0092] Step S442: Based on the fluctuation amplitude of the deviation value sequence and the predicted load curve shape of the future period in the load forecast data, the error propagation path between the input layer nodes and the hidden layer nodes is calculated to generate the weight gradient correction direction of the hidden layer nodes.

[0093] The fluctuation amplitude of the deviation value sequence and the shape of the predicted load curve reflect the prediction error of the load forecasting model. Based on this information, the error propagation path between the input layer nodes and the hidden layer nodes is calculated to determine the weight gradient correction direction of the hidden layer nodes. The weight gradient correction direction indicates the direction in which the weights of the hidden layer nodes need to be adjusted in the gradient descent algorithm to reduce the prediction error of the model. When calculating the error propagation path, the chain rule can be used. The chain rule is a derivation method in calculus that is used to calculate the derivative of a composite function. In a neural network, the process of error propagation from the output layer to the input layer can be regarded as a derivation process of a composite function. Through the chain rule, the partial derivative of the error with respect to each connection weight can be calculated to determine the error propagation path. According to the fluctuation amplitude of the error propagation path and the deviation value sequence, and the shape of the predicted load curve, the weight gradient correction direction of the hidden layer nodes is generated to guide the adjustment of the connection weights.

[0094] Step S443: the initial weight distribution ratio is iteratively adjusted layer by layer according to the weight gradient correction direction, and the deviation value change rate between the updated predicted load curve and the actual load curve is obtained after each iteration.

[0095] Layer-by-layer iterative adjustment of the initial weight allocation ratio according to the weight gradient correction direction is the core step of the gradient descent algorithm. After each iteration, calculate the change rate of the deviation value between the updated predicted load curve and the actual load curve to evaluate the adjustment effect of the model. If the change rate of the deviation value shows a downward trend, it indicates that the adjustment of the model is effective; if the change rate of the deviation value does not decrease or even increases, it may be necessary to adjust the parameters of the gradient descent algorithm. During the iterative adjustment process, after updating the connection weights according to the weight gradient correction direction each time, recalculate the output of the load prediction model to obtain the updated predicted load curve. Then calculate the change rate of the deviation value between the updated predicted load curve and the actual load curve. The change rate of the deviation value can be obtained by calculating the difference between the average values of the deviation value sequences in two adjacent iterations. For example, if the average value of the deviation value sequence in the k-th iteration is Ek and the average value of the deviation value sequence in the k + 1-th iteration is Ek+1, then the change rate of the deviation value is (Ek - Ek+1) / Ek. By monitoring the change rate of the deviation value, the adjustment effect of the model can be evaluated to determine whether it is necessary to continue adjusting the connection weights.

[0096] Step S444: If the change rate of the deviation value does not show a downward trend in consecutive N iterations, dynamically switch the step size parameter of the gradient descent algorithm according to the trend change direction of the actual load curve, and recalculate the weight gradient correction direction, where N > 2.

[0097] If the change rate of the deviation value does not show a downward trend in consecutive N iterations, it indicates that the gradient descent algorithm may be trapped in a local optimum or the step size parameter is inappropriate. At this time, dynamically switch the step size parameter of the gradient descent algorithm according to the trend change direction of the actual load curve, such as increasing or decreasing the step size, and then recalculate the weight gradient correction direction to jump out of the local optimum and continue to search for better weight parameters. When the change rate of the deviation value does not decrease in consecutive N iterations, it indicates that the gradient descent algorithm may be trapped in a local optimum or the step size parameter is inappropriate. At this time, dynamically switch the step size parameter according to the trend change direction of the actual load curve. If the actual load curve shows an upward trend, the step size can be appropriately increased to speed up the adjustment speed of the model; if the actual load curve shows a downward trend, the step size can be appropriately decreased to avoid over-adjustment of the model. Then recalculate the weight gradient correction direction and continue the iterative adjustment to jump out of the local optimum and search for better connection weights.

[0098] Step S445: When the average value of the deviation value sequence reaches the preset error threshold, lock the connection weights after the current iteration, and verify the fitting degree between the connection weights and the predicted load data output by the activation function of the load prediction model, and terminate the adjustment process after the fitting degree meets the preset stability condition.

[0099] When the average value of the deviation value sequence reaches the preset error threshold, it indicates that the prediction error of the model has been reduced to an acceptable range. At this time, lock the connection weights after the current iteration and verify the fitting degree between the connection weights and the predicted load data output by the activation function of the load prediction model. The fitting degree represents the matching degree between the connection weights and the predicted load data. If the fitting degree meets the preset stability condition, it is considered that the adjustment of the model has been completed and the adjustment process is terminated. The preset error threshold is an error limit set according to actual requirements and the accuracy requirements of the model. When the average value of the deviation value sequence reaches this threshold, it indicates that the prediction error of the model has been reduced to an acceptable range. At this time, lock the connection weights after the current iteration, and then verify the fitting degree between the connection weights and the predicted load data output by the activation function. The fitting degree can be measured by calculating indicators such as the correlation coefficient or the mean square error. If the fitting degree meets the preset stability condition, it means that the matching degree between the connection weights and the predicted load data is good, the adjustment of the model has been completed, and the adjustment process is terminated.

[0100] Step S500: Calibrate the optimization rules of the parameter optimization model according to the updated weight parameters and the operation feedback data of the air conditioner group to generate a calibrated air conditioner energy-saving control model.

[0101] The updated weight parameters are obtained after the adjustment of the load prediction model, which reflects the better adaptation of the model to the actual situation. The operation feedback data of the air conditioner group contains various information of multiple air conditioner devices during actual operation, such as actual energy consumption, temperature change, etc. Calibrating the optimization rules of the parameter optimization model according to these data is to enable the parameter optimization model to generate more accurate parameter adjustment strategies for air conditioner devices and further improve the air conditioner energy-saving effect.

[0102] As an implementation method, in step S500, calibrate the optimization rules of the parameter optimization model according to the updated weight parameters and the operation feedback data of the air conditioner group to generate a calibrated air conditioner energy-saving control model, which may specifically include:

[0103] Step S510: Obtain the actual energy consumption value of the device, the number of temperature deviations, and the user manual intervention frequency in the operation feedback data of the air conditioner group.

[0104] The actual energy consumption value of the device refers to the electric energy consumed by the air conditioner device during actual operation, which reflects the actual energy consumption situation of the air conditioner device. The number of temperature deviations refers to the number of deviations between the actual temperature and the set temperature of the air conditioner device, which may affect the user's comfort. The user manual intervention frequency refers to the frequency of the user's manual operation on the air conditioner device, such as manually adjusting the temperature setting value, operation mode, etc., which reflects the user's satisfaction with the current control strategy.

[0105] Step S520: Calculate the energy consumption deviation ratio based on the actual energy consumption value of the device and the predicted energy consumption value in the optimized subset of device operating parameters.

[0106] The energy consumption deviation ratio refers to the degree of deviation between the actual energy consumption value of the device and the predicted energy consumption value. By calculating the energy consumption deviation ratio, the accuracy of the predicted energy consumption value generated by the parameter optimization model can be evaluated. For example, if the energy consumption deviation ratio is large, it indicates that there is a significant difference between the predicted energy consumption value and the actual energy consumption value, and the optimization rules of the parameter optimization model need to be adjusted. Suppose in the optimized subset of device operating parameters, the predicted energy consumption value of an air conditioner device for the next day is 20 kWh. After actual operation for one day, the actual energy consumption value of the device is obtained as 22 kWh. According to the formula: Energy consumption deviation ratio = (Actual energy consumption value - Predicted energy consumption value) ÷ Predicted energy consumption value × 100%, the energy consumption deviation ratio of this air conditioner device can be calculated as (22 - 20) ÷ 20 × 100% = 10%. If the actual energy consumption value is lower than the predicted energy consumption value, such as 18 kWh, the energy consumption deviation ratio is (18 - 20) ÷ 20 × 100% = -10%.

[0107] Step S530: If the energy consumption deviation ratio exceeds the preset calibration threshold, determine the calibration amount of the parameter optimization model in the temperature setpoint adjustment strategy according to the number of temperature deviations and the user's manual intervention frequency.

[0108] The preset calibration threshold is an energy consumption deviation limit set according to the actual situation. When the energy consumption deviation ratio exceeds this threshold, it indicates that there may be problems with the optimization rules of the parameter optimization model and calibration is required. Determining the calibration amount of the parameter optimization model in the temperature setpoint adjustment strategy according to the number of temperature deviations and the user's manual intervention frequency is to make the temperature setpoint adjustment strategy more in line with the actual situation and improve the user's comfort and energy-saving effect.

[0109] Step S540: Add a user behavior correction factor to the parameter optimization model and dynamically adjust the influence coefficient of the correction factor according to the user's manual intervention frequency.

[0110] The user behavior correction factor is introduced to consider the impact of the user's manual intervention behavior on the operation of the air conditioner device. Dynamically adjusting the influence coefficient of the correction factor according to the user's manual intervention frequency can make the parameter optimization model more flexible to adapt to the user's needs. For example, if the user's manual intervention frequency is high, it indicates that the user is not satisfied with the current control strategy, and the influence coefficient of the correction factor needs to be increased to consider the user's manual operation more.

[0111] Step S550: Integrate the calibration amount and the correction factor into the optimization rules to generate a calibrated air conditioner energy-saving control model.

[0112] Integrate the determined calibration quantity and the adjusted correction factor into the optimization rules of the parameter optimization model to form a calibrated air-conditioning energy-saving control model. This model can generate more accurate parameter adjustment strategies for air-conditioning equipment according to the actual situation, achieving better energy-saving effects and user comfort.

[0113] As an implementation, in step S540, add a user behavior correction factor to the parameter optimization model and dynamically adjust the influence coefficient of the correction factor according to the user's manual intervention frequency. Specifically, it can include the following steps:

[0114] Step S541: Count the number of times the temperature set value changes due to the user's manual operation for each air-conditioning equipment in the future period to generate a user intervention record set.

[0115] When counting the number of times the temperature set value changes due to the user's manual operation for each air-conditioning equipment in the future period, data collection can be carried out with the help of the intelligent control system built into the air-conditioning equipment. This intelligent control system can accurately record the time of each user's manual operation, the operation type (such as temperature increase or decrease), and the temperature set values before and after the operation. Through the collation and analysis of these data, the number of times the temperature set value changes for each air-conditioning equipment is summarized, and then a user intervention record set is generated. For example, in a commercial area with 100 air-conditioning equipment, the user's manual operation data of each air-conditioning equipment within the next week is collected through the intelligent control system, and then the number of times the temperature set value changes for each equipment is counted to form a user intervention record set containing 100 records.

[0116] Step S542: Determine the type of user behavior pattern according to the distribution density and time distribution characteristics of the user intervention record set.

[0117] The distribution density of the user intervention record set reflects the difference in the frequency of user manual intervention among different air-conditioning equipment, and the time distribution characteristics reflect the law of the user's manual operation in different time periods. The clustering analysis method can be used to process the user intervention record set, and the user behaviors with similar distribution density and time distribution characteristics are classified into the same type. For example, if some air-conditioning equipment has frequent user manual interventions from 9 am to 11 am and from 2 pm to 4 pm on weekdays, and fewer interventions in other time periods, and the distribution of the intervention times of these equipment is relatively concentrated, then this type of user behavior can be classified as "high-frequency intervention type in specific weekdays"; if some air-conditioning equipment has relatively uniform and small interventions in all time periods of the day, it can be classified as "low-frequency and uniform intervention type throughout the day".

[0118] Step S543: If the type of user behavior pattern is the frequent intervention type, increase the weight ratio of the correction factor in the temperature set value adjustment strategy.

[0119] When it is determined that the user behavior pattern is of the frequent intervention type, it indicates that the user is not very satisfied with the current temperature set value adjustment strategy and hopes to achieve a more comfortable environment through manual operations. To better meet the user's needs, the weight ratio of the correction factor in the temperature set value adjustment strategy is increased. For example, originally the weight ratio of the correction factor in the temperature set value adjustment strategy was 20%, and after it is determined to be of the frequent intervention type, its weight ratio is increased to 40%. In this way, when the parameter optimization model generates the temperature set value adjustment strategy, it will consider the user's manual operation history more, making the adjustment strategy more in line with the user's actual needs.

[0120] Step S544: If the user behavior pattern type is of the low-frequency intervention type, then reduce the weight ratio of the correction factor and preferentially adopt the parameter adjustment strategy generated from the load prediction data.

[0121] For the user behavior pattern of the low-frequency intervention type, it indicates that the user is relatively satisfied with the current temperature set value adjustment strategy and makes fewer manual interventions. At this time, the weight ratio of the correction factor can be reduced, for example, from the original 20% to 10%. At the same time, preferentially adopt the parameter adjustment strategy generated from the load prediction data, because in this case, adjusting the parameters according to the load prediction data can better achieve the energy-saving goal, and the user's manual intervention has a relatively small impact on the overall energy-saving effect.

[0122] Step S545: According to the adjusted weight ratio, recalculate the optimization priority of the parameter optimization model for the temperature set value and the operating power threshold.

[0123] After adjusting the weight ratio of the correction factor, it will affect the optimization objectives and constraints of the parameter optimization model for the temperature set value and the operating power threshold. According to the adjusted weight ratio, recalculate the optimization priority. For example, in the case of increasing the weight ratio of the correction factor, the optimization priority of the temperature set value may relatively increase because more consideration should be given to the user's manual operation habits; while the optimization priority of the operating power threshold may be appropriately reduced, but still need to achieve energy saving as much as possible on the premise of ensuring the user's comfort. A multi-objective optimization algorithm, such as the particle swarm optimization algorithm or the ant colony optimization algorithm, can be used to recalculate the optimization priority of the temperature set value and the operating power threshold according to the adjusted weight ratio and the actual constraints to generate a more reasonable parameter adjustment strategy.

[0124] As an implementation manner, the method provided by the embodiment of the present invention may further include:

[0125] Step S600: Real-time monitor the power grid frequency fluctuation value and the voltage stability index of the target area.

[0126] Real-time monitoring of the power grid frequency fluctuation value and voltage stability index in the target area is crucial for ensuring the safe and stable operation of the power grid and the normal operation of air conditioning equipment. High-precision power grid monitoring devices, such as frequency sensors and voltage sensors, can be installed at key nodes of the power grid in the target area. These sensors can collect real-time power grid frequency and voltage data and transmit the data to the data processing center. The data processing center performs real-time analysis and processing on the collected data to calculate the power grid frequency fluctuation value and voltage stability index. For example, the power grid frequency fluctuation value is obtained by calculating the change amplitude of the power grid frequency over a period of time, and the voltage stability index is obtained by analyzing the amplitude and phase changes of the voltage.

[0127] Step S700: Obtain the real-time change rate of the power grid frequency fluctuation value and the phase offset of the voltage stability index, and determine the fluctuation type of the power grid frequency fluctuation value according to the real-time change rate. The fluctuation types include instantaneous spike fluctuation or continuous low-frequency fluctuation.

[0128] The real-time change rate of the power grid frequency fluctuation value can be obtained by performing differential calculation on the power grid frequency fluctuation values at adjacent time points. The phase offset of the voltage stability index can be obtained by analyzing the phase of the voltage signal. The fluctuation type is determined according to the real-time change rate of the power grid frequency fluctuation value. If the real-time change rate is large and the fluctuation duration is short, it can be determined as an instantaneous spike fluctuation, which is usually caused by the sudden connection or disconnection of a large load; if the real-time change rate is small and the fluctuation duration is long, it is determined as a continuous low-frequency fluctuation, which may be due to power imbalance in the power grid or abnormal operation of some equipment. For example, when the power grid frequency fluctuation value rapidly rises by 2Hz from the normal range and then quickly recovers within a short time, it can be judged as an instantaneous spike fluctuation; if the power grid frequency fluctuation value slowly fluctuates at an amplitude of 0.1Hz for a long time, it is a continuous low-frequency fluctuation.

[0129] Step S800: Match the preset emergency adjustment strategy library according to the fluctuation type to determine the first adjustment strategy corresponding to the instantaneous spike fluctuation or the second adjustment strategy corresponding to the continuous low-frequency fluctuation.

[0130] The preset emergency adjustment strategy library is a series of adjustment strategies formulated in advance according to different types of power grid fluctuations and the characteristics of air-conditioning equipment. After determining the type of power grid frequency fluctuation, the corresponding adjustment strategy is matched from the emergency adjustment strategy library. For instantaneous spike fluctuations, the first adjustment strategy may include quickly reducing the operating power of some non-critical air-conditioning equipment to relieve the instantaneous load pressure on the power grid. For example, after detecting an instantaneous spike fluctuation, immediately reduce the operating power of the air-conditioning equipment in the public area by 30%. For continuous low-frequency fluctuations, the second adjustment strategy may be to balance the power of the power grid by gradually adjusting the temperature setting values and operating modes of high-priority air-conditioning equipment. For instance, during continuous low-frequency fluctuations, gradually increase the temperature setting value of the air-conditioning equipment in the data center by 1°C and switch the operating mode to the energy-saving mode.

[0131] Step S900: If the first adjustment strategy is determined, extract the set of current operating power thresholds of the low-priority equipment group from the air-conditioning group control strategy, and generate a set of dynamic power reduction subtasks for the low-priority equipment group according to the phase offset of the voltage stability index.

[0132] When it is determined to adopt the first adjustment strategy, extract the set of current operating power thresholds of the low-priority equipment group from the air-conditioning group control strategy. These thresholds reflect the maximum power allowed for the low-priority equipment group during normal operation. Generate a set of dynamic power reduction subtasks for the low-priority equipment group according to the phase offset of the voltage stability index. The larger the phase offset, the worse the voltage stability of the power grid, and more power needs to be reduced to restore stability. For example, when the phase offset of the voltage stability index exceeds 5°, the generated set of dynamic power reduction subtasks may require each device in the low-priority equipment group to reduce its operating power by 20%; if the phase offset is between 2° and 5°, reduce the operating power by 10%.

[0133] Step S1000: If the second adjustment strategy is determined, synchronously obtain the adjustment margin of the temperature setting value of the high-priority equipment group and the user comfort feedback data, and generate a set of hybrid adjustment subtasks that combine temperature setting value increase and partial equipment power reduction.

[0134] When it is determined to adopt the second adjustment strategy, the temperature setpoint adjustment margin of the high-priority device group and the user comfort feedback data are obtained synchronously. The temperature setpoint adjustment margin refers to the range within which the temperature setpoint of the high-priority device group can be increased without affecting the normal operation of the devices and the basic comfort of the users. The user comfort feedback data can be collected through comfort sensors installed in high-priority areas or user feedback systems. Based on these data, a set of hybrid adjustment subtasks that combines temperature setpoint increase and power reduction of some devices is generated. For example, if the temperature setpoint adjustment margin of the high-priority device group is 2°C and the users feedback that the current ambient temperature is slightly high, the temperature setpoints of some high-priority devices can be increased by 1°C, and at the same time, the operating power of some non-critical high-priority devices can be reduced by 10% to form a set of hybrid adjustment subtasks.

[0135] Step S1100: According to the dynamic power reduction subtask set or the hybrid adjustment subtask set, determine the execution order of the emergency adjustment instructions for each air-conditioning device, and dynamically adjust the priority of the execution order based on the real-time change of the voltage stability index.

[0136] As an implementation, in step S1100, according to the dynamic power reduction subtask set or the hybrid adjustment subtask set, determine the execution order of the emergency adjustment instructions for each air-conditioning device, and dynamically adjust the priority of the execution order based on the real-time change of the voltage stability index, which may specifically include the following steps:

[0137] Step S1110: Analyze the target power reduction amplitude and the allowed execution time window corresponding to each subtask in the dynamic power reduction subtask set, and generate a first task queue.

[0138] When analyzing the dynamic power reduction subtask set, it is necessary to clarify the target power reduction amplitude and the allowed execution time window corresponding to each subtask. The target power reduction amplitude refers to the amount of power that each air-conditioning device needs to reduce, and the allowed execution time window specifies the time range during which the subtask can be executed. By sorting and arranging this information, a first task queue is generated. For example, for a dynamic power reduction subtask set containing 50 subtasks, determine the target power reduction amplitude (such as 1kW, 2kW, etc.) and the allowed execution time window (such as 9:00 - 9:10, 9:10 - 9:20, etc.) for each subtask respectively, and then sort these subtasks in the order of the allowed execution time window to form a first task queue.

[0139] Step S1120: Analyze the temperature setpoint increase amplitude, associated device identifier, and power reduction ratio in the hybrid adjustment subtask set, and generate a second task queue.

[0140] For the hybrid regulation sub-task set, it is necessary to parse the increase amplitude of the temperature set value, the associated device identifier, and the power reduction ratio. The increase amplitude of the temperature set value refers to the number of degrees by which the temperature set value of the air-conditioning device needs to be increased. The associated device identifier is used to determine the specific air-conditioning device for which the regulation task needs to be performed. The power reduction ratio represents the percentage of the power that the device needs to reduce relative to its current operating power. By processing this information, a second task queue is generated. For example, in a hybrid regulation sub-task set, there is a sub-task requiring the temperature set value of air-conditioning devices numbered 001 - 010 to be increased by 1°C and the operating power to be reduced by 15%. These sub-tasks are sorted according to the set rules (such as the device number order) to form the second task queue.

[0141] Step S1130: According to the phase shift direction of the voltage stability index, determine the task types to be processed first in the first task queue and the second task queue. Among them, if the phase shift direction is a positive shift, the temperature set value increase operation in the hybrid regulation sub-task set is preferentially executed; if it is a negative shift, the power reduction operation in the dynamic power reduction sub-task set is preferentially executed.

[0142] The phase shift direction of the voltage stability index reflects the change trend of the grid voltage. When the phase shift direction is a positive shift, it indicates that the grid voltage has an upward trend. At this time, the temperature set value increase operation in the hybrid regulation sub-task set is preferentially executed because increasing the temperature set value can, to a certain extent, reduce the power consumption of the air-conditioning device and at the same time reduce the impact on the voltage increase. For example, when it is detected that the phase of the voltage stability index is positively shifted, the sub-task of increasing the temperature set value in the second task queue is preferentially executed. If the phase shift direction is a negative shift, it indicates that the grid voltage has a downward trend. At this time, the power reduction operation in the dynamic power reduction sub-task set is preferentially executed to relieve the grid load and stabilize the voltage. For instance, during a negative shift, the power reduction sub-task in the first task queue is preferentially processed.

[0143] Step S1140: Based on the task type priority, perform an interleaved sorting on the first task queue and the second task queue to generate an emergency regulation instruction sequence containing the execution timestamp and the device identifier.

[0144] Cross - sort the first task queue and the second task queue according to the determined task type priorities. During the sorting process, the execution timestamp and device identifier of each subtask need to be considered. The execution timestamp is used to determine the execution order of subtasks, and the device identifier is used to uniquely identify each air - conditioner device. Through cross - sorting, an emergency adjustment instruction sequence containing the execution timestamp and device identifier is generated. For example, if the power - reduction task in the first task queue is given priority, during sorting, first insert the power - reduction subtasks in the first task queue that meet the current time range into the emergency adjustment instruction sequence, and then insert the appropriate subtasks in the second task queue to ensure that the entire sequence is arranged in chronological order and task priorities.

[0145] Step S1150: During the execution of the emergency adjustment instruction sequence, continuously monitor the recovery trajectory of the grid frequency fluctuation value. If the trajectory deviates from the preset recovery curve, re - sort the unexecuted instructions according to the device priority identifier. Among them, advance the unexecuted temperature - setting value adjustment instructions in the high - priority device group, and reduce the power - reduction instruction intensity of the low - priority device group.

[0146] During the execution of the emergency adjustment instruction sequence, continuously monitor the recovery trajectory of the grid frequency fluctuation value. The preset recovery curve is a pre - set grid frequency recovery trend based on the characteristics of the power grid and the adjustment target. If it is monitored that the recovery trajectory of the grid frequency fluctuation value deviates from the preset recovery curve, it indicates that the current adjustment strategy may need to be adjusted. At this time, re - sort the unexecuted instructions according to the device priority identifier. Advancing the unexecuted temperature - setting value adjustment instructions in the high - priority device group is to ensure the normal operation of high - priority devices and user comfort. At the same time, reducing the power - reduction instruction intensity of the low - priority device group can avoid excessive power reduction having a greater impact on the environment in the low - priority area. For example, originally planned to reduce the power of a certain device in the low - priority device group by 30%, after re - sorting, reduce the power - reduction instruction intensity to 20%.

[0147] Step S1160: When the grid frequency fluctuation value recovers to the preset safe range, freeze the current unexecuted emergency adjustment instructions and convert them into subsequent optimization input parameters for the parameter optimization model.

[0148] When the grid frequency fluctuation value returns to the preset safe range, it indicates that the power grid has returned to stability. At this time, freeze the current unexecuted emergency adjustment instructions and stop further adjustment operations. Convert these unexecuted emergency adjustment instructions into subsequent optimization input parameters of the parameter optimization model, so that in the subsequent operation process, the parameter optimization model can take into account these unexecuted adjustment tasks and optimize the operating parameters of the air-conditioning equipment more reasonably. For example, input the relevant parameters (such as target power, target temperature, etc.) in the unexecuted power reduction instructions and temperature setting value adjustment instructions into the parameter optimization model as a reference for the next round of optimization.

[0149] Step S1200: During the execution of the emergency adjustment instructions, real-time collect the compressor response delay data of the air-conditioning equipment and the grid frequency recovery rate. If the recovery rate does not reach the preset threshold, perform instruction intensity enhancement processing on the unexecuted adjustment subtasks.

[0150] When executing the emergency adjustment instructions, the compressor response delay data of the air-conditioning equipment can be real-time collected through sensors, which reflects the time from when the compressor receives the adjustment instruction to actually making a response. At the same time, use the power grid monitoring equipment to obtain the grid frequency recovery rate. For example, the preset grid frequency recovery rate threshold is to recover 0.5 Hz every 5 minutes. If after executing the adjustment instructions for a period of time, it is found that the grid frequency only recovers 0.3 Hz every 5 minutes, which does not reach the threshold, then enhance the instruction intensity of the unexecuted adjustment subtasks. For example, increase the instruction of originally reducing the power by 10% to reducing the power by 15% to accelerate the grid frequency recovery.

[0151] As an implementation manner, in step S500, after calibrating the optimization rules of the parameter optimization model according to the updated weight parameters and the operation feedback data of the air-conditioning group, the method provided by the embodiment of the present invention may further include:

[0152] Step S1300: Obtain the equipment response delay parameters and the user manual intervention record set generated during the execution of the air-conditioning group control strategy, and generate a strategy execution deviation data set.

[0153] During the execution of the air conditioner group control strategy, the device response delay parameter reflects the time required for the air conditioner device to actually execute a control instruction after receiving it. The device response delay parameter can be calculated by installing a time monitoring module in the air conditioner device to record the instruction sending time and the actual response time of the device. The user manual intervention record set records the relevant information of the user's manual operation on the air conditioner device during the strategy execution, such as the operation time, operation content, etc. By sorting and analyzing the device response delay parameter and the user manual intervention record set, a strategy execution deviation data set is generated. This data set can intuitively reflect the deviation degree between the air conditioner group control strategy and the expected situation during the execution process. For example, in an area with 200 air conditioner devices, by collecting the response delay parameters and user manual intervention records of each device, a strategy execution deviation data set containing 200 records is generated.

[0154] Step S1400: Determine the calibration priorities of the parameter optimization model for the temperature setting value adjustment strategy and the operating power threshold allocation strategy according to the delay parameter distribution characteristics and intervention frequencies in the strategy execution deviation data set.

[0155] Analyze the delay parameter distribution characteristics in the strategy execution deviation data set, such as the average value and standard deviation of the delay time, to understand the overall situation of the device response delay. The intervention frequency reflects the user's acceptance and intervention degree of the strategy. Determine the calibration priorities of the parameter optimization model for the temperature setting value adjustment strategy and the operating power threshold allocation strategy based on this information. If the delay parameter distribution is relatively scattered and the average delay time is long, it indicates that there are major problems with the device response, and it may be necessary to calibrate the operating power threshold allocation strategy first to ensure that the device can respond to control instructions more promptly. If the user intervention frequency is high, it may be necessary to calibrate the temperature setting value adjustment strategy first to better meet the user's needs. For example, through statistical analysis, it is found that the standard deviation of the delay parameter is large and the user intervention frequency reaches 30%, then it is determined to calibrate the temperature setting value adjustment strategy first.

[0156] Step S1500: Based on the calibration priorities, extract the actual energy consumption deviation rate and the number of temperature deviations corresponding to the high-priority calibration items from the air conditioner group operation feedback data to generate a dynamic calibration parameter set.

[0157] Extract relevant information from the operation feedback data of the air conditioner group according to the determined calibration priorities. For high-priority calibration items, if it is the temperature set value adjustment strategy, extract the actual energy consumption deviation rate and the number of temperature deviations. The actual energy consumption deviation rate refers to the deviation ratio between the actual energy consumption and the predicted energy consumption, and the number of temperature deviations refers to the number of deviations between the actual temperature and the set temperature of the air conditioner equipment. Organize this information into a dynamic calibration parameter set. For example, if the temperature set value adjustment strategy is calibrated preferentially, extract the actual energy consumption deviation rate and the number of temperature deviations of each air conditioner equipment from the operation feedback data to form a dynamic calibration parameter set containing multiple parameters.

[0158] Step S1600: Input the dynamic calibration parameter set into the parameter optimization model, and iteratively correct the temperature set value constraint conditions and power distribution ratios in the optimization rules item by item to generate the corrected optimization rules.

[0159] As an implementation, in step S1600, input the dynamic calibration parameter set into the parameter optimization model, and iteratively correct the temperature set value constraint conditions and power distribution ratios in the optimization rules item by item, which may specifically include the following steps:

[0160] Step S1610: Extract the set of device identifiers corresponding to the actual energy consumption deviation rate from the dynamic calibration parameter set, and determine the historical temperature set value adjustment records and current operation modes of each air conditioner equipment in the set of device identifiers.

[0161] Extract the set of device identifiers corresponding to the actual energy consumption deviation rate from the dynamic calibration parameter set. These device identifiers are used to determine the air conditioner equipment that needs to be analyzed in detail. For each air conditioner equipment in the set of device identifiers, obtain its historical temperature set value adjustment record, which may specifically include information such as adjustment time and adjustment amplitude, and at the same time determine its current operation mode (such as cooling mode, heating mode, energy-saving mode, etc.). For example, the dynamic calibration parameter set contains the identifiers of 50 air conditioner equipments with a relatively large actual energy consumption deviation rate. By querying the database, obtain the historical temperature set value adjustment records and current operation modes of these 50 equipments.

[0162] Step S1620: Calculate the correction amount of the temperature set value constraint condition according to the deviation direction and amplitude in the historical temperature set value adjustment record, and determine the adjustable range of the power distribution ratio based on the current operation mode.

[0163] Adjust the deviation direction (upward or downward) and amplitude in the record according to the historical temperature set value, and calculate the correction amount of the temperature set value constraint condition. If the actual energy consumption deviation rate increases after the historical temperature set value is increased, it indicates that the upper limit of the temperature set value may need to be reduced, and vice versa, the lower limit may need to be increased. Determine the adjustable range of the power distribution ratio based on the current operating mode. Different operating modes have different power requirements. For example, the power requirement may be relatively large in the cooling mode and relatively small in the energy-saving mode. Therefore, the adjustable range of the power distribution ratio is also different in different operating modes. For example, in the cooling mode, the adjustable range of the power distribution ratio may be relatively small, while in the energy-saving mode, it may be relatively large.

[0164] Step S1630: Input the correction amount and the adjustable range into the rule engine of the parameter optimization model, traverse each constraint condition in the optimization rules, and synchronously update the upper limit, lower limit of the temperature set value and the power distribution weight related to the device identification set.

[0165] Input the calculated correction amount of the temperature set value constraint condition and the adjustable range of the power distribution ratio into the rule engine of the parameter optimization model. The rule engine will traverse each constraint condition in the optimization rules and synchronously update the upper limit, lower limit of the temperature set value and the power distribution weight related to the device identification set. For example, if the correction amount of the upper limit of the temperature set value of a certain device is to be reduced by 1°C, the rule engine will correspondingly reduce the upper limit of the temperature set value of this device by 1°C in the optimization rules. At the same time, according to the adjustable range of the power distribution ratio, adjust the power distribution weight of this device to ensure more reasonable power distribution.

[0166] Step S1640: During the synchronous update process, detect the policy conflict between the temperature set value constraint condition and the power distribution ratio. If there is a conflict, according to the high-frequency adjustment device identification in the user manual intervention record set, preferentially retain the update result of the temperature set value constraint condition.

[0167] During the synchronous update of the temperature set value constraint condition and the power distribution ratio, a policy conflict may occur. For example, reducing the upper limit of the temperature set value may lead to an increase in power demand, while reducing the power distribution ratio at the same time may cause the device to fail to reach the set temperature. At this time, according to the high-frequency adjustment device identification in the user manual intervention record set, preferentially retain the update result of the temperature set value constraint condition. Because the high-frequency adjustment device identification indicates that the user pays more attention to the temperature setting of these devices, preferentially retaining the update result of the temperature set value constraint condition can better meet the user's needs. For example, if a certain device appears in the user manual intervention record set with high-frequency adjustment and there is a conflict between the temperature set value constraint condition and the power distribution ratio during the update process, then preferentially retain the update result of the temperature set value constraint condition of this device.

[0168] Step S1650: Match and verify the updated result after conflict detection with the future period load distribution in the load prediction data. If the verification passes, integrate the updated constraint conditions and allocation ratios into the revised optimization rule.

[0169] Match and verify the updated result after conflict detection with the future period load distribution in the load prediction data. The purpose of the match verification is to ensure that the updated constraint conditions and allocation ratios can adapt to the grid load conditions in the future period. For example, check whether the updated power allocation ratio will cause the grid load to be too high during peak load periods. If the verification passes, it indicates that the updated constraint conditions and allocation ratios are reasonable, and integrate them into the revised optimization rule. In this way, the optimization rule of the parameter optimization model can be made more in line with the actual situation and improve the effect of air-conditioning energy-saving control.

[0170] Step S1700: Recalculate the air-conditioning group control strategy for the future period according to the revised optimization rule and execute the updated control strategy in the target area.

[0171] According to the revised optimization rule, use the parameter optimization model to recalculate the air-conditioning group control strategy for the future period. During the calculation process, factors such as equipment response delay, user manual intervention, and actual energy consumption deviation are considered, making the new control strategy more accurate and reasonable. Then execute the updated control strategy in the target area. Send control instructions to each air-conditioning device through the intelligent control system to achieve precise control of the air-conditioning devices. For example, in a commercial park, calculate the air-conditioning group control strategy for the next 24 hours according to the revised optimization rule, and then send control instructions to each air-conditioning device through the intelligent air-conditioning control system of the park to ensure that the air-conditioning devices operate according to the new strategy, achieving better energy-saving and comfort effects.

[0172] Please refer to Figure 2 , Figure 2Schematic structural diagram of an air conditioner energy-saving control system provided by an embodiment of the present invention. The air conditioner energy-saving control system may be a computer system, such as a background server remotely connected to an air conditioner cluster, or a computer device deployed in an air conditioner control machine room. The air conditioner energy-saving control system at least includes a processor 101, a communication interface 102, and a memory 103. Among them, the processor 101, the communication interface 102, and the memory 103 can be connected through a bus or other means. Among them, the processor 101 (or Central Processing Unit (CPU)) is the computing core and control core of the air conditioner energy-saving control system, which can parse various instructions in the air conditioner energy-saving control system and process various data of the air conditioner energy-saving control system. The communication interface 102 may optionally include a standard wired interface, a wireless interface (such as WI-FI, a mobile communication interface, etc.), and under the control of the processor 101, it can be used for sending and receiving data; the communication interface 102 can also be used for the transmission and interaction of internal data of the air conditioner energy-saving control system. The memory 103 (Memory) is a memory device in the air conditioner energy-saving control system, used to store programs and data. It can be understood that the memory 103 here can include both the built-in memory of the air conditioner energy-saving control system, and of course, it can also include the extended memory supported by the air conditioner energy-saving control system. The memory 103 provides a storage space, and the operating system of the air conditioner energy-saving control system is stored in this storage space, which may include but is not limited to: Android system, iOS system, Windows Phone system, etc., and the present invention does not make any limitations in this regard.

[0173] In one embodiment, the processor 101 executes the air conditioner energy-saving control method based on a virtual power plant provided above in the embodiments of the present invention by running a computer program in the memory 103.

Claims

1. An air conditioning energy-saving control method based on a virtual power plant, characterized in that: The method comprises: Obtaining operating parameter sets and real-time grid load data of multiple air-conditioning devices in a target area, and determining air-conditioning energy-saving control models corresponding to the multiple air-conditioning devices; the air-conditioning energy-saving control model includes a load prediction model and a parameter optimization model, wherein the load prediction model is used to generate load prediction data for a future period based on historical load data and real-time environmental parameters, and the parameter optimization model is used to generate parameter adjustment strategies corresponding to each air-conditioning device based on the load prediction data; Obtain the current temperature setting value, operating power threshold and device priority identification of each air-conditioning device in the target area; Dividing a plurality of load adjustment intervals in the future period according to the peak load period and the valley load period in the load forecast data; In the parameter optimization model, a maximum load value and a minimum load value corresponding to the power grid load fluctuation threshold are set, and based on the maximum load value and the minimum load value, an air conditioning equipment opening and closing strategy corresponding to each load adjustment interval is generated; Based on the equipment priority identifier, the multiple air-conditioning equipment are grouped to determine a high-priority equipment group and a low-priority equipment group, and according to the air-conditioning equipment on / off strategy, the operating power threshold of the low-priority equipment group is lowered during the peak load period, and the temperature setting value of the high-priority equipment group is increased during the valley load period; Combining the adjusted operating power threshold and temperature setting value into an optimized device operating parameter subset; According to the optimized subset of equipment operating parameters, adjusting the temperature setting values ​​and operating modes of the plurality of air-conditioning equipment in the future period, and generating an air-conditioning group control strategy corresponding to the future period; Acquire actual grid load data and air conditioning group operation feedback data in the future period, and update the weight parameter of the load prediction model according to the deviation value between the actual grid load data and the load prediction data; According to the updated weight parameters and the air-conditioning group operation feedback data, the optimization rules of the parameter optimization model are calibrated to generate a calibrated air-conditioning energy-saving control model.

2. The method according to claim 1, characterized in that The step of grouping the plurality of air-conditioning devices based on the device priority identifier to determine a high-priority device group and a low-priority device group includes: Obtain historical usage frequency data of each air conditioning device, user-preset preferred temperature range and equipment energy consumption level; Determine the usage activity score of each air-conditioning device according to the historical usage frequency data, and generate a comprehensive priority score for each device in combination with the user's preset preferred temperature range and the device energy consumption level; Classify the first device set whose comprehensive priority score is greater than a preset threshold as the high-priority device group, and classify the remaining second device set as the low-priority device group; In the parameter optimization model, assigning a first optimization weight to the high-priority device group and a second optimization weight to the low-priority device group, wherein the first optimization weight is greater than the second optimization weight; According to the first optimization weight and the second optimization weight, parameter adjustment operations are performed on the high-priority device group and the low-priority device group within the load adjustment interval respectively.

3. The method according to claim 2, characterized in that The performing parameter adjustment operations on the high-priority device group and the low-priority device group in the load adjustment interval according to the first optimization weight and the second optimization weight, respectively, includes: During the peak load period, determining the maximum power reduction ratio allowed for the low-priority device group according to the second optimization weight thereof, and generating a corresponding power adjustment instruction set based on the maximum power reduction ratio; During the valley load period, according to the first optimization weight of the high-priority equipment group, the allowable temperature setting value increase range thereof is determined, and a corresponding temperature adjustment instruction set is generated based on the user-preset tendency temperature range; Inputting the power adjustment instruction set and the temperature adjustment instruction set into the parameter optimization model for conflict detection, if it is detected that there is a resource conflict between the temperature adjustment instruction of the high-priority device group and the power adjustment instruction of the low-priority device group, the temperature adjustment instruction of the high-priority device group is executed first; The optimized device operating parameter subset is generated according to the adjustment instruction after the conflict detection.

4. The method according to claim 1, characterized in that The updating of the weight parameter of the load prediction model according to the deviation value between the actual power grid load data and the load prediction data comprises: Extracting the predicted load change sequence corresponding to the future time period in the load forecast data, and obtaining the actual load change sequence for the same time period in the actual power grid load data; Calculating the load difference between the predicted load change sequence and the actual load change sequence at each time point to generate a deviation value sequence; Determining the weighted error amount of the load forecasting model on the temperature-related parameters and the time-related parameters according to the fluctuation amplitude and duration of the deviation value sequence; Using a gradient descent algorithm to back-propagate the weight error, and adjusting the connection weights of the neural network layer in the load forecasting model; The adjusted connection weights are smoothed with the historical weight data to generate updated weight parameters.

5. The method according to claim 4, characterized in that The step of back-propagating the weight error using a gradient descent algorithm to adjust the connection weights of the neural network layer in the load forecasting model includes: According to the distribution characteristics of historical load data, real-time ambient temperature, humidity and user behavior characteristics in the input layer nodes, the initial weight distribution ratio corresponding to each neuron in the hidden layer nodes is determined; Based on the fluctuation amplitude of the deviation value sequence and the predicted load curve shape of the future period in the load forecast data, the error propagation path between the input layer node and the hidden layer node is calculated to generate the weight gradient correction direction of the hidden layer node; Iteratively adjusting the initial weight distribution ratio layer by layer according to the weight gradient correction direction, and obtaining the deviation value change rate between the updated predicted load curve and the actual load curve after each iteration; If the deviation value change rate does not show a downward trend during N consecutive iterations, the step size parameter of the gradient descent algorithm is dynamically switched according to the trend change direction of the actual load curve, and the weight gradient correction direction is recalculated, N>2; When the average value of the deviation value sequence reaches a preset error threshold, the connection weight after the current iteration is locked, and the fit between the connection weight and the predicted load data output by the activation function of the load forecasting model is verified, so as to terminate the adjustment process after the fit meets the preset stability condition.

6. The method according to claim 1, characterized in that The step of calibrating the optimization rules of the parameter optimization model according to the updated weight parameters and the air conditioning group operation feedback data to generate the calibrated air conditioning energy-saving control model includes: Obtaining the actual energy consumption value of the equipment, the number of temperature deviations, and the frequency of manual intervention by the user from the operation feedback data of the air conditioning group; Calculating an energy consumption deviation ratio according to the actual energy consumption value of the equipment and the predicted energy consumption value in the optimized equipment operation parameter subset; If the energy consumption deviation ratio exceeds a preset calibration threshold, determining a calibration amount of the parameter optimization model on the temperature setting value adjustment strategy according to the number of temperature deviations and the user's manual intervention frequency; Adding a user behavior correction factor to the parameter optimization model, and dynamically adjusting the influence coefficient of the correction factor according to the user's manual intervention frequency; The calibration amount and the correction factor are integrated into the optimization rule to generate the calibrated air conditioning energy-saving control model.

7. The method according to claim 6, characterized in that The adding of the user behavior correction factor in the parameter optimization model and dynamically adjusting the influence coefficient of the correction factor according to the user manual intervention frequency includes: Counting the number of temperature setting value changes of each air-conditioning device due to manual operation by the user in the future period, and generating a user intervention record set; Determining the user behavior pattern type according to the distribution density and time distribution characteristics of the user intervention record set; If the user behavior pattern type is a frequent intervention type, increasing the weight ratio of the correction factor in the temperature setting value adjustment strategy; If the user behavior pattern type is a low-frequency intervention type, the weight ratio of the correction factor is reduced, and the parameter adjustment strategy generated by the load forecast data is preferentially adopted; According to the adjusted weight ratio, the optimization priority of the parameter optimization model on the temperature setting value and the operating power threshold is recalculated.

8. The method according to claim 1, characterized in that The method further comprises: Real-time monitoring of power grid frequency fluctuation values ​​and voltage stability indicators in the target area; Acquire the real-time change rate of the power grid frequency fluctuation value and the phase offset of the voltage stability index, and determine the fluctuation type of the power grid frequency fluctuation value according to the real-time change rate, wherein the fluctuation type includes instantaneous peak fluctuation or continuous low-frequency fluctuation; According to the fluctuation type, a preset emergency adjustment strategy library is matched to determine a first adjustment strategy corresponding to the instantaneous peak fluctuation or a second adjustment strategy corresponding to the continuous low-frequency fluctuation; If the first adjustment strategy is determined, extracting the current operating power threshold set of the low-priority device group from the air-conditioning group control strategy, and generating a dynamic power reduction subtask set for the low-priority device group according to the phase offset of the voltage stability indicator; If the second regulation strategy is determined, the temperature setting value adjustment margin and user comfort feedback data of the high-priority device group are synchronously obtained to generate a hybrid regulation subtask set that combines temperature setting value increase with power reduction of some devices; Determine the execution order of the emergency adjustment instructions of each air-conditioning device according to the dynamic power reduction subtask set or the mixed adjustment subtask set, and dynamically adjust the priority of the execution order based on the real-time change of the voltage stability index; During the execution of the emergency adjustment instruction, the compressor response delay data and the grid frequency recovery rate of the air-conditioning equipment are collected in real time. If the recovery rate does not reach a preset threshold, the instruction strength enhancement processing is performed on the unexecuted adjustment subtask.

9. An air conditioning energy-saving control system, characterized in that: include: a memory, wherein a computer program is stored in the memory; A processor is used to load the computer program to implement the air conditioning energy-saving control method based on a virtual power plant as described in any one of claims 1-8.

Images