Air conditioner dispatching method and device based on virtual power plant response

Abstract

The invention provides an air conditioner dispatching method and device based on virtual power plant response. The method comprises the steps that a real-time operation parameter set of all air conditioner terminals in a target area in a preset time window is obtained; performing feature alignment processing on the real-time operation parameter set and a dynamic regulation and control instruction issued by a virtual power plant to generate a joint input feature; the combined input features are input into a pre-trained multi-target optimization model, hierarchical weight distribution is conducted on the combined input features through a feature fusion layer in the multi-target optimization model, an air conditioner regulation and control priority sequence is generated, and a target scheduling strategy is generated based on the priority sequence; and the target scheduling strategy is issued to each air conditioner terminal in the target area to be executed, a feedback parameter set after execution is collected in real time, and incremental parameter updating is conducted on the multi-target optimization model through the feedback parameter set. According to the invention, the accuracy, real-time performance and sustainability of air conditioner terminal scheduling can be improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to an air conditioning scheduling method and device based on virtual power plant response. Background Art

[0002] In the prior art, when virtual power plants achieve grid load regulation by aggregating air-conditioning terminals, they mostly adopt static scheduling methods based on fixed rules or single-objective optimization, which have significant defects. There may be dynamic conflicts between multiple objectives such as grid peak-shaving demand, user comfort, and equipment energy efficiency. Traditional strategies use fixed priorities (such as grid priority), which leads to rigid strategies and cannot adapt to the coordinated needs of real-time electricity price fluctuations, environmental parameter changes, and user preference migration. It is easy to cause a sharp drop in user satisfaction or an increase in energy consumption caused by frequent equipment start-up and shutdown; secondly, static models rely on offline training of historical data, and the input features are limited to static parameters such as rated power and fixed electricity prices. There is a lack of modeling of real-time response capabilities. When sudden control instructions are issued, the strategy execution deviation causes the load adjustment to fail. In addition, the existing methods lack a closed-loop mechanism from strategy execution to feedback learning, and the problem of model performance degradation in long-term operation cannot be solved. In summary, the existing technologies are difficult to meet the requirements of virtual power plants for the accuracy, real-time performance, and sustainability of air-conditioning terminal scheduling. Summary of the invention

[0003] The present invention provides an air conditioning scheduling method and device based on virtual power plant response.

[0004] According to one aspect of the present invention, there is provided an air conditioning scheduling method based on virtual power plant response, the method comprising: Acquire a real-time operating parameter set of each air-conditioning terminal in the target area within a preset time window, wherein the real-time operating parameter set includes current energy consumption characteristics, ambient temperature and humidity characteristics, and user preference setting characteristics, wherein the user preference setting characteristics are used to describe the user's priority configuration of the air-conditioning operating mode; Performing feature alignment processing on the real-time operating parameter set and the dynamic control instructions issued by the virtual power plant to generate a joint input feature, wherein the dynamic control instructions include the load distribution strategy of the virtual power plant to the target area and the real-time electricity price fluctuation characteristics; The combined input features are input into a pre-trained multi-objective optimization model, and the combined input features are weighted in layers through a feature fusion layer in the multi-objective optimization model to generate an air conditioning control priority sequence, wherein the priority sequence is used to describe the scheduling order and parameter adjustment range of each air conditioning terminal under a dynamic control instruction; Generate a target scheduling strategy based on the priority sequence, the target scheduling strategy including the start and stop time configuration of each air-conditioning terminal, the temperature setting threshold adjustment range and the energy consumption constraint condition; The target scheduling strategy is sent to each air-conditioning terminal in the target area for execution, and a feedback parameter set after execution is collected in real time, and the feedback parameter set is used to perform incremental parameter update on the multi-objective optimization model.

[0005] According to another aspect of the present invention, there is provided an air conditioning scheduling device based on virtual power plant response, comprising: A parameter acquisition module is used to acquire a real-time operating parameter set of each air-conditioning terminal in a target area within a preset time window, wherein the real-time operating parameter set includes current energy consumption characteristics, ambient temperature and humidity characteristics, and user preference setting characteristics, wherein the user preference setting characteristics are used to describe the user's priority configuration of the air-conditioning operation mode; A feature alignment module, used to perform feature alignment processing on the real-time operating parameter set and the dynamic control instructions issued by the virtual power plant to generate a joint input feature, wherein the dynamic control instructions include the load distribution strategy of the virtual power plant to the target area and the real-time electricity price fluctuation characteristics; A priority generation module, used for inputting the joint input features into a pre-trained multi-objective optimization model, performing hierarchical weight allocation on the joint input features through a feature fusion layer in the multi-objective optimization model, and generating an air conditioning control priority sequence, wherein the priority sequence is used to describe the scheduling order and parameter adjustment range of each air conditioning terminal under a dynamic control instruction; A strategy generation module, used to generate a target scheduling strategy based on the priority sequence, wherein the target scheduling strategy includes the start and stop time configuration of each air-conditioning terminal, the temperature setting threshold adjustment range and the energy consumption constraint condition; The incremental update module is used to send the target scheduling strategy to each air-conditioning terminal in the target area for execution, collect the feedback parameter set after execution in real time, and use the feedback parameter set to perform incremental parameter update on the multi-objective optimization model.

[0006] The present invention has at least the following beneficial effects: The air conditioning scheduling method based on the virtual power plant response provided by the present invention generates joint input features by acquiring the real-time operating parameter set of each air conditioning terminal in the target area and combining it with the dynamic control instructions issued by the virtual power plant; hierarchical weight allocation is performed on the joint input features based on the multi-objective optimization model to generate an air conditioning control priority sequence, and further generate a target scheduling strategy including start and stop time, temperature threshold and energy consumption constraints; the model is incrementally updated by executing the feedback parameter set to form a dynamic optimization closed loop. Based on this, the real-time operating parameter set integrates the physical operating status, environmental parameters and user preferences of the air conditioning terminal, and after feature alignment with the dynamic control instructions of the virtual power plant, the generated joint input features simultaneously cover multi-dimensional information of power grid demand, equipment capacity and user behavior. By performing hierarchical weight allocation on the above features through the multi-objective optimization model, the conflict between the peak-shaving demand of the power grid, user comfort and equipment energy efficiency can be dynamically balanced, avoiding the policy bias problem caused by single-objective optimization in traditional methods, and significantly improving the adaptability of the scheduling strategy in complex scenarios. In addition, by mapping the load adjustment target value with the energy consumption fluctuation curve in the real-time operation parameters in time segments and integrating the gradient change characteristics of electricity price fluctuations, the generated joint input features can accurately characterize the spatiotemporal correlation between power grid demand and terminal operation status. Compared with traditional static strategies, this mechanism can capture the impact of electricity price fluctuations on user preferences in real time, and dynamically adjust the equipment scheduling order and parameter adjustment range in the priority sequence, so as to achieve minute-level response when the power grid load is urgently adjusted, and reduce the risk of scheduling failure caused by instruction delay. The multi-objective optimization model learns the trade-off between power grid stability, user satisfaction and energy efficiency in historical data through pre-training. When generating the priority sequence, it not only allocates the equipment scheduling order based on the current joint input features, but also embeds the prediction information of the equipment response capability through the parameter adjustment range. In this way, the target scheduling strategy can reserve the temperature adjustment buffer space allowed by user preferences while meeting the hard constraints of virtual power plant load distribution, avoiding the interruption of strategy execution caused by equipment response delay or sudden change of user behavior, and realizing the unity of strategy rigid demand and elastic fault tolerance. Finally, by collecting the feedback parameter set after the strategy is executed in real time and comparing it with the model prediction results, a multi-objective training loss value is generated to drive the incremental update of the model parameters, so that the model can continuously learn the dynamic changes such as power grid operation mode switching, equipment aging and attenuation, and user preference migration, avoiding the performance degradation of traditional models caused by environmental time changes. At the same time, the incremental update process only optimizes local parameters for abnormal data points, which improves the adaptability of the model while retaining the ability to generate stable strategies in historical training.In summary, the present invention solves the technical problem of real-time coordinated optimization of power grid demand, user comfort and equipment energy consumption in air-conditioning scheduling in a virtual power plant environment through the synergistic effect of multi-source data fusion, dynamic feature alignment, multi-objective elastic strategy generation and closed-loop incremental update, and significantly improves the accuracy, response speed and long-term stability of the scheduling strategy.

[0007] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The accompanying drawings exemplarily illustrate the embodiments and constitute a part of the specification, and together with the text description of the specification, are used to explain the exemplary implementation of the embodiments. The embodiments shown are for illustrative purposes only and do not limit the scope of the claims. In all drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0009] Figure 1 A schematic diagram of an application scenario of an air conditioning scheduling method based on virtual power plant response according to an embodiment of the present invention is shown.

[0010] Figure 2 A flow chart of an air conditioning scheduling method based on virtual power plant response according to an embodiment of the present invention is shown.

[0011] Figure 3 A schematic diagram of the functional module architecture of an air conditioning scheduling device based on virtual power plant response according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0012] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be appreciated by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present invention. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0013] In the present invention, unless otherwise specified, the use of the terms "first", "second", etc. to describe various elements is not intended to limit the positional relationship, temporal relationship, or importance relationship of these elements, and such terms are only used to distinguish one element from another element. In some examples, the first element and the second element may refer to the same instance of the element, and in some cases, based on the description of the context, they may also refer to different instances.

[0014] Figure 1The application scenario of the embodiment of the present invention is shown, and specifically provides an air conditioning scheduling system 100. The air conditioning scheduling system 100 includes one or more air conditioning terminals 101, a server 120, and one or more communication networks 110 coupling the one or more air conditioning terminals 101 to the server 120.

[0015] In an embodiment of the present invention, the server 120 may run one or more services or software applications that enable the execution of an air conditioning scheduling method based on virtual power plant response.

[0016] exist Figure 1 In the configuration shown, the server 120 may include one or more components that implement the functions performed by the server 120. These components may include software components, hardware components, or a combination thereof that are executable by one or more processors.

[0017] Network 110 may be any type of network known to those skilled in the art that may support data communications using any of a variety of available protocols, including but not limited to TCP / IP, SNA, IPX, etc.

[0018] Server 120 may include one or more general purpose computers, dedicated server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running virtual operating systems, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that may be virtualized to maintain a server's virtual storage device). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0019] The air conditioning scheduling system 100 may also include one or more databases 130. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 130 may be used to store historical data. The databases 130 may reside in various locations. For example, the database used by the server 120 may be local to the server 120, or may be remote from the server 120 and may communicate with the server 120 via a network-based or dedicated connection.

[0020] Please refer to Figure 2 The air conditioning scheduling method based on virtual power plant response provided by the embodiment of the present invention may include the following steps: Step S100: Obtain the real-time operating parameter set of each air-conditioning terminal in the target area within a preset time window. The real-time operating parameter set includes current energy consumption characteristics, ambient temperature and humidity characteristics, and user preference setting characteristics. The user preference setting characteristics are used to describe the user's priority configuration of the air-conditioning operating mode.

[0021] In the target area, the preset time window is, for example, a continuous time interval set by a virtual power plant, and its length can be dynamically adjusted according to the grid load forecast demand, for example, set to a variable period of 5 minutes to 30 minutes. The current energy consumption characteristics are collected in real time through the current sensor and power metering module of the air-conditioning terminal, specifically including parameters such as the compressor operating frequency, air supply wind speed, and the difference sequence between the set temperature and the actual temperature. For example, a certain air-conditioning terminal records that the compressor operating frequency is 45Hz, the air supply wind speed is 3m / s, and the difference between the set temperature of 24℃ and the actual temperature of 26℃ has not converged for 2 minutes. The environmental temperature and humidity characteristics are obtained through the temperature and humidity sensors deployed in the space where the air-conditioning terminal is located, including indoor temperature, humidity and outdoor environmental temperature gradient change data, for example, the indoor temperature of a certain area drops linearly from 28℃ to 26℃ within the time window, and the humidity maintains a fluctuation of 60%±2%. The user preference setting characteristics are collected through the mobile terminal application or the air-conditioning control panel, specifically manifested as the temperature sensitivity coefficient set by the user, the energy-saving priority flag, and the allowed adjustment time period. For example, user A sets the temperature sensitivity coefficient to 0.8 (0-1 range, the higher the value, the lower the tolerance for temperature fluctuations), activates the energy-saving priority flag, and specifies 14:00-16:00 every day as the time period allowed for regulation. The above three types of features are transmitted to the regional data center through the Internet of Things communication protocol, and after timestamp alignment processing, a real-time operating parameter set with spatiotemporal consistency is formed. The data structure of the parameter set is a multidimensional time series matrix, in which each air-conditioning terminal corresponds to a row in the matrix, and each characteristic parameter is arranged as a column vector according to the sampling frequency within the time window. This step S100 uses a data synchronization mechanism to ensure that the operating status and environmental parameters of all air-conditioning terminals can be correlated and analyzed under a unified time reference, providing complete input conditions for subsequent dynamic regulation.

[0022] As an implementation manner, the above step S100, obtaining the real-time operating parameter set of each air-conditioning terminal in the target area within a preset time window, may specifically include: Step S110: The operating status data of each air-conditioning terminal in the target area in multiple continuous time segments are collected through the Internet of Things gateway. The operating status data includes the compressor operating frequency, air supply speed, and the difference sequence between the set temperature and the actual temperature.

[0023] As the data intermediary node between the air-conditioning terminal and the central control system in the target area, the IoT gateway uses the MQTT protocol or Modbus protocol to establish a two-way communication link with each terminal controller, and collects operating status data at a fixed time segment. The length of the time segment is dynamically configured according to the scheduling requirements of the virtual power plant, and is typically set to an adjustable range of 30 seconds to 5 minutes. For example, in the scenario where the grid frequency regulation needs to be urgent, the time segment is shortened to 30 seconds to ensure high-frequency collection of instantaneous changes in the compressor operating frequency, while in the normal scheduling mode it is extended to 5 minutes to reduce the communication load. The compressor operating frequency is obtained in real time through the inverter control module of the air-conditioning terminal, which represents the speed fluctuation of the compressor motor per unit time. For example, a terminal records that the operating frequency rises linearly from 45Hz to 48Hz and remains stable within the time segment. The air supply speed is measured by the Hall sensor of the built-in fan of the air conditioner, which is specifically expressed as the air flow speed generated by the rotation of the fan impeller. For example, the air supply speed of a terminal maintains a steady-state value of 3.2m / s±0.1m / s in multiple consecutive time segments. The difference sequence between the set temperature and the actual temperature is obtained by periodic sampling of the temperature sensor. The difference sequence is sorted by timestamp to form discrete time series data. For example, if the set temperature of a terminal is 24°C, and the actual temperature in three consecutive time segments is 25.3°C, 24.8°C, and 24.5°C, the difference sequence is [1.3, 0.8, 0.5]. This step uses the multiplexing communication mechanism of the IoT gateway to synchronously collect the operating status data of all air-conditioning terminals in the target area, and encapsulates it into JSON format data packets for transmission to the edge computing node to ensure that the data collection process meets the requirements of low latency and high reliability.

[0024] Step S120: Perform outlier elimination and timestamp alignment processing on the operating status data to generate a standardized operating parameter set; wherein, the timestamp alignment processing includes uniformly mapping sensor data of different sampling frequencies to the same time reference axis, and outlier elimination includes identifying compressor current values ​​or temperature and humidity readings that exceed preset physical thresholds.

[0025] Exemplarily, the outlier removal process is implemented based on a preset physical threshold library, which stores the rated operating parameter ranges of various air-conditioning terminals. For example, the normal range of compressor current of a certain model of air-conditioning is 3A-6A. When the instantaneous current value collected is 8A, the system determines it as abnormal data and triggers the removal operation, and fills the missing values ​​with linear interpolation. The abnormal detection of temperature and humidity readings adopts a dynamic threshold adjustment strategy. For example, when the reported value of the indoor temperature sensor rises by 5°C compared with the previous period and exceeds the range of the outdoor temperature gradient change, it is determined to be a sensor failure and the data point is removed. The timestamp alignment process is implemented through a time reference axis mapping algorithm, which resamples data with different sampling frequencies based on the global clock of the IoT gateway. For example, the compressor operating frequency data is sampled at intervals of 10 seconds, while the air supply speed data is sampled at intervals of 30 seconds. During the alignment process, the two types of data are uniformly mapped to the time reference axis with an interval of 1 minute, and the data at the missing time points are supplemented by cubic spline interpolation. The generation of standardized operating parameter sets includes two stages: data normalization and unit unification: the normalization process scales each parameter to the [0,1] interval, such as mapping the compressor operating frequency 45Hz-55Hz to 0.45-0.55; the unit unification process converts parameters of different dimensions into standard units, such as converting the temperature difference from Fahrenheit to Celsius. The processed standardized operating parameter set is a structured data matrix, whose row dimension represents the air conditioning terminal identifier and the column dimension contains the normalized parameter values ​​under the timestamp sequence, ensuring that the subsequent analysis has temporal and spatial consistency.

[0026] Step S130: extracting the preference setting instruction submitted by the user through the mobile terminal, the preference setting instruction including the temperature sensitivity coefficient, the energy saving priority flag and the allowed adjustment time period.

[0027] Exemplarily, the user preference setting instruction is input through the interactive interface of the mobile terminal application and transmitted to the user preference database of the virtual power plant in an encrypted form. The temperature sensitivity coefficient is defined as a quantitative indicator of the user's tolerance for temperature fluctuations, with a value range of 0-1, where 0 indicates that the user can accept larger temperature fluctuations, and 1 indicates that the temperature is required to be strictly maintained at the set value. For example, if user A sets the temperature sensitivity coefficient to 0.9, the system will give priority to ensuring the temperature stability of its air-conditioning terminal during scheduling. The energy-saving priority flag is a Boolean parameter. When its value is true, it indicates that the user allows the virtual power plant to automatically adjust the air-conditioning operation mode during peak electricity price periods to reduce energy consumption. For example, when the flag is activated, the air-conditioning terminal can accept a 10% reduction in the compressor frequency. The allowed control time period is a time interval specified by the user that is acceptable to external scheduling intervention. For example, user B sets 14:00-16:00 every day as the allowed control time period. Outside this time period, the system shall not modify the operating parameters of its air-conditioning terminal. The extraction process of preference setting instructions includes three steps: data decryption, format conversion and validity verification. The decryption step uses an asymmetric encryption algorithm to restore the user's original input. The format conversion converts text instructions into numerical feature vectors. For example, the allowed adjustment time period "14:00-16:00" is converted into a start timestamp of 1620000000 and an end timestamp of 1620007200. The validity verification detects the legitimacy of the input parameters through the rule engine. For example, when the temperature sensitivity coefficient exceeds 1, the system automatically truncates it to 1 and generates an alarm log.

[0028] Step S140: Feature splicing of the standardized operating parameter set and the preference setting instruction to generate a real-time operating parameter set containing temporal and spatial correlation.

[0029] Exemplarily, the feature splicing process can be implemented by a matrix expansion algorithm, and the two-dimensional matrix of the standardized operating parameter set is aligned with the three-dimensional feature vector of the user preference setting instruction in the time and space dimensions. Specifically, the parameter row vector corresponding to each timestamp of the standardized operating parameter set (such as [0.45, 3.2, 0.5] represents the compressor frequency, air supply wind speed and temperature difference) is horizontally spliced ​​with the corresponding user's preference feature vector (such as [0.9, 1, 1620000000] represents the temperature sensitivity coefficient, the energy-saving priority flag and the start time of the control permission) to form an expanded composite feature vector. The establishment of spatiotemporal correlation depends on the joint encoding of the timestamp and the spatial position. For example, a unique geographic location code (such as GPS coordinates) is assigned to each air-conditioning terminal in the target area, and it is used together with the timestamp as a composite index key. The real-time operating parameter set is finally stored as a spatiotemporal tensor structure, whose dimensions include four parts: timestamp sequence, air-conditioning terminal spatial coordinates, operating status parameters and user preference parameters. For example, the parameter entry of a terminal at timestamp 1620000000 can be expressed as (1620000000, (116.404, 39.915), 0.48, 3.1, 0.6, 0.9, 1, 1620000000), where the first three values ​​are standardized operating parameters and the last three values ​​are user preference parameters. This step enables the subsequent scheduling algorithm to consider the real-time status of the device, environmental factors and user personalized needs at the same time through the deep fusion of feature space, laying a data foundation for generating high-precision control strategies.

[0030] Step S200: perform feature alignment processing on the real-time operating parameter set and the dynamic control instructions issued by the virtual power plant to generate a joint input feature. The dynamic control instructions include the load distribution strategy of the virtual power plant for the target area and the real-time electricity price fluctuation characteristics.

[0031] Exemplarily, the dynamic control instruction is a set of control requirements generated by the central controller of the virtual power plant, and its load distribution strategy specifically includes the total load value to be adjusted in the target area, the load change rate limit and the load distribution ratio in different time periods. For example, the virtual power plant requires that the total load in the target area be reduced by 50kW within the next 15-minute time window, and the load reduction rate shall not exceed 10kW / minute, of which 30% load reduction shall be completed in the first 5 minutes. The real-time electricity price fluctuation characteristics are obtained through the power market trading platform, which are expressed as the time-of-use electricity price curve and its gradient change rate. For example, the electricity price in the current period is 0.5 yuan / kWh, and it is predicted that it will rise to 0.6 yuan / kWh in the next period. The feature alignment process first parses the time constraint parameters in the dynamic control instruction and maps and matches them with the time window of the real-time operation parameter set. For example, the 5-minute initial load reduction target required in the load distribution strategy is associated with the historical energy consumption data of the air-conditioning terminal in the first 5 minutes of the current time window. The equipment response rate data is obtained by calculating the product of the average command response time in the historical dispatch record of the air-conditioning terminal and the current compressor operating frequency. For example, if the historical average response delay of a terminal is 20 seconds and the current compressor frequency is 50Hz, the response rate is quantified as 20 seconds × 50Hz = 1000 response units. The feature superposition process breaks down the load adjustment target into multiple time segments, and performs matrix convolution operations with the equipment response capabilities of the corresponding time period to generate a load adjustment vector sequence. The real-time electricity price fluctuation feature is element-wise weighted and spliced ​​with the load adjustment vector through time dimension expansion, where the weight coefficient is dynamically calculated by the power grid stability assessment model. For example, when the power grid frequency deviation exceeds 0.1Hz, the weight of the electricity price feature is increased to 0.7 to ensure that the economic goal is subject to the stability requirement. The final generated joint input feature is a high-dimensional tensor structure, whose dimensions include multiple axes such as time segment index, air-conditioning terminal identification, load adjustment amount, electricity price weight and environmental parameters, realizing the deep fusion of multi-source heterogeneous data.

[0032] As an implementation mode, step S200 performs feature alignment processing on the real-time operation parameter set and the dynamic control instruction issued by the virtual power plant to generate a joint input feature, which may specifically include: Step S210: parse the load distribution strategy in the dynamic control instruction, extract the load adjustment target value related to the air-conditioning terminal and the corresponding time constraint parameters, the time constraint parameters include the start time of the load adjustment, the maximum allowed delay time and the staged adjustment ratio.

[0033] Exemplarily, the parsing process of the dynamic control instruction is based on the structured data message issued by the central control system of the virtual power plant, in which the load distribution strategy is encapsulated in JSON format, for example, and may include the total load adjustment amount, adjustment direction (increase / decrease) and constraints of the target area within the specified time window. The load adjustment target value represents the power change that the air-conditioning terminal group needs to complete collaboratively, for example, requiring the target area to reduce the total load by 50kW in the next scheduling cycle. The starting time in the time constraint parameter is the triggering time point of the load adjustment operation, in the form of a Unix timestamp accurate to milliseconds, for example, the starting time is set to 1620000000 (corresponding to 00:00:00 UTC on May 3, 2021). The maximum allowed delay time refers to the maximum tolerable time from the issuance of the instruction to the completion of the parameter adjustment of the air-conditioning terminal, for example, it is set to 120 seconds. If a terminal does not reach the target adjustment value within 120 seconds, it is marked as a response timeout. The phased adjustment ratio decomposes the total load adjustment target into multiple sub-targets according to time segments, for example, requiring 30% load reduction (15kW) to be completed within the first 5 minutes, and the remaining 70% to be completed within the next 10 minutes. During the parsing process, a regular expression matching algorithm is used to extract key parameters, and type conversion and range verification are performed on numerical data, such as verifying whether the sum of the phased adjustment ratios is 100%. If an abnormality is detected, the command retransmission mechanism is triggered.

[0034] Step S220: extracting a current energy consumption feature subset that matches the time constraint parameters from the real-time operation parameter set, wherein the current energy consumption feature subset includes the energy consumption fluctuation curve and equipment response rate data of each air-conditioning terminal within a preset time window before the start time; wherein the equipment response rate data is obtained by calculating the product of the average time from receiving the instruction to completing the parameter adjustment of the air-conditioning terminal in the historical scheduling record and the operating frequency of the compressor under the current operating state.

[0035] Exemplarily, the preset time window can be dynamically set according to the start time of load adjustment, and its length is, for example, the historical data backtracking cycle. For example, the energy consumption data 30 minutes before the start time is extracted to analyze the recent fluctuation trend. The energy consumption fluctuation curve is composed of the power sampling values ​​of the air-conditioning terminal arranged in time series. For example, the power value sequence of a terminal in the time window is [3.2kW, 3.0kW, 2.8kW], reflecting the trend of its energy consumption decreasing over time. The calculation of the equipment response rate data depends on the historical scheduling record database, which stores the delay time and execution effect of each terminal's previous command execution. For example, in the past 10 schedulings, the average delay time from receiving the shutdown command to the complete stop of the compressor of a terminal is 25 seconds. The current compressor operating frequency is 45Hz, and the equipment response rate is quantified as 25 seconds × 45Hz = 1125 seconds · Hz. This value represents the adjustment ability of the terminal under a specific operating state. The higher the value, the lower the response efficiency. The extraction process uses a timestamp matching algorithm to filter out energy consumption data segments that are aligned with the current time constraint parameters. For example, when the starting time is 1620000000, all power sampling points from 1620000000 to 1620001800 (30 minutes) are extracted, and the data missing period caused by communication interruption is eliminated.

[0036] Step S230: splitting the load adjustment target value into sub-load adjustment values ​​corresponding to a plurality of continuous time segments based on the time constraint parameter, and mapping each time segment to a corresponding fluctuation period in the energy consumption fluctuation curve.

[0037] Exemplarily, the time segment division is determined based on the phased adjustment ratio and the maximum allowed delay time. For example, the total load adjustment target of 50kW is completed in two stages: 15kW is reduced within 5 minutes in the first stage, and 35kW is reduced within 10 minutes in the second stage. The sub-load adjustment value of each time segment is allocated to each air-conditioning terminal through a linear programming algorithm to ensure that the cumulative adjustment amount meets the stage target. When mapping to the energy consumption fluctuation curve, a sliding window matching mechanism is used to identify the fluctuation period corresponding to the time segment. For example, if a terminal shows periodic changes in the historical energy consumption fluctuation curve (such as a power peak every 10 minutes), the sub-load adjustment value of the first stage of 5 minutes is mapped to its next expected trough period. In this process, the start and end timestamps of the time segment are aligned with the phase of the energy consumption fluctuation curve, for example, the 5 minutes of the first stage (1620000000-1620000300) are mapped to the falling edge period of the energy consumption fluctuation of a terminal (the power drops from 3.5kW to 3.0kW during 1620000000-1620000300). This mapping relationship is achieved by calculating the time series similarity through the dynamic time warping algorithm (DTW), ensuring that the sub-load adjustment value is coordinated with the actual energy consumption change trend of the terminal.

[0038] Step S240: Superimpose the sub-load adjustment value of each time segment with the equipment response rate data in the corresponding fluctuation period to generate a load adjustment vector sequence including time sensitivity.

[0039] Exemplarily, the feature superposition process can use matrix convolution operation to perform element-by-element weighted fusion of the sub-load adjustment value matrix and the equipment response rate matrix according to the time dimension. For example, the sub-load adjustment value of a certain time segment is 5kW, and the equipment response rate of the corresponding terminal is 900 seconds·Hz. The superimposed feature vector contains the adjustment amount (5kW), the response rate (900 seconds·Hz) and the product of the two (4500kW·seconds·Hz). Time sensitivity is quantified by introducing a time attenuation factor. For example, the time segment closer to the current moment is given a higher weight coefficient. Specifically, for the t-th time segment, the attenuation factor can be set to 1 / (1+0.1t), so that the adjustment task adjacent to the operation window obtains a more significant decision weight. The generated load adjustment vector sequence is a three-dimensional tensor structure, whose dimensions include the time segment index, the air conditioning terminal identifier and the fused feature value (adjustment amount, response rate, time sensitivity). For example, the vector of terminal A in the first phase time segment (index 0) can be expressed as [0, A, 5kW, 900 seconds·Hz, 0.91], where 0.91 is the calculated attenuation factor.

[0040] Step S250: Dimensionally expand and fuse the load adjustment vector sequence with the real-time electricity price parameters in the dynamic control instructions. The dimensional expansion and fusion includes splitting the real-time electricity price parameters into gradient change segments according to time segments and performing element-level weighted splicing with the positions with the same time index in the vector sequence to generate a joint input feature; wherein the weight coefficient of the weighted splicing is determined by the grid stability priority under the current operating mode of the virtual power plant.

[0041] Exemplarily, real-time electricity price parameters can be obtained from the electricity market information platform and split into gradient change segments with the same time granularity as the load adjustment vector sequence according to time segments. For example, if the electricity price corresponding to a time segment rises from 0.5 yuan / kWh to 0.6 yuan / kWh, it is split into three step values ​​[0.5, 0.55, 0.6], and each step value is spliced ​​with the load adjustment feature corresponding to the time index in the vector sequence. The weight coefficient of element-level weighted splicing is dynamically calculated by the power grid stability assessment model, which outputs the stability priority (0-1 interval) based on parameters such as real-time frequency deviation and voltage fluctuation rate. For example, when the frequency deviation is detected to exceed 0.1Hz, the power grid stability priority rises to 0.8, at which time the weight coefficient of the electricity price feature is reduced to 0.2, and the weight of the load adjustment feature is increased to 0.8 to ensure that the stability goal takes precedence over the economic goal. The spliced ​​joint input feature is a four-dimensional tensor, including the time segment, terminal identifier, load-electricity price fusion feature and stability weight. For example, the joint features of terminal B in time segment 1 can be expressed as [1, B, 3kW, 800s·Hz, 0.85, 0.55 yuan / kWh, 0.2], where 0.2 is the electricity price weight coefficient. This feature structure provides complete input conditions for the subsequent multi-objective optimization model to take into account both grid operation safety and economic dispatch.

[0042] Step S300: Input the joint input features into the pre-trained multi-objective optimization model, and perform hierarchical weight assignment on the joint input features through the feature fusion layer in the multi-objective optimization model to generate an air-conditioning control priority sequence. The priority sequence is used to describe the scheduling order and parameter adjustment range of each air-conditioning terminal under the dynamic control instructions.

[0043] Exemplarily, the pre-trained multi-objective optimization model can be constructed using a deep reinforcement learning framework, and its feature fusion layer includes a composite structure of a multi-head attention mechanism and a gated recurrent unit. The hierarchical weight allocation process first calculates the importance score of each air-conditioning terminal feature through the attention mechanism, for example, based on the current electricity price fluctuation characteristics, a higher attention weight is given to the terminal with higher energy consumption. The feature fusion layer further uses a temporal convolutional network to extract the temporal dependency of the joint input features, for example, identifying the trend characteristics of a terminal's compressor frequency continuing to rise in three consecutive time segments. The generation of the air-conditioning control priority sequence includes two core stages: initial priority score calculation and dynamic correction. The initial score is calculated by the policy network based on the output of the feature fusion layer, specifically including the scheduling order score (0-1 interval, the higher the value, the higher the priority scheduling) and the recommended value of the parameter adjustment amplitude (such as the temperature setting value adjustment ±2℃). The dynamic correction process introduces the device response delay feature, and models the historical scheduling data through the gated recurrent unit. For example, if a terminal has an average delay of 8 seconds in the past three schedulings, the current priority score will be attenuated and compensated according to the delay variance. The urgency parameter is issued by the virtual power plant in real time. When the load adjustment demand emergency threshold exceeds the preset threshold (such as the grid frequency deviation exceeds 0.2Hz), the terminal with a score higher than 0.8 in the priority sequence will be marked as an immediate response device, and its parameter adjustment range is subject to the maximum allowable offset in the user preference setting feature. For example, if the maximum allowable temperature offset set by user B is 3°C, even if the model recommends an adjustment of 4°C, the actual dispatch instruction will be automatically truncated to 3°C. The final generated priority sequence is a structured data table, which contains the terminal identification, dispatch time (accurate to seconds), temperature setting value adjustment amount, and maximum allowable energy consumption limit.

[0044] As an implementation manner, in step S300, the priority sequence generation process may include the following steps: Step S310: sorting the joint input features by feature importance through the attention mechanism layer in the multi-objective optimization model to generate an initial priority score for each air-conditioning terminal.

[0045] Exemplarily, as described above, the attention mechanism layer in the multi-objective optimization model can be a multi-head self-attention structure, which determines the key influencing factors of each air-conditioning terminal under the dynamic control instruction by calculating the associated weights of different dimensions in the joint input features. The joint input features are used as the input tensor of the attention layer, and its dimensions include time segment index, air-conditioning terminal identifier, load adjustment amount, real-time electricity price parameters and grid stability weight. For example, for the input feature vector [0.6kW, 0.55 yuan / kWh, 0.8] of a certain air-conditioning terminal in time segment T1, the attention mechanism calculates its correlation weight with other terminals in the same time period through query-key value matching. Specifically, the query vector is generated by linear transformation of the load adjustment amount and the real-time electricity price parameter, the key vector is composed of the device response rate and the grid stability weight, and the attention score is normalized to the [0,1] interval by the Softmax function. The initial priority score is generated by the dot product operation of the attention weight and the feature vector. For example, if the load adjustment weight of a terminal is 0.7 and the electricity price parameter weight is 0.3, the initial score is 0.7×0.6kW+0.3×0.55 yuan / kWh = 0.645. This score reflects the comprehensive contribution of the terminal to the stability and economy of the power grid within the current regulation window. The higher the score, the higher the dispatch priority.

[0046] Step S320: Input the initial priority score into the gated loop unit, perform timing correction based on the device response delay characteristics in the historical scheduling records, and generate a corrected dynamic priority sequence; wherein the device response delay characteristics include the average delay time and variance data of each air-conditioning terminal from receiving the instruction to completing the execution in the historical scheduling.

[0047] The input layer of the gated recurrent unit (GRU) receives the initial priority score sequence, and its hidden state dimension is aligned with the device response delay feature in the historical scheduling record. The device response delay feature is represented in the form of a two-dimensional vector, which contains the historical average delay time and delay variance. For example, the average delay time of a terminal in the historical scheduling is 25 seconds, and the variance is 4 seconds. 2, then its delay feature vector is [25, 4]. During the timing correction process, the update gating unit of the GRU dynamically adjusts the influence weight of the historical delay feature according to the initial priority score of the current time segment and the hidden state of the previous moment. For example, if the current initial score is 0.8 and the hidden state of the previous moment indicates a high delay risk, the update gating coefficient is set to 0.6 and the reset gating coefficient is set to 0.4, then the corrected priority score is 0.6×0.8+0.4×(1-25 / 50) = 0.62, where the 25-second delay is converted to a 0.5 delay impact factor through normalization. The output of the dynamic priority sequence is a timestamp-aligned corrected score matrix. For example, the score sequence of terminal A in the time window [14:00-14:05] is [0.62, 0.71, 0.68], reflecting the result of its priority being dynamically adjusted with the device's response capability.

[0048] Step S330: According to the load adjustment urgency parameter issued by the virtual power plant, the dynamic priority sequence is threshold segmented, and the air-conditioning terminals with priority scores higher than the emergency threshold are marked as immediate response devices.

[0049] Exemplarily, the load adjustment urgency parameter is issued in real time by the central control system of the virtual power plant, and its value range is [0,1]. The larger the value, the higher the scheduling urgency. For example, when the grid frequency deviation exceeds 0.2Hz, the urgency parameter is set to 0.9. The emergency threshold is dynamically calculated by a linear interpolation algorithm, and the formula is: threshold = basic threshold + urgency parameter × adjustment coefficient. Assuming that the basic threshold is 0.6 and the adjustment coefficient is 0.3, the threshold corresponding to the urgency parameter 0.9 is 0.6+0.9×0.3 = 0.87. Terminals with scores higher than 0.87 in the dynamic priority sequence are marked as immediate response devices, and their identification information and parameter adjustment instructions are preferentially added to the real-time scheduling queue. For example, if the corrected score of a terminal in time segment T2 is 0.89, which exceeds the threshold of 0.87, it is marked as an immediate response device and triggers a 10% frequency reduction control operation of the compressor. The device list after threshold segmentation is broadcast to all terminals in the target area through the message queue protocol of the edge controller to ensure real-time synchronization of scheduling instructions.

[0050] Step S340: Gradient limit the parameter adjustment range of the immediate response device so that the temperature setting value does not change more than the maximum allowable offset in the user preference setting feature.

[0051] Exemplarily, the gradient limiting module truncates the temperature setting value adjustment suggestion according to the maximum allowable offset in the user preference setting feature. For example, if the maximum allowable temperature offset set by user B is ±2°C, and the multi-objective optimization model recommends an adjustment of -3°C, the actual adjustment range is limited to -2°C. The limiting algorithm uses the projected gradient descent method to map the out-of-bounds adjustment value to the allowed range. Specifically, let the original adjustment suggestion be ΔT and the allowed offset be ΔT max , then the corrected adjustment amount ΔT clipped = sign(ΔT) × min(|ΔT|, ΔT max ). For example, ΔT = -3°C, ΔT max = 2°C, then ΔT clipped = -2℃. The gradient limiting process synchronously updates the parameter adjustment amplitude field in the priority sequence, for example, the temperature setting value of terminal C is adjusted from "-2.5℃" to "-2.0℃", and a limit flag is added to the dispatch instruction. This mechanism ensures that the user comfort constraint is not exceeded while maintaining the maximum feasible execution effect of the grid dispatch demand.

[0052] Step S400: Generate a target scheduling strategy based on the priority sequence, the target scheduling strategy including the start and stop time configuration of each air-conditioning terminal, the temperature setting threshold adjustment range and the energy consumption constraint condition.

[0053] The generation process of the target scheduling strategy first parses the immediate response device identifier in the priority sequence and extracts the corresponding start and stop time configuration parameters. The start and stop time configuration is generated by the timing planning algorithm of the edge controller. For example, a terminal needs to shut down the compressor within 30 seconds after the start of scheduling and restart it in a reduced frequency mode after 5 minutes. The temperature setting threshold adjustment range is determined based on the recommended amplitude in the priority sequence and the user preference setting characteristics. For example, the model recommends adjusting -2℃, and the user allows a maximum deviation of -3℃, then the actual adjustment range is set to the optimal value in the interval [-2℃, -3℃]. The energy consumption constraint is calculated by a linear programming algorithm to ensure that the cumulative load adjustment of all air-conditioning terminals in the target area meets the load distribution strategy of the virtual power plant, and the energy consumption of a single terminal does not exceed 85% of its rated power. For example, if the rated power of a terminal is 3kW, its maximum allowable power during scheduling is limited to 2.55kW. After receiving the target scheduling strategy, the edge controller converts it into a device control instruction set, which contains a time-stamped drive signal sequence, such as sending a compressor stop instruction at Unix timestamp 1620000000, and sending an instruction to adjust the set temperature to 26°C at 1620000300. The execution timing optimization algorithm further introduces a buffer interval parameter to prevent multiple devices from starting and stopping at the same time and causing grid shocks. For example, a 50-millisecond time offset trigger mechanism is set for terminals under the same distribution circuit.

[0054] As an implementation manner, in step S400, the execution process of the target scheduling strategy may specifically include: Step S410: parse the air-conditioning terminal identifiers marked as immediate response devices in the priority sequence, extract the corresponding start / stop time configurations and temperature setting threshold adjustment ranges, and generate a device control instruction set.

[0055] Exemplarily, the generation process of the target scheduling strategy takes the priority sequence as the core input, and generates an executable set of equipment control instructions by parsing the scheduling order and parameter adjustment range of each air-conditioning terminal. The start and stop time configuration is determined according to the scheduling time field in the priority sequence. For example, a terminal is assigned to shut down the compressor at timestamp 1620000300 and restart in reduced frequency mode at 1620000600. The temperature setting threshold adjustment range is constrained by the recommended range in the priority sequence and the maximum allowable offset in the user preference setting feature. For example, the priority sequence recommends an adjustment of -2.5℃, and the user allows a maximum offset of -3℃, then the actual adjustment range is set to the optimal value in the interval [-2.5℃, -3℃]. The energy consumption constraint condition is calculated by a linear programming algorithm to ensure that the cumulative load adjustment of all terminals in the target area meets the load distribution strategy of the virtual power plant, and the instantaneous power of a single terminal does not exceed 85% of its rated power. For example, if the rated power of a terminal is 3kW, the maximum allowable power limit during its scheduling is 2.55kW, and it is calibrated in real time by the power monitoring module of the edge controller.

[0056] Exemplarily, the immediate response device identifier is identified by the emergency flag bit in the priority sequence. For example, if the priority score of a terminal is 0.89 and the flag bit is "urgent", its identifier is added to the immediate response queue. The start and stop time configuration is extracted from the scheduling time field of the priority sequence, specifically including the compressor shutdown time, restart time and frequency gradient parameters in the transition phase. For example, the shutdown time of terminal A is 1620000300, the restart time is 1620000600, and the frequency drops linearly from 45Hz to 40Hz. The temperature setting threshold adjustment range is parsed from the parameter adjustment field of the priority sequence. For example, the temperature setting value of terminal B is adjusted from 24℃ to 22℃, and the maximum hysteresis is allowed to be ±0.5℃. The device control instruction set is encapsulated in JSON format, including the terminal identifier, operation type (start / stop / temperature adjustment), execution timestamp and parameter threshold, for example {"ID":"AC-001", "Action":"SetTemp", "Time":1620000300, "Value":22}.

[0057] Step S420: sort the device control instruction set by timestamp and send it to the edge controller in the target area. The edge controller adjusts the start and stop time of the compressor according to the buffer interval parameter in the instruction and generates a drive signal set including the execution timing.

[0058] Exemplarily, the timestamp sorting algorithm can be arranged in ascending order to ensure that the instructions are executed according to the planned timing. For example, the shutdown instruction (1620000300) of terminal A takes precedence over the temperature adjustment instruction (1620000450) of terminal B. Among them, the buffer interval parameter can be dynamically set by the virtual power plant to prevent the simultaneous start and stop of multiple devices from causing grid shocks, such as setting the start and stop interval of equipment under the same distribution circuit to 50 milliseconds. The edge controller adjusts the actual execution time according to the buffer interval. For example, the shutdown instruction of terminal A originally planned to be executed at 1620000300 is postponed to 1620000300.050 due to the buffer interval requirement. After the drive signal set is generated, it is converted into a Modbus RTU protocol frame, which includes the device address, function code and data field. For example, the shutdown instruction frame of terminal A is [01 06 0001 0000 CRC].

[0059] Step S430: Real-time collection of actual operating parameters of each air-conditioning terminal after executing the driving signal set, including compressor state switching delay, temperature setting value execution deviation and instantaneous energy consumption fluctuation range, to generate a dynamic adjustment parameter set.

[0060] The delay time of compressor state switching records the time difference between the instruction sending and the execution completion through a high-precision clock stamp. For example, terminal A plans to shut down at 1620000300.050, and the actual completion time is 1620000302.100, with a delay of 2.05 seconds. The execution deviation of the temperature setting value is calculated by the difference between the actual temperature and the target temperature fed back by the air conditioning control board. For example, if the setting is 22℃ and the actual temperature is stable at 22.5℃, the deviation is 0.5℃. The instantaneous energy consumption fluctuation amplitude is quantified by the standard deviation of the sampling value of the power sensor. For example, the standard deviation of the power fluctuation of terminal B within 5 minutes after adjustment is 0.3kW. The dynamic adjustment parameter set is stored in a time series database. The data structure includes timestamp, terminal identifier, delay time, temperature deviation and power fluctuation value, for example, {"Time":1620000300, "ID":"AC-001", "Delay":2.05, "TempDev":0.5, "PowerStd":0.3}.

[0061] Step S440: Compare the dynamically adjusted parameter set with the parameter adjustment range in the priority sequence, identify the abnormal device identifier whose execution deviation exceeds the allowable threshold, and generate a policy adjustment trigger event.

[0062] Optionally, the threshold can be set adaptively according to the device type and user preference, for example, the compressor delay threshold is set to 3 seconds, the temperature deviation threshold is set to 1°C, and the power fluctuation threshold is set to 10% of the rated power. The comparison process adopts a window sliding mechanism, for example, the dynamic adjustment parameter set is checked every 5 minutes, and if the delay of terminal A reaches 3.2 seconds (exceeding the 3-second threshold), it is marked as an abnormal device. The policy adjustment trigger event contains the abnormal device identification, deviation type and severity level, for example {"EventID":"E-001", "Device":"AC-001", "Type":"DelayExceed", "Level":"High"}, and is sent to the virtual power plant dispatch center through the message queue service.

[0063] Step S450: According to the policy adjustment trigger event, the multi-objective optimization model is called to locally recalculate the current priority sequence, generate an updated target scheduling strategy and reissue it for execution, and at the same time, the feedback parameter set associated with the abnormal device identifier is marked as a high-priority sample for incremental update.

[0064] Exemplarily, the local recalculation process only updates the joint input features of the sub-area where the abnormal device is located. For example, the feedback parameter set of terminal A is injected into the model input layer, triggering the weight fine-tuning of the feature fusion layer. The updated target scheduling strategy is optimized by the differential evolution algorithm, for example, reducing the priority score of terminal A from 0.89 to 0.75 and extending its buffer interval to 100 milliseconds. Incrementally updated high-priority samples are added to the training data set with ten times the weight, for example, the feedback parameters of terminal A are copied 10 times in the historical data pool to accelerate the model's adaptation to abnormal scenarios.

[0065] As an implementation mode, the process of handling the policy adjustment triggering event may specifically include: Step S451: parsing the execution deviation type corresponding to the abnormal device identification, the execution deviation type including at least one of compressor response delay exceeding the standard, temperature setting value lock failure or instantaneous energy consumption exceeding the limit.

[0066] Exemplarily, the deviation type parsing can be achieved by matching the abnormal event description field through regular expressions. For example, the event description "DelayExceed" corresponds to the compressor response delay exceeding the standard, and "TempLockFail" corresponds to the temperature setting value lock failure. The basis for determining whether the compressor response delay exceeds the standard is that the actual delay exceeds the preset threshold, for example, the 3-second threshold is exceeded. The temperature setting value lock failure means that the air conditioning control board fails to adjust the temperature as instructed, for example, the feedback temperature continues to deviate from the set value by more than 2°C for 5 minutes. The instantaneous energy consumption exceeds the limit by comparing the power fluctuation amplitude with the percentage of the rated power. For example, the terminal rated power is 3kW, and the instantaneous power reaches 3.3kW (10% over the limit).

[0067] Step S452: Match the preset adjustment rule library according to the deviation type, and extract candidate adjustment strategies associated with the current virtual power plant operation mode, where the candidate adjustment strategies include equipment parameter compensation coefficients, execution timing redundancy or priority weight correction values.

[0068] Exemplarily, the adjustment rule base can be organized in a decision tree structure, where the nodes are combinations of deviation types and grid operation modes. For example, when encountering a "DelayExceed" deviation in "Emergency" mode, candidate strategies include a device parameter compensation coefficient of 1.2 (extending the buffer interval by 20%), increasing the execution timing margin by 50 milliseconds, and reducing the priority weight correction value by 0.1. The device parameter compensation coefficient is obtained through historical data regression analysis. For example, for every 1 second increase in delay exceeding the standard, the compensation coefficient is increased by 0.1. The execution timing margin is dynamically calculated based on the variance of the device response rate, for example, a variance of 4 seconds. 2 The corresponding redundancy is 50 milliseconds.

[0069] Step S453: perform constraint verification on the candidate adjustment strategies and the actual operating parameters in the dynamic adjustment parameter set, eliminate invalid strategies that violate the user preference setting characteristics or the grid stability boundary, and generate a feasible strategy set.

[0070] Exemplarily, the constraint verification process can use a linear programming solver to verify the feasibility of the strategy. For example, the candidate strategy recommends adjusting the temperature setting value of terminal A to increase the compensation factor by 1.2, but the maximum offset allowed in the user preference setting is -3°C. If the original adjustment of -2.5°C becomes -3.0°C after compensation, the strategy is valid; if it is compensated to -3.1°C, it will be eliminated due to crossing the boundary. The grid stability boundary check simulates the grid state after the strategy is executed through the power flow calculation model. For example, if a strategy causes the distribution line load rate to exceed 95%, it will be marked as invalid.

[0071] Step S454: Input the feasible strategy set into the feature fusion layer of the multi-objective optimization model, perform a preliminary assessment of the strategy benefits in combination with the real-time electricity price fluctuation characteristics in the current joint input features, and generate an updated priority sequence candidate set.

[0072] Exemplarily, the feature fusion layer encodes feasible strategies and generates strategy feature vectors. For example, the strategy "increase buffer interval by 50 milliseconds" is encoded as [0, 50, 0, 0], and the strategy "reduce priority weight by 0.1" is encoded as [0, 0, -0.1,0]. The real-time electricity price fluctuation characteristics extract trend information through the time convolution network, for example, the electricity price rises by 0.05 yuan / kWh per hour during the current period. The benefit pre-evaluation calculates the comprehensive score of each strategy through the value network. For example, strategy A scores 0.85 (grid 0.4, user 0.3, energy consumption 0.15), and strategy B scores 0.78.

[0073] Step S455: Perform sandbox simulation on the priority sequence candidate set through the edge controller, collect the power grid load forecast data and user satisfaction forecast score after the simulation execution, and select the candidate strategy with the best comprehensive evaluation as the new target scheduling strategy for execution.

[0074] Exemplarily, the sandbox simulation environment builds a digital twin model of the target area, injects candidate strategies and runs simulations. The grid load forecast data is generated by the improved gray model, for example, it is predicted that the total load will be reduced by 65kW±2kW after the execution of strategy A. The user satisfaction prediction score is calculated based on the transfer learning model of historical feedback data, for example, strategy A is expected to get 4.2 / 5.0 points. The comprehensive evaluation adopts the weighted summation method, for example, the grid indicator weight is 0.5, the user weight is 0.3, the energy consumption weight is 0.2, the total score of strategy A is 0.85×0.5+4.2×0.3+0.78×0.2=1.923, and the total score of strategy B is 1.845. Finally, strategy A is selected to update the target scheduling strategy.

[0075] Step S500: Send the target scheduling strategy to each air-conditioning terminal in the target area for execution, collect the feedback parameter set after execution in real time, and use the feedback parameter set to perform incremental parameter update on the multi-objective optimization model.

[0076] For example, the target scheduling strategy can be sent to each air-conditioning terminal through an industrial Internet of Things protocol (such as MQTT or OPC UA), and the terminal controller generates a drive signal after parsing the instruction. The feedback parameter set includes the actual delay time of the compressor state switching, the temperature setting value execution deviation, and the instantaneous power fluctuation data. For example, a terminal plans to shut down the compressor within 30 seconds, and the actual completion time is 32 seconds, resulting in a 2-second delay record. The incremental parameter update adopts an online learning mechanism. First, the feedback parameters are compared with the model prediction values ​​to generate an error distribution matrix. For example, terminal data with a temperature setting value execution deviation exceeding 1°C is marked as a high-weight sample. The model update process adopts a selective neuron adjustment strategy, and only the hidden layer neurons associated with abnormal data are optimized by gradient descent, such as the weight matrix related to the delay feature in the third-layer GRU unit. The version fallback mechanism maintains two model copies on the edge server: the main version accepts real-time updates, and the shadow version is synchronized every 5 minutes. When the scheduling accuracy of the main version on the validation set drops by more than 3%, the system automatically switches to the shadow version and triggers an alarm. The updated model parameters are transmitted to each edge node through AES-256 encryption to ensure that the data transmission process complies with the safety specifications of the power monitoring system. This step forms a closed-loop optimization mechanism, which enables the multi-objective optimization model to dynamically adapt to external disturbance factors such as equipment aging and user behavior changes, and continuously improve the accuracy and robustness of the scheduling strategy.

[0077] As an implementation manner, in step S500, the incremental parameter update process may specifically include the following steps: Step S510: Compare the feedback parameter set with the expected execution effect data to generate a model prediction error distribution graph.

[0078] Exemplarily, the feedback parameter set may include the actual operating data of the air-conditioning terminal after executing the scheduling strategy, including the delay time of the compressor state switching, the execution deviation of the temperature setting value, and the instantaneous energy consumption fluctuation amplitude. The expected execution effect data is derived from the predicted value of the multi-objective optimization model when generating the priority sequence. For example, the model predicts that the delay time of a terminal is 2 seconds, the temperature deviation is 0.3°C, and the standard deviation of energy consumption fluctuation is 0.2kW. The difference comparison algorithm uses tuple-level difference calculation to align the actual parameters and the predicted values ​​by timestamp and then calculate the difference item by item to generate an error sequence. For example, the error between the actual delay of 3 seconds and the predicted 2 seconds is +1 second, and the error between the temperature deviation of 0.5°C and the predicted 0.3°C is +0.2°C. The error distribution diagram is generated by the kernel density estimation (KDE) algorithm, with the horizontal axis representing the error value range and the vertical axis representing the probability density of the error. For example, the delay error is concentrated in the interval of [-0.5 seconds, +1.5 seconds], and the temperature deviation error is distributed in the range of [-0.3℃, +0.7℃], forming a multi-peak distribution pattern, which intuitively reflects the prediction deviation tendency of the model under different working conditions.

[0079] Step S520: extracting abnormal data points that exceed the tolerance interval in the error distribution graph, and tracing back to the generation logic path of the corresponding scheduling strategy.

[0080] Exemplarily, the tolerance interval can be adaptively set according to the equipment type and scheduling objectives, for example, the compressor delay tolerance interval is [-1 second, +3 seconds], the temperature deviation tolerance interval is [-0.5℃, +1.0℃], and the energy consumption fluctuation tolerance interval is [-0.5kW, +0.5kW]. Abnormal data points are identified through a boundary detection algorithm. For example, if the delay error of a terminal +3.2 seconds exceeds the upper limit, it is marked as abnormal. The tracing process relies on the timestamp index of the scheduling log and the model decision tree path to locate the policy generation link corresponding to the abnormal data. For example, the delay exceeding standard data is associated with the third layer of the GRU unit of the feature fusion layer in the log. Backtracking finds that the equipment response rate characteristics were not correctly weighted when the scheduling strategy was generated. The tracing results are stored in the form of a logical path diagram, including the timestamp of the abnormal data, the identifier of the impact model layer, and the associated weight matrix index.

[0081] Step S530: performing local gradient descent optimization on the neuron connection weights associated with the abnormal data points in the multi-objective optimization model.

[0082] Local gradient descent optimization focuses on specific network layers affected by abnormal data, such as the weight matrix of the second attention head of the feature fusion layer. The optimization process calculates the partial derivative of the error corresponding to the abnormal data with respect to the target weight, which is expressed as ∂L / ∂W i j, where L is the loss function, W i j is the weight of the i-th row and j-th column. The learning rate is set to 1 / 10 of the global learning rate. For example, when the global learning rate is 0.001, the local learning rate is 0.0001 to prevent overfitting. For example, for abnormal data with excessive delay error, the weight matrix W3 of the third fully connected layer of the policy network is adjusted through error back propagation, and the update amount is ΔW3=η·(y_true-y_pred)·x i nput, where η = 0.0001, x i nput is the corresponding joint input feature vector. After optimization, the delay prediction error of the model for similar working conditions is reduced to within 0.8 seconds.

[0083] Step S540: retain the historical parameter version that performs stably on the preset validation set, and automatically roll back to the most recent stable version when the performance of the incrementally updated model drops by more than a threshold.

[0084] Exemplarily, historical parameter versions can be stored in the version control database in the form of snapshots, and each version is accompanied by a validation set evaluation indicator (such as a grid benefit score, a user satisfaction score). The validation set contains 10% of the data that was not involved in the training in the historical period, such as 24 hours of data randomly selected each week in the past three months. The performance degradation threshold is set to a relative decrease of 5% in the comprehensive evaluation index. For example, if the comprehensive score of the current version is 0.85 and drops to 0.8075 (a decrease of 5.0%) after the incremental update, the rollback mechanism is triggered. The rollback process compares the model hash value of the latest version with the most recent stable version, and loads the parameter file of the stable version on the edge controller. For example, after the performance of version V2.1 degrades due to overfitting, the system automatically rolls back to V2.0 and generates an alarm log.

[0085] Step S550: encrypt and transmit the updated model parameters to the edge computing node for deployment.

[0086] Exemplarily, encrypted transmission can be doubly protected by the AES-256 algorithm combined with the TLS 1.3 protocol, and the encryption key is dynamically generated and regularly updated by the central management system of the virtual power plant. After the model parameters are serialized into binary files, they are transmitted in blocks to the edge nodes in the target area through the HTTPS protocol, and each data packet is attached with a digital signature to verify the integrity. The edge node deployment process includes: 1) verifying that the hash value of the received parameter file is consistent with the central system record; 2) stopping the currently running model instance; 3) loading new parameters into memory and initializing the model inference engine; 4) switching traffic to the new model and monitoring the initial operating status. For example, after receiving the encrypted parameter file, an edge node restores it to a weight matrix file through a preset decryption key, and completes the hot update within 200 milliseconds to ensure the continuity of the scheduling service.

[0087] As an implementation manner, the method provided in the embodiment of the present invention further includes a pre-training process of the multi-objective optimization model, which may specifically include the following steps: Step S10: Obtain the operating parameter set of each air-conditioning terminal in the target area within the historical period and the corresponding virtual power plant historical control instruction set, the operating parameter set including historical energy consumption fluctuation characteristics, environmental parameter change curves and user setting change records.

[0088] The historical period refers to a pre-set time span, such as the complete scheduling record of the past 12 months, and the length of the time window is dynamically configured according to the model training requirements. The historical energy consumption fluctuation characteristics are collected through the power metering module of each air-conditioning terminal in the target area, including the fluctuation data of the compressor operating frequency, air supply wind speed, and the difference sequence between the set temperature and the actual temperature on the historical time axis. For example, the energy consumption curve of an air-conditioning terminal in the summer historical period shows a power peak from 14:00 to 16:00 every day, and the peak power reaches 3.5kW. The environmental parameter change curve is recorded by the temperature and humidity sensors deployed in the area where the air-conditioning terminal is located, including indoor and outdoor temperature gradients, humidity change rate, and light intensity time series data. For example, the average room temperature in a certain area in the summer during the historical period is 28℃±2℃, 20℃±1.5℃ in the winter, and the humidity fluctuation range is 45%-65%. The user setting change record is extracted through the log database of the mobile terminal application, specifically including the user's modification records of the temperature sensitivity coefficient, energy-saving priority flag, and allowed adjustment time period during the historical period. For example, user A adjusted the temperature sensitivity coefficient three times in the historical period, gradually increasing it from the initial value of 0.7 to 0.9. The operating parameter set and the historical control instruction set are stored in a distributed database through timestamp association to ensure data integrity. For example, each control instruction record contains the issuance timestamp, load distribution target value and execution status code, which forms a mapping relationship with the air conditioning terminal operating parameters in the same time period.

[0089] Step S20: parsing the historical control instruction set for instruction features, extracting the load distribution target value, electricity price constraint conditions and power grid emergency event markers contained therein, and generating a historical control feature set.

[0090] Exemplarily, the instruction feature parsing process can be to extract structured features from unstructured historical control instruction texts using regular expressions and semantic analysis algorithms. The load distribution target value represents the power adjustment amount set by the virtual power plant for each region during the historical period. For example, a certain instruction requires the target region to reduce the total load by 80kW during 14:00-15:00 on July 15, 2023. The electricity price constraint condition is parsed as the gradient change parameter of the time-of-use electricity price curve, including the peak period, valley period and real-time floating rate of electricity price. For example, in a historical instruction, the electricity price rises linearly from 0.5 yuan / kWh to 0.65 yuan / kWh during the period of 14:00-15:00, and the gradient change rate is 0.03 yuan / kWh·minute. The power grid emergency event marking is realized by parsing the status code in the instruction. For example, the status code "E1" indicates a frequency over-limit event, and "E2" indicates a voltage sag event. After the historical control feature set is generated, it is stored as a three-dimensional tensor structure, whose dimensions include timestamp, region identifier and feature vector. For example, the feature vector of a region at timestamp 1650000000 can be expressed as [80kW, 0.5-0.65 yuan / kWh, E1], which correspond to the load distribution target value, electricity price constraint condition and power grid emergency event mark respectively.

[0091] Step S30: Align the operating parameter set with the historical control feature set in time dimension, intercept parameter segments within the same time interval through a sliding window mechanism, and generate a historical joint input feature set; wherein, the time dimension alignment includes interpolating the air-conditioning terminal data with different sampling frequencies and the control instruction timestamps to ensure that the input features of each time segment contain a complete correspondence between the operating parameters and the control instructions.

[0092] Exemplarily, the time dimension alignment algorithm can use bilinear interpolation to unify data with different sampling frequencies. For example, the air conditioning terminal data is sampled at intervals of 30 seconds, and the control instructions are issued at intervals of 5 minutes. During alignment, the control instruction timestamp is expanded to a 30-second granularity, and the instruction features of the missing time points are filled by neighboring interpolation. The sliding window mechanism intercepts parameter fragments of continuous time intervals with a fixed duration, for example, setting the window length to 1 hour and the step length to 30 minutes, and extracting overlapping time series fragments from historical data. The generated historical joint input feature set is a five-dimensional tensor structure, which includes a time window index, an air conditioning terminal identifier, operating parameters, control features, and environmental parameters. For example, the fragment with a window index of W001 contains the compressor frequency sequence [45Hz, 46Hz, 47Hz] of a terminal during 14:00-15:00 on July 15, 2023, the corresponding load distribution target value of 80kW, and the room temperature data of 28℃-29℃ during the same period. Interpolation processing ensures that each 30-second timestamp contains a complete mapping relationship between operating parameters and control instructions, eliminating the problem of feature discontinuity caused by data asynchrony.

[0093] Step S40: Input the historical joint input feature set into the initialized deep reinforcement learning network, generate a historical priority sequence through the policy network, and evaluate the comprehensive benefit score of the historical priority sequence through the value network.

[0094] Exemplarily, the initialized deep reinforcement learning network consists of a policy network and a value network, and its input layer dimension matches the historical joint input feature set. The policy network uses a gated recurrent unit (GRU) structure to process time series features and outputs the scheduling priority score and parameter adjustment suggestions for each air-conditioning terminal in the historical time window. For example, the output for a terminal in time window W001 is a priority score of 0.85 (0-1 interval, the higher the value, the higher the priority), and a recommended temperature setting value adjustment of -1.5℃. The value network contains three parallel fully connected layers, which calculate the grid benefit sub-score, user satisfaction sub-score, and energy efficiency sub-score respectively. The grid benefit sub-score is generated by comparing the mean square error between the actual load curve and the command target value. For example, the error between the actual load reduction of 75kW and the target of 80kW is 5kW, corresponding to a score of 0.78. The user satisfaction sub-score is calculated based on the negative correlation between the historical feedback score and the temperature setting value execution deviation. For example, when a terminal executes a deviation of 1℃, the user score drops by 20%, and the score is 0.65. The energy efficiency sub-score is calculated by the dynamic time warping distance between the actual energy consumption sequence and the theoretical adjustment value. The smaller the distance, the higher the score. The comprehensive benefit score is the weighted sum of the three sub-scores. The initial weight value is set to [0.4, 0.3, 0.3] and is dynamically optimized during the training process.

[0095] Step S50: Collect the grid load change data, user feedback data and equipment energy consumption data after the actual execution of the historical priority sequence, compare the data with the comprehensive benefit score, and generate a multi-objective training loss value.

[0096] Exemplarily, the grid load change data is exported from the energy management system (EMS) of the virtual power plant, including the actual power curve of the target area after historical scheduling. For example, after the priority sequence is executed, the total load of the area is actually reduced by 78kW from 14:00 to 15:00, which is 2kW away from the target value of 80kW. User feedback data is collected through the scoring interface of the mobile application, including the five-point satisfaction score and the sentiment analysis results of the text evaluation. For example, user A scored 4 points after scheduling, and the sentiment analysis result is "basically satisfied". The equipment energy consumption data is transmitted back through the metering module of the air-conditioning terminal, recording parameters such as real-time power and compressor start and stop times after the priority sequence is executed. The difference comparison process uses a multimodal data fusion algorithm to compare the actual data with the comprehensive benefit score predicted by the value network item by item. For example, the predicted value of the grid benefit sub-score is 0.78, and the actual load curve matching degree is calculated to be 0.75, so the difference value is 0.03. The predicted value of the user satisfaction sub-score is 0.65, and the actual score average is 3.8 / 5.0 (converted to 0.76), with a difference value of -0.11. The energy efficiency sub-score is predicted to be 0.70, and the actual energy consumption fluctuation is calculated to be 0.68 with a deviation from the target, with a difference of 0.02. The above difference values ​​constitute the initial component of the multi-objective training loss value.

[0097] Step S60: Adopting the gradient back propagation algorithm to iteratively update the policy network parameters of the deep reinforcement learning network until the multi-objective training loss value reaches the convergence threshold, thereby obtaining a pre-trained multi-objective optimization model.

[0098] For example, the gradient back-propagation process uses the Adam optimizer, and the learning rate is initially set to 1e-4, decaying by 10% every 1000 training steps. The update direction of the policy network parameters is determined by the total gradient of the multi-objective training loss value, and the total gradient is calculated by the chain rule as the sum of the partial derivatives of each loss component with respect to the network weight. For example, the gradient partial derivative of the power grid stability loss component is ∂L grid / ∂W, the partial derivative of the user satisfaction loss component is ∂L user / ∂W, and the weighted sum of the two is used to update the strategy network weights. The convergence threshold is set according to the loss curve on the validation set. For example, the training is terminated when the loss decreases by less than 1e-5 for 10 consecutive training cycles. After pre-training, the multi-objective optimization model can generate a priority sequence that takes into account grid security, user comfort, and energy efficiency. For example, it can achieve a comprehensive performance of 92% load adjustment target matching, 4.2 / 5.0 user satisfaction score, and 15% energy efficiency improvement on the test set.

[0099] As an implementation method, the process of generating the multi-objective training loss value may specifically include: Step S501: Decompose the comprehensive benefit score output by the value network into a grid benefit sub-score, a user satisfaction sub-score and an energy efficiency sub-score, each sub-score corresponding to the three dimensions of the virtual power plant control target; among them, the grid benefit sub-score is generated by analyzing the matching degree between the historical load curve and the control instruction, the user satisfaction sub-score is generated by statistically analyzing the positive correlation between the historical feedback score and the priority sequence execution effect, and the energy efficiency sub-score is generated by calculating the deviation attenuation rate between the historical energy consumption fluctuation and the target adjustment value.

[0100] Exemplarily, the calculation of the grid benefit sub-score can use the dynamic time warping (DTW) algorithm to measure the morphological similarity between the actual load curve and the command target value. For example, the target load curve for a certain period of time is a linear decrease of 80kW, and the actual curve shows a step-by-step decrease. The DTW distance is 12kW·second, and the score after normalization is 0.82. The user satisfaction sub-score can be quantified by the Pearson correlation coefficient, for example, the priority sequence execution deviation (such as temperature setting error) is negatively correlated with the user score (r=-0.85), and the score is mapped to (1+r) / 2=0.575. The energy efficiency sub-score can use the exponential decay function to calculate the deviation decay rate, the formula is score=exp(-λ·ΔE), where ΔE is the root mean square error between the actual energy consumption and the target value, and λ is the decay coefficient. For example, when ΔE=5kW, λ=0.1, the score is exp(-0.5)=0.606.

[0101] Step S502: extracting the matching degree between the actual load curve and the expected load curve in each time segment from the power grid load change data, performing a difference comparison between the matching degree and the power grid benefit sub-score, and generating a power grid stability loss component.

[0102] For example, the matching degree calculation can be performed by the piecewise integration method, for example, the load curve is divided into several intervals according to the time segment, and the area overlap between the actual value and the expected value in each interval is calculated. For example, in a 5-minute time segment, the expected load decreases by 15kW, and the actual decrease is 12kW. The overlap area is 72% of the expected curve, and the matching degree is 0.72. The grid stability loss component is calculated by the mean square error (MSE), and the formula is L grid =Σ(matching degree - grid benefit sub-score) 2 / N. For example, if the matching degree sequence in a window is [0.72, 0.85, 0.91], and the grid benefit sub-score prediction is [0.78, 0.82, 0.88], then the loss component is ((0.72-0.78) 2 +(0.85-0.82) 2 +(0.91-0.88)2 ) / 3=0.0013.

[0103] Step S503: extracting the actual satisfaction score of each user from the user feedback data, performing probability density difference analysis on the score distribution curve and the user satisfaction sub-score, and generating a user satisfaction loss component.

[0104] For example, the probability density difference analysis uses the Kullback-Leibler (KL) divergence to measure the degree of deviation between the predicted score distribution and the actual score. For example, the predicted distribution of the user satisfaction sub-score is a normal distribution with a mean of 0.7 and a variance of 0.1, and the actual score distribution is a mean of 0.76 and a variance of 0.05. The KL divergence is calculated to be 0.15. The user satisfaction loss component is defined as the weighted sum of the KL divergence, and the formula is L user =Σ w i ·D KL (P i ||Q i ), where P i is the actual distribution, Q i is the predicted distribution, w i is the user weight coefficient. For example, the weight of high-value users is set to 1.5, and that of ordinary users is 1.0.

[0105] Step S504: extracting the actual energy consumption sequence from the equipment energy consumption data, calculating the trend alignment between the sequence fluctuation characteristics and the energy efficiency sub-score, and generating an energy consumption deviation loss component.

[0106] Exemplarily, the trend alignment can be measured by cosine similarity to measure the trend consistency between the actual energy consumption sequence and the target adjustment value. For example, the actual energy consumption sequence shows a V-shaped trend of first decreasing and then increasing within the time window, while the target value is monotonically decreasing, and the cosine similarity is -0.3. The energy consumption deviation loss component is defined as (1-similarity) / 2, so that the loss is 0 when it is completely consistent and 1 when it is completely opposite. For example, the loss component corresponding to the similarity of -0.3 is (1-(-0.3)) / 2=0.65.

[0107] Step S505: Input the grid stability loss component, the user satisfaction loss component and the energy consumption deviation loss component into the dynamic weight allocator, and adjust the weight ratio of each component based on the grid operation mode mark of the historical control instruction set, wherein the weight ratio is synchronized with the evaluation weights of the three dimensions of the value network and updated; the parameter update of the dynamic weight allocator shares the back propagation link with the gradient descent process of the policy network.

[0108] Exemplarily, the dynamic weight allocator can be implemented using an attention mechanism, where the query vector is the current grid operation mode tag (such as "normal", "emergency", "economy"), and the key-value pair is the historical weight of each loss component. For example, in the "emergency" mode, the grid stability loss weight is increased from 0.4 to 0.7, the user satisfaction weight is reduced to 0.2, and the energy efficiency weight is 0.1. The weight coefficients are normalized by the Softmax function to ensure that the sum is 1. During the parameter update process, the gradient ∂L of the dynamic weight allocator total / ∂w i Gradient ∂L with the policy network total / ∂θ participate in back propagation together to achieve end-to-end joint optimization. Step S506: Input the weighted loss component into the nonlinear fusion unit, calculate the total gradient value through the back propagation path of the value network, and generate a multi-objective training loss value consistent with the optimization direction of the comprehensive benefit score.

[0109] The nonlinear fusion unit is composed of two layers of fully connected neural networks, and the activation function is LeakyReLU. The weighted loss components are transformed nonlinearly by the fusion unit, and the total loss value L is output. total =α·L grid +β·L user +γ·L energy , where α, β, and γ are dynamic weight coefficients. The total gradient value is calculated by the chain rule, and the formula is ∂L total / ∂θ=∂L total / ∂L grid ∂L grid / ∂θ+∂L total / ∂L user ∂L user / ∂θ+∂L total / ∂L energy ∂L energy / ∂θ. The back-propagation path shares the parameter space of the value network and the policy network, ensuring that the gradient update direction is consistent with the Pareto optimal solution of the comprehensive benefit score. For example, when the grid stability loss weight is high, the gradient update prioritizes the performance of the policy network in load tracking.

[0110] In some optional implementations, after the incremental parameter is updated in step S500, the method provided by the embodiment of the present invention may further include the following derivative steps: Step S600: Predictive priority is pre-generated for the air-conditioning terminals in the target area that have not executed the scheduling strategy based on the updated multi-objective optimization model, wherein the pre-generation process includes: extracting the actual execution delay characteristics and energy consumption adjustment efficiency characteristics of each air-conditioning terminal in the feedback parameter set, generating a device response capability portrait, wherein the device response capability portrait includes the historical average instruction parsing time, the success probability of compressor state switching and the execution accuracy of the temperature setting value, dynamically correlating and matching the device response capability portrait with the real-time electricity price fluctuation characteristics in the current joint input characteristics, and predicting the potential priority score of the non-executed equipment in the next scheduling cycle.

[0111] The predictive priority pre-generation mechanism uses the incrementally updated multi-objective optimization model to perform forward-looking priority evaluation on air-conditioning terminals that have not received control instructions in the current scheduling cycle. Air-conditioning terminals that have not executed the scheduling strategy refer to the set of terminals that have not been marked as immediate response devices due to insufficient urgency or user preference restrictions. Their identification information is obtained through real-time screening of the device status database of the edge controller. The pre-generation process simulates the scenario by constructing a joint input feature of the future time window, combining historical response capabilities with real-time grid status data, and generating potential priority scores to guide the strategy preparation for the next scheduling cycle. For example, a terminal is not scheduled because its current priority score of 0.72 is lower than the emergency threshold of 0.8, but it is predicted that its score may increase to 0.85 when the electricity price rises to 0.7 yuan / kWh in the next cycle, thus being included in the preparation strategy queue.

[0112] The actual execution delay feature is calculated by counting the average time difference from the command issuance to the execution completion in the historical scheduling records. For example, a terminal has an average delay of 28 seconds in the past 50 schedulings, with a standard deviation of 3 seconds. The energy consumption adjustment efficiency feature is defined as the ratio of the actual energy consumption change rate of the terminal to the target adjustment rate. For example, the target requires a power reduction of 1kW within 5 minutes, and the actual completion time is 5 minutes and 20 seconds, and the efficiency is (1kW / 320 seconds) = 0.003125kW / s. The equipment response capability portrait is stored in a structured data table, with each terminal corresponding to a row of records, including: 1) the historical average command parsing time, which is calculated by the average difference between the command receiving timestamp and the execution confirmation timestamp, for example, the average parsing time of terminal A is 2.5 seconds; 2) the probability of successful switching of the compressor state, which is the percentage of the number of times the compressor is started and stopped according to the command in the historical scheduling, for example, the number of successful times is 48 / 50=96%; 3) the execution accuracy of the temperature set value, which is defined as the root mean square error (RMSE) between the actual temperature and the target temperature, for example, the RMSE of terminal B is 0.4℃. The portrait data is dynamically updated through the time series database of the edge computing node to ensure that it reflects the latest status of the device. The dynamic association matching algorithm adopts a fusion architecture of the attention mechanism and the time series prediction model. The real-time electricity price fluctuation characteristics are divided into gradient segments according to the time granularity of the next scheduling cycle. For example, the electricity price will increase from 0.6 yuan / kWh to 0.65 yuan / kWh in the next 30 minutes. The historical average instruction parsing time in the equipment responsiveness portrait is convolved with the gradient change rate of the electricity price to generate a time sensitivity weight. For example, the parsing time of terminal A is 28 seconds, which corresponds to a time sensitivity coefficient of 0.7 in the electricity price increase stage. It is predicted that the priority score increase during the peak electricity price period is 0.7×0.05 yuan / kWh=0.035. The potential priority score is calculated by the preparatory strategy branch network of the multi-objective optimization model. The input features include the equipment responsiveness portrait vector, real-time electricity price parameters, and grid frequency prediction data. For example, the input feature vector of terminal B is [28 seconds, 96%, 0.4°C, 0.6-0.65 yuan / kWh, 49.98Hz], and the output potential priority score is 0.82.

[0113] Step S700: pre-adjust the start and stop time configuration of the unexecuted equipment according to the potential priority score, and generate a candidate scheduling strategy set including a pre-buffer time slot; the pre-buffer time slot is dynamically calculated and generated according to the equipment response delay characteristics and the power grid frequency fluctuation trend.

[0114] Exemplarily, the pre-adjustment algorithm can use the potential priority score as the sorting basis to virtually allocate the start and stop time windows for the unexecuted devices. For example, the potential score of terminal C is 0.78, which corresponds to the 120-150 second time window after the start of the next cycle. The length of the pre-buffer time slot is dynamically determined by the product of the device response delay characteristic and the grid frequency fluctuation rate, and the formula is: pre-buffer time slot = historical average delay × (1 + frequency fluctuation rate). Assuming that the historical delay of terminal C is 30 seconds and the current grid frequency fluctuation rate is 0.2% (relative to 50Hz), the pre-buffer time slot = 30 × (1 + 0.002) = 30.06 seconds, rounded to 30 seconds. The candidate scheduling strategy set includes multiple virtual time configuration schemes. For example, strategy 1 sets the start time of terminal C to 120 seconds after the start of the next cycle, and strategy 2 sets it to 135 seconds, each with a pre-buffer time slot of 30 seconds and 35 seconds, forming an optional timing arrangement set.

[0115] Step S800: Input the candidate scheduling strategy set into the sandbox simulation environment of the edge controller, and collect the power grid load forecast curve and user satisfaction forecast value after the simulation execution.

[0116] The sandbox simulation environment constructs a virtual grid topology model and digital twin of the air-conditioning terminal in the target area, and runs a high-precision simulation after injecting candidate scheduling strategies. The grid load forecast curve is generated by the improved ARIMA model. For example, after the simulation execution of strategy 1, the total load drops linearly from 500kW to 470kW in the next cycle, and strategy 2 drops to 465kW. The user satisfaction prediction value is calculated based on the regression model of historical feedback data and temperature setting value execution accuracy. For example, the prediction score of strategy 1 is 4.1 / 5.0, and that of strategy 2 is 4.3 / 5.0. The simulation results are stored as a multi-dimensional evaluation matrix, which includes the load change rate, user score mean, and equipment energy efficiency index of each strategy. For example, the matrix items of strategy 1 are [470kW, 4.1, 0.85], and those of strategy 2 are [465kW, 4.3, 0.82].

[0117] Step S900: Sort candidate strategies according to the deviation between the prediction curve and the prediction value, select the strategy with the smallest deviation as the reserve strategy for the next scheduling cycle and connect it with the current target scheduling strategy in time sequence.

[0118] For example, the deviation calculation can use the Mahalanobis distance to measure the multi-dimensional difference between the prediction result and the ideal target value. For example, the ideal target value is set as the Pareto frontier point with the maximum load reduction, the highest user score and the best energy efficiency. For example, the load deviation of strategy 1 = |470-450| = 20kW, the user score deviation = |4.1-5.0| = 0.9, and the corresponding deviation of strategy 2 is 15kW and 0.7, and the Mahalanobis distance is √(20 2 +0.9 2)=20.02 and√(15 2 +0.7 2 )=15.02, strategy 2 is selected as the reserve strategy due to its small deviation. The timing connection process is implemented through the scheduling timing coordination module of the edge controller, which seamlessly embeds the time window of the reserve strategy into the current execution queue. For example, if the end time of the current strategy is 1620003600, the start time of the reserve strategy is set to 1620003600+buffer interval 30 seconds=1620003630.

[0119] As an implementation manner, in step S900, the process of performing timing connection may include the following steps: Step S901: parse the remaining execution time period of each air-conditioning terminal in the current target scheduling strategy, and extract the device identifiers that overlap and conflict with the pre-adjusted start and stop time in the preparation strategy.

[0120] The remaining execution time period is dynamically calculated through the timestamp queue of the current strategy. For example, the planned downtime period of terminal D is 1620003000-1620003600, and the start time of terminal E in the preparation strategy is set to 1620003300, then the overlapping conflict period is 1620003300-1620003600. The conflicting device identifier is identified by the time interval intersection detection algorithm. For example, terminals D and E have overlapping instructions during the period of 1620003300-1620003600, and are marked as a conflicting device pair (D, E). The conflict information is recorded as a triple data structure, including the conflicting device identifier, the start and end timestamps of the overlapping period, and the conflict type (such as start-stop conflict, power superposition conflict).

[0121] Step S902: Perform bidirectional sliding window matching on the execution time period of the conflicting device, and search for an idle time slot that allows the insertion of the preliminary strategy operation within the remaining execution time period; wherein the bidirectional sliding window matching includes forward searching for the idle time period after the end time of the current strategy and backward searching for the compressible time period before the start time of the current strategy.

[0122] Exemplarily, the bidirectional sliding window algorithm takes the conflict period as the center and expands the search for available time slots in both the forward and backward directions. The forward search range extends from the end time of the current strategy to the start time of the next cycle, for example, looking for continuous idle windows between 1620003600-1620007200; the backward search compresses the execution period of non-critical equipment in the current strategy, for example, advancing the shutdown time of terminal F from 1620003200 to 1620003100 to free up a 100-second idle window. The availability check of the idle time slot is determined by the power grid flow calculation and the equipment response delay constraint. For example, a 100-second time slot must meet the following requirements: 1) the load rate of the distribution line after inserting the reserve strategy is ≤90%; 2) the equipment pre-buffer time slot is ≥ the historical average delay × 1.2. If both conditions are met, it is marked as an available time slot.

[0123] Step S903: adjusting the pre-buffer time slot in the preparation strategy according to the length of the idle time slot, and generating a time-aligned connection strategy instruction set.

[0124] Exemplarily, the pre-buffer time slot adjustment algorithm is dynamically scaled based on the margin of the idle time slot. For example, if the original pre-buffer time slot of the preparation strategy is 30 seconds, and the length of the available idle time slot is 85 seconds, the pre-buffer time slot is adjusted to 30 seconds (maintaining the lower limit) to retain a 55-second safety margin; if the idle time slot is only 25 seconds, the pre-buffer time slot is compressed to 25 seconds and the device response rate check is triggered. The timestamps of the connection strategy instruction set are rearranged. For example, the startup time of terminal E is adjusted from 1620003630 to 1620003550, the pre-buffer time slot is reduced from 30 seconds to 20 seconds, and a new instruction frame [ID: E, Action: Start, Time: 1620003550, Buffer: 20] is generated.

[0125] Step S904: Real-time monitoring of grid load fluctuation data, and triggering a dynamic strategy switching operation when it is detected that the execution period in the connection strategy instruction set matches the grid frequency adjustment demand; the triggering conditions of the dynamic strategy switching operation include the urgency of the grid frequency regulation instruction exceeding a preset threshold or the user comfort feedback value being lower than the safety critical line.

[0126] Exemplarily, the grid frequency adjustment demand is obtained through the frequency deviation signal of the virtual power plant. For example, when the frequency drops to 49.8Hz, the urgency parameter of the frequency regulation command is triggered to rise to 0.9 (threshold 0.8). The user comfort safety threshold is set to a satisfaction score of 3.5 / 5.0. If the real-time feedback mean drops to 3.4, the switch is triggered. The dynamic switching operation is executed by the instant decision module of the edge controller. For example, when the frequency deviation is detected to exceed 0.2Hz, the current non-critical strategy is forcibly interrupted and the instructions with a priority score ≥ 0.85 in the connection strategy are immediately executed.

[0127] Step S905: embed the switched connection strategy instruction set into the execution queue of the current target scheduling strategy according to the timestamp, and update the time constraint parameters in the input features of the multi-objective optimization model.

[0128] Exemplarily, the timestamp embedding process adopts a priority preemptive scheduling algorithm, for example, the shutdown instruction (priority 0.88) of terminal G in the connection strategy is inserted before the instruction with priority 0.75 in the current queue. After the queue reconstruction is executed, the time constraint parameters are synchronously updated in the input features of the multi-objective optimization model, for example, the start time of the next scheduling cycle is adjusted from the original 1620007200 to 1620003630, and the time window length is shortened from 3600 seconds to 3570 seconds. The timing axis of the model input features is re-interpolated and aligned to ensure that the subsequent optimization process generates strategies based on the latest scheduling timing constraints.

[0129] According to another aspect of the present invention, there is also provided an air conditioning scheduling device based on a virtual power plant response, see Figure 3 , the device 900 comprises: The parameter acquisition module 910 is used to obtain the real-time operation parameter set of each air-conditioning terminal in the target area within a preset time window, wherein the real-time operation parameter set includes current energy consumption characteristics, ambient temperature and humidity characteristics, and user preference setting characteristics, wherein the user preference setting characteristics are used to describe the user's priority configuration of the air-conditioning operation mode; A feature alignment module 920 is used to perform feature alignment processing on the real-time operating parameter set and the dynamic control instructions issued by the virtual power plant to generate a joint input feature, wherein the dynamic control instructions include the load distribution strategy of the virtual power plant to the target area and the real-time electricity price fluctuation characteristics; A priority generation module 930 is used to input the joint input features into a pre-trained multi-objective optimization model, and to perform hierarchical weight allocation on the joint input features through a feature fusion layer in the multi-objective optimization model to generate an air conditioning control priority sequence, wherein the priority sequence is used to describe the scheduling order and parameter adjustment range of each air conditioning terminal under a dynamic control instruction; A strategy generation module 940 is used to generate a target scheduling strategy based on the priority sequence, wherein the target scheduling strategy includes the start and stop time configuration of each air-conditioning terminal, the temperature setting threshold adjustment range and the energy consumption constraint condition; The incremental update module 950 is used to send the target scheduling strategy to each air-conditioning terminal in the target area for execution, collect the feedback parameter set after execution in real time, and use the feedback parameter set to perform incremental parameter update on the multi-objective optimization model.

[0130] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiment of the present invention can be used to execute the method described in the above method embodiment. For technical details not disclosed in the device embodiment of the present invention, please refer to the description of the method embodiment of the present invention for understanding.

[0131] It should be noted that, in the embodiment of the present invention, if the above-mentioned air conditioning scheduling method based on virtual power plant response is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), disk or optical disk, etc., various media that can store program codes. In this way, the embodiment of the present invention is not limited to any specific hardware, software or firmware, or any combination of hardware, software and firmware.

[0132] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document is not limited here.

[0133] Although the embodiments or examples of the present invention have been described with reference to the accompanying drawings, it should be understood that the above-mentioned methods, systems and devices are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but only by the authorized claims and their equivalents. Various elements in the embodiments or examples may be omitted or replaced by their equivalents. In addition, the steps may be performed in an order different from that described in the present invention. Further, the various elements in the embodiments or examples may be combined in various ways. It is important that with the evolution of technology, many of the elements described herein may be replaced by equivalent elements that appear after the present invention.

Claims

1. An air conditioning scheduling method based on virtual power plant response, characterized in that: The method comprises: Acquire a real-time operating parameter set of each air-conditioning terminal in the target area within a preset time window, wherein the real-time operating parameter set includes current energy consumption characteristics, ambient temperature and humidity characteristics, and user preference setting characteristics, wherein the user preference setting characteristics are used to describe the user's priority configuration of the air-conditioning operating mode; Performing feature alignment processing on the real-time operating parameter set and the dynamic control instructions issued by the virtual power plant to generate a joint input feature, wherein the dynamic control instructions include the load distribution strategy of the virtual power plant to the target area and the real-time electricity price fluctuation characteristics; The combined input features are input into a pre-trained multi-objective optimization model, and the combined input features are weighted in layers through a feature fusion layer in the multi-objective optimization model to generate an air conditioning control priority sequence, wherein the priority sequence is used to describe the scheduling order and parameter adjustment range of each air conditioning terminal under a dynamic control instruction; Generate a target scheduling strategy based on the priority sequence, the target scheduling strategy including the start and stop time configuration of each air-conditioning terminal, the temperature setting threshold adjustment range and the energy consumption constraint condition; The target scheduling strategy is sent to each air-conditioning terminal in the target area for execution, and a feedback parameter set after execution is collected in real time, and the feedback parameter set is used to perform incremental parameter update on the multi-objective optimization model.

2. The method according to claim 1, characterized in that The step of obtaining a real-time operating parameter set of each air-conditioning terminal in the target area within a preset time window includes: The operation status data of each air-conditioning terminal in the target area in multiple continuous time segments are collected through the Internet of Things gateway, and the operation status data includes the compressor operating frequency, air supply speed, and the difference sequence between the set temperature and the actual temperature; Perform outlier elimination and timestamp alignment processing on the operating status data to generate a standardized operating parameter set; Extracting a preference setting instruction submitted by a user through a mobile terminal, wherein the preference setting instruction includes a temperature sensitivity coefficient, an energy-saving priority flag, and an allowable control time period; Performing feature splicing on the standardized operating parameter set and the preference setting instruction to generate the real-time operating parameter set including time-space correlation; The timestamp alignment process includes uniformly mapping sensor data of different sampling frequencies to the same time reference axis, and the outlier removal includes identifying compressor current values ​​or temperature and humidity readings that exceed preset physical thresholds.

3. The method according to claim 1, characterized in that The step of performing feature alignment processing on the real-time operating parameter set and the dynamic control instructions issued by the virtual power plant to generate a joint input feature includes: Parsing the load distribution strategy in the dynamic control instruction, extracting the load adjustment target value and the corresponding time constraint parameter related to the air-conditioning terminal, wherein the time constraint parameter includes the start time of the load adjustment; Extracting a current energy consumption feature subset matching the time constraint parameter from the real-time operation parameter set, the current energy consumption feature subset comprising energy consumption fluctuation curves and device response rate data of each air-conditioning terminal within a preset time window before the start time; Based on the time constraint parameter, the load adjustment target value is split into sub-load adjustment values ​​corresponding to a plurality of continuous time segments, and each time segment is mapped to a corresponding fluctuation period in the energy consumption fluctuation curve; The sub-load adjustment value of each time segment is superimposed with the equipment response rate data in the corresponding fluctuation period to generate a load adjustment vector sequence containing time sensitivity; Performing dimension expansion and fusion on the load adjustment vector sequence and the real-time electricity price parameter in the dynamic control instruction, wherein the dimension expansion and fusion includes splitting the real-time electricity price parameter into gradient change segments according to time segments and performing element-level weighted splicing with the positions with the same time index in the vector sequence to generate the joint input feature; Among them, the equipment response rate data is obtained by calculating the product of the average time from the air-conditioning terminal receiving the instruction to completing the parameter adjustment in the historical scheduling record and the compressor operating frequency under the current operating state, and the weight coefficient of the weighted splicing is determined by the grid stability priority under the current operating mode of the virtual power plant.

4. The method according to claim 1, characterized in that: The pre-training process of the multi-objective optimization model includes: Obtaining the operating parameter set of each air-conditioning terminal in the target area within the historical period and the corresponding virtual power plant historical control instruction set, wherein the operating parameter set includes historical energy consumption fluctuation characteristics, environmental parameter change curves, and user setting change records; Performing instruction feature analysis on the historical control instruction set, extracting the load distribution target value, electricity price constraint conditions and power grid emergency event markers contained therein, and generating a historical control feature set; Aligning the operating parameter set with the historical control feature set in time dimension, intercepting parameter segments within the same time interval through a sliding window mechanism, and generating a historical joint input feature set; Inputting the historical joint input feature set into an initialized deep reinforcement learning network, generating a historical priority sequence through a policy network, and evaluating a comprehensive benefit score of the historical priority sequence through a value network; Collecting grid load change data, user feedback data, and equipment energy consumption data after the actual execution of the historical priority sequence, performing a difference comparison between the data and the comprehensive benefit score, and generating a multi-objective training loss value; Iteratively updating the policy network parameters of the deep reinforcement learning network using a gradient back-propagation algorithm until the multi-objective training loss value reaches a convergence threshold, thereby obtaining the pre-trained multi-objective optimization model; The time dimension alignment includes interpolating the air-conditioning terminal data with different sampling frequencies and the control instruction timestamps to ensure that the input features of each time segment contain the complete corresponding relationship between the operating parameters and the control instructions.

5. The method according to claim 4, characterized in that The generation process of the multi-objective training loss value includes: Decomposing the comprehensive benefit score output by the value network into a grid benefit sub-score, a user satisfaction sub-score and an energy efficiency sub-score, each sub-score corresponding to the three dimensions of the virtual power plant control target; Extracting the matching degree between the actual load curve and the expected load curve in each time segment from the power grid load change data, performing a difference comparison between the matching degree and the power grid benefit sub-score, and generating a power grid stability loss component; Extracting the actual satisfaction score of each user from the user feedback data, performing probability density difference analysis on the score distribution curve and the user satisfaction sub-score, and generating a user satisfaction loss component; Extracting an actual energy consumption sequence from the equipment energy consumption data, calculating trend alignment between the sequence fluctuation characteristics and the energy efficiency sub-score, and generating an energy consumption deviation loss component; Input the grid stability loss component, the user satisfaction loss component and the energy consumption deviation loss component into a dynamic weight allocator, and adjust the weight ratio of each component based on the grid operation mode mark of the historical control instruction set, wherein the weight ratio is synchronized with the evaluation weight of the value network on the three dimensions; Inputting the weighted loss components into the nonlinear fusion unit, calculating the total gradient value through the back propagation path of the value network, and generating a multi-objective training loss value consistent with the optimization direction of the comprehensive benefit score; Among them, the grid benefit sub-score is generated by analyzing the matching degree between the historical load curve and the control instructions, the user satisfaction sub-score is generated by statistically analyzing the positive correlation between the historical feedback score and the execution effect of the priority sequence, and the energy consumption efficiency sub-score is generated by calculating the deviation attenuation rate between the historical energy consumption fluctuation and the target adjustment value. The parameter update of the dynamic weight allocator shares the back-propagation link with the gradient descent process of the policy network.

6. The method according to claim 1, characterized in that The priority sequence generation process includes: sorting the joint input features by the attention mechanism layer in the multi-objective optimization model to generate an initial priority score for each air-conditioning terminal; Inputting the initial priority score into a gated loop unit, performing timing correction in combination with the device response delay characteristics in the historical scheduling record, and generating a corrected dynamic priority sequence; According to the load adjustment urgency parameter issued by the virtual power plant, the dynamic priority sequence is threshold segmented, and the air-conditioning terminal with a priority score higher than the emergency threshold is marked as an immediate response device; The parameter adjustment amplitude of the immediate response device is gradient-limited so that the temperature setting value does not change beyond the maximum offset allowed in the user preference setting feature.

7. The method according to claim 1, characterized in that The execution process of the target scheduling strategy includes: Parsing the air-conditioning terminal identifier marked as an immediate response device in the priority sequence, extracting the corresponding start and stop time configuration and temperature setting threshold adjustment range, and generating a device control instruction set; The device control instruction set is sorted by timestamp and sent to the edge controller of the target area, and the edge controller adjusts the start and stop time of the compressor according to the buffer interval parameter in the instruction to generate a drive signal set including the execution timing; Real-time collection of actual operating parameters of each air-conditioning terminal after executing the drive signal set, including compressor state switching delay, temperature setting value execution deviation and instantaneous energy consumption fluctuation range, to generate a dynamic adjustment parameter set; Comparing the dynamically adjusted parameter set with the parameter adjustment range in the priority sequence, identifying abnormal device identifiers whose execution deviations exceed an allowable threshold, and generating a policy adjustment trigger event; According to the policy adjustment trigger event, the multi-objective optimization model is called to locally recalculate the current priority sequence, generate an updated target scheduling strategy and re-issue it for execution, and at the same time, the feedback parameter set associated with the abnormal device identifier is marked as a high-priority sample for incremental update.

8. The method according to claim 7, characterized in that The process of handling the policy adjustment triggering event includes: parsing the execution deviation type corresponding to the abnormal device identifier, the execution deviation type comprising at least one of excessive compressor response delay, temperature setting value lock failure, or excessive instantaneous energy consumption; According to the deviation type, a preset adjustment rule library is matched to extract a candidate adjustment strategy associated with the current virtual power plant operation mode, wherein the candidate adjustment strategy includes a device parameter compensation coefficient, an execution timing redundancy, or a priority weight correction value; Perform constraint checking on the candidate adjustment strategies and actual operating parameters in the dynamic adjustment parameter set, eliminate invalid strategies that violate user preference setting characteristics or grid stability boundaries, and generate a feasible strategy set; Inputting the feasible strategy set into the feature fusion layer of the multi-objective optimization model, performing a preliminary evaluation of the strategy benefits in combination with the real-time electricity price fluctuation feature in the current joint input feature, and generating an updated priority sequence candidate set; The edge controller performs sandbox simulation on the priority sequence candidate set, collects power grid load forecast data and user satisfaction forecast scores after the simulation, and selects the candidate strategy with the best comprehensive evaluation as the new target scheduling strategy for execution.

9. The method according to claim 1, characterized in that: The incremental parameter updating process includes: Compare the feedback parameter set with the expected execution effect data to generate a model prediction error distribution graph; Extracting abnormal data points that exceed the tolerance interval in the error distribution graph, and tracing them back to the generation logic path of the corresponding scheduling strategy; Performing local gradient descent optimization on the neuron connection weights associated with the abnormal data points in the multi-objective optimization model; Keep the historical parameter versions that are stable on the preset validation set, and automatically roll back to the most recent stable version when the model performance after incremental update drops beyond a threshold; The updated model parameters are encrypted and transmitted to the edge computing nodes for deployment.

10. An air conditioning dispatching device based on virtual power plant response, characterized in that: include: A parameter acquisition module is used to acquire a real-time operating parameter set of each air-conditioning terminal in a target area within a preset time window, wherein the real-time operating parameter set includes current energy consumption characteristics, ambient temperature and humidity characteristics, and user preference setting characteristics, wherein the user preference setting characteristics are used to describe the user's priority configuration of the air-conditioning operation mode; A feature alignment module, used to perform feature alignment processing on the real-time operating parameter set and the dynamic control instructions issued by the virtual power plant to generate a joint input feature, wherein the dynamic control instructions include the load distribution strategy of the virtual power plant to the target area and the real-time electricity price fluctuation characteristics; A priority generation module, used for inputting the joint input features into a pre-trained multi-objective optimization model, performing hierarchical weight allocation on the joint input features through a feature fusion layer in the multi-objective optimization model, and generating an air conditioning control priority sequence, wherein the priority sequence is used to describe the scheduling order and parameter adjustment range of each air conditioning terminal under a dynamic control instruction; A strategy generation module, used to generate a target scheduling strategy based on the priority sequence, wherein the target scheduling strategy includes the start and stop time configuration of each air-conditioning terminal, the temperature setting threshold adjustment range and the energy consumption constraint condition; The incremental update module is used to send the target scheduling strategy to each air-conditioning terminal in the target area for execution, collect the feedback parameter set after execution in real time, and use the feedback parameter set to perform incremental parameter update on the multi-objective optimization model.

Images