An alternating current power supply system

Abstract

The application relates to the technical field of power supply control, and particularly discloses an alternating current power supply system, which comprises an evolution module, a construction module and a prediction module. The evolution module acquires historical operation data of the alternating current power supply system in each natural day within a historical specified period, constructs a long-term trend prediction model, and determines a load prediction sequence and a plurality of possible power evolution sequences in a specified time period. The construction module acquires thermal response data and constructs a multi-module coupled thermal impedance field model. The prediction module acquires a device to be exited from the alternating current power supply system, determines a candidate exit time set and a predicted junction temperature curve of each power change device within an exit time window. The exit module constructs a power thermal margin field, determines a feasible exit time set and an exit time. The alternating current power supply system device exit time dynamic optimization driven by long-term load trend prediction and multi-module coupled thermal impedance field is realized, and the operation reliability and thermal safety redundancy of the alternating current power supply system in the device exit process are improved.

Description

Technical Field

[0001] This invention relates to the field of power control technology, and in particular to an AC power supply system. Background Technology

[0002] Optimizing the timing of equipment shutdown in AC power systems is a key issue in ensuring their long-term reliable operation. Early methods relied on periodic maintenance plans or manual judgment, determining the shutdown time based on current load and temperature conditions, lacking consideration of future power variation trends. With the development of thermal management technology, lumped parameter models based on thermal resistance networks are widely used for power device junction temperature estimation, and multi-chip thermal coupling matrix models have enabled quantitative characterization of thermal interactions between chips within a module. At the equipment management level, technologies such as temperature monitoring derating protection, lifespan prediction-based shutdown triggering, and module rotation control are gradually being applied, achieving real-time monitoring and basic protection of the device's thermal status. However, existing technologies have the following shortcomings: their protection mechanisms are mostly reactive, meaning they only trigger shutdown when the temperature reaches a threshold or the module ages; they also lack systematic modeling capabilities for the lateral thermal coupling formed between multiple independent power change devices through shared heat sinks, air duct heat transfer, and thermal radiation, and they lack decision-making methods to systematically search and optimize feasible shutdown timing based on comprehensive thermal margins over a continuous time axis, making it difficult to balance the safety and economy of long-term operation when selecting shutdown timing; they also fail to fully exploit the value of industrial big data accumulated throughout the entire lifecycle of AC power systems, resulting in a lack of data-driven, forward-looking operation and maintenance decision-making capabilities.

[0003] Therefore, the present invention proposes an AC power supply system. Summary of the Invention

[0004] This invention provides an AC power supply system that acquires historical operating data for each natural day within a specified historical period, constructs a long-term trend prediction model, and determines the load prediction sequence and multiple possible power evolution sequences for the AC power supply system within the specified prediction time period. It also acquires thermal response data of the AC power supply system and constructs a multi-module coupled thermal impedance field model; acquires devices to be decommissioned from the AC power supply system and determines the predicted junction temperature curve for each power-changing device within the decommissioning time window; and constructs a power supply thermal margin field to determine the set of feasible decommissioning opportunities and the execution decommissioning time. By deeply coupling the thermal state evolution prediction of the power supply system with the device decommissioning decision, it achieves dynamic optimization of the AC power supply system's device decommissioning timing based on the joint drive of long-term load trend prediction and multi-module coupled thermal impedance field, thereby improving the operational reliability and thermal safety redundancy of the AC power supply system during the device decommissioning process.

[0005] This invention provides an AC power supply system, comprising: Evolution Module: Acquires historical operating data of the AC power system for each natural day within a specified historical period, constructs a long-term trend prediction model, and determines the load prediction sequence and multiple possible power evolution sequences of the AC power system within the specified prediction time period. Module construction: Acquire thermal response data of AC power system, establish three-node thermal resistance network model structure for each power change device, determine thermal parameter data of each power change device and mutual thermal resistance parameter data between power change devices, and construct multi-module coupled thermal impedance field model. Prediction module: acquires the devices to be decommissioned in the AC power system, determines the candidate decommissioning timing set based on the current time and the predicted specified time period, and determines the predicted junction temperature curve of each power change device within the decommissioning time window based on the candidate decommissioning timing set and the multi-module coupled thermal impedance field model. Exit Module: Based on all possible power evolution sequences and the predicted junction temperature curves of all power change devices within the exit time window, construct the power supply thermal margin field, determine the set of feasible exit opportunities, and execute the exit time.

[0006] Preferably, an AC power supply system, an evolution module, includes: Historical Operation Data Unit: Acquires historical operation data of the AC power system for each natural day within a specified historical period. The historical operation data includes historical daily power time series, cyclic operation patterns, historical load rate time series of each power change device, and historical exit data. Cyclic operation patterns include start-up and shutdown time series of multiple power change devices, rotation cycle data, and priority data. Historical exit data includes multiple exit events of power change devices, the time of occurrence of each exit, the reason for exit, the thermal state before exit, the number of remaining devices after exit, load redistribution data, overheat alarm tags, and load reduction protection tags. Long-term trend prediction model unit: Based on historical operating data of all natural days within a specified historical period, a long-term trend prediction model is constructed. The long-term trend prediction model includes at least a feature extraction module and a time series decomposition extrapolation module. The feature extraction module extracts features from the cyclical operating patterns, historical load rate time series of each power change device, and historical exit data in the historical operating data of all natural days within the specified historical period. The time series decomposition extrapolation module performs seasonal decomposition, trend extraction, and residual modeling on the historical daily power series, and extrapolates the decomposition results to the specified prediction time period.

[0007] Preferably, an AC power supply system, an evolution module, further includes: Load forecast sequence unit: Based on a long-term trend forecast model, determine the load forecast sequence of the AC power system within a specified forecast time period; Short-term dispatch data unit: acquires short-term dispatch data from multiple information sources of the AC power system, and labels each information source with information tags, including definite information and uncertain information; First power increment sequence unit: Transforms short-term mobilization data of each information source with information labeled as definite information into a power increment sequence for predicting a specified time period; The second power increment sequence unit: Monte Carlo sampling is used to construct a deviation probability distribution model based on the short-term mobilization data of each information source with uncertain information label, and generate multiple possible power increment sequences for each information source with uncertain information label within the specified prediction time period; Power Evolution Sequence Unit: Based on the load prediction sequence of the AC power system, the power increment sequence of all information sources with definite information labels, and multiple possible power increment sequences of all information sources with uncertain information labels, multiple possible power evolution sequences of the AC power system are generated within the specified prediction time period.

[0008] Preferably, an AC power supply system, comprising a building module, including: Thermal response data unit: Acquires thermal response data of AC power system, including historical power disturbance sequence, junction temperature time series, case temperature time series, radiator temperature time series, ambient temperature time series, and inter-device thermal coupling data, including heat dissipation coupling data, air duct coupling data and radiation coupling data of each radiator. Model structure unit: Establish a three-node thermal resistance network model structure for each power change device. The three nodes include junction nodes, shell nodes, and heat sink nodes. The nodes are connected by thermal resistance and thermal capacity to form a heat flow conduction path.

[0009] Preferably, an AC power supply system, the building module further includes: Coupling Path Unit: Based on the heat dissipation coupling data in the inter-device thermal coupling data of the thermal response data, a heat dissipation conduction coupling path is generated for each radiator; based on the air duct coupling data in the inter-device thermal coupling data of the thermal response data, an airflow thermal coupling path is generated; and simultaneously, based on the radiation coupling data in the inter-device thermal coupling data of the thermal response data, a thermal radiation coupling path is generated. Thermal parameter data unit: A step power perturbation sequence is sequentially injected into each power change device. Except for the power change device, each power change device maintains a constant power. The junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of the power change device are collected. The step power perturbation sequence is used as input, and the junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of the power change device are used as output. The least squares algorithm is used to determine the thermal parameter data of each power change device. The thermal parameter data includes junction-to-case thermal resistance, case-to-radiator thermal resistance, radiator-to-ambient thermal resistance, junction heat capacity, case heat capacity, and radiator heat capacity. Mutual thermal resistance parameter data unit: A pseudo-random binary power perturbation sequence is sequentially injected into each power change device. Except for the power change device, each power change device maintains a constant power and does not perform additional active temperature intervention. The junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of each power change device are collected. The pseudo-random binary power perturbation sequence is cross-correlated with the junction temperature response curve, case temperature response curve, and radiator temperature response curve of each power change device. Transfer function fitting is performed to determine the mutual thermal resistance parameter vector from the power change device to each power change device. The mutual thermal resistance parameter data is determined. The mutual thermal resistance parameter vector includes junction-to-junction mutual thermal resistance parameters, junction-to-case mutual thermal resistance parameters, and junction-to-radiator mutual thermal resistance parameters. Multi-module coupled thermal impedance field model unit: The thermal parameter data of each power change device is embedded into the corresponding three-node thermal resistance network model structure, and the mutual thermal resistance parameter data is connected to the nodes of the corresponding power change device as a lateral coupling branch to construct a multi-module coupled thermal impedance field model.

[0010] Preferably, an AC power supply system includes a prediction module, comprising: Candidate Exit Timing Set Unit: Based on the current time and the predicted termination time of a specified time period, determine the exit time window, and based on the exit time window and the set time step, determine the candidate exit timing set; Acquisition Unit: Acquires the devices to be decommissioned in the AC power system, and acquires the current power share of each power-changing device in the AC power system at the current moment; Update power share unit: Determine the updated power share of the device to be withdrawn as 0. Based on the power share of the device to be withdrawn in the AC power system and all power change devices, determine the normalized updated power share of all power change devices after the withdrawal of the device to be withdrawn. Power loss allocation value unit: For each candidate time in the candidate exit timing set, based on the updated power share undertaken by each power change device and the expected power value at the predicted time corresponding to the candidate time, the power loss allocation value of each power change device is input into the power-loss mapping model of the power change device to determine the power loss allocation value of each power change device at each candidate time. Power loss allocation sequence unit: Based on the power loss allocation value of each power change device at all candidate times, determine the power loss allocation sequence of each power change device; Predicted junction temperature curve unit: Input the power loss allocation sequence of all power change devices into the multi-module coupled thermal impedance field model to determine the predicted junction temperature curve of each power change device within the exit time window.

[0011] Preferably, an AC power supply system includes an exit module, comprising: Selection Unit: Extract all power values ​​at each prediction time within the specified prediction time period from all power evolution sequences and sort them to determine the power evolution sequence at each prediction time within the specified prediction time period. Select the median of the power evolution sequence as the expected power value at the prediction time. Select the upper quantile of the set confidence level as the upper limit envelope of the power at the prediction time. Select the lower quantile of the set confidence level as the lower limit envelope of the power at the prediction time. Maximum power increment unit: For each candidate time in the candidate exit timing set, calculate the difference between the upper limit envelope of the power and the expected power value of the predicted time corresponding to the candidate time, and determine the maximum power increment of the candidate time; Maximum power sub-increment unit: Based on the normalized updated power share of all power change devices after the exit of the device to be exited, and the maximum power increment at each candidate moment in the candidate exit timing set, the maximum power sub-increment of each power change device is determined. Equivalent junction temperature increment unit: Based on mutual thermal resistance parameter data and the maximum power sub-increment and thermal parameter data of each power change device, calculate the equivalent junction temperature increment of each power change device; Thermal margin unit: Obtain the safe junction temperature threshold for each power change device, and determine the thermal margin of each power change device at each candidate moment within the exit time window based on the equivalent junction temperature increment, safe junction temperature threshold, and predicted junction temperature curve within the exit time window for each power change device. Power supply thermal margin field unit: Based on the thermal margin of all candidate moments within the exit time window of all power change devices with non-zero updated power share, a power supply thermal margin field is constructed. Exit tag unit: The thermal margin of all power change devices at each candidate time in the power thermal margin field is judged. If the thermal margin of all power change devices at the candidate time is greater than 0, and the minimum thermal margin of all power change devices at the candidate time is greater than the set safety threshold, the exit tag of the candidate time is determined to be feasible. Feasible exit timing set unit: Based on all exit tags within the exit time window being feasible candidate times, determine the set of feasible exit timings.

[0012] Preferably, an AC power supply system, including an exit module, further includes: Triggering Unit: If the feasible exit timing set is empty, trigger an early warning and request external load reduction or activation of a backup cooling source; Execution Exit Time Unit: If the set of feasible exit opportunities is not empty, the execution exit time is determined based on the set of feasible exit opportunities, the thermal margin field, and the lower limit envelope of the power at each predicted time within the specified time period. The device to be exited in the AC power system will exit at the execution exit time.

[0013] The beneficial effects of this invention compared to existing technologies are as follows: It acquires historical operating data of the AC power system for each natural day within a specified historical period, constructs a long-term trend prediction model, and determines the load prediction sequence and multiple possible power evolution sequences of the AC power system within the specified prediction time period; it acquires thermal response data of the AC power system and constructs a multi-module coupled thermal impedance field model; it acquires devices to be decommissioned from the AC power system and determines the predicted junction temperature curve of each power-changing device within the decommissioning time window; it constructs a power thermal margin field and determines the set of feasible decommissioning opportunities and the execution decommissioning time. By deeply coupling the prediction of the thermal state evolution of the power system with the equipment decommissioning decision, it achieves dynamic optimization of the AC power system equipment decommissioning timing based on the joint drive of long-term load trend prediction and multi-module coupled thermal impedance field, improving the operational reliability and thermal safety redundancy of the AC power system during equipment decommissioning; it fully releases the data value of industrial big data in power thermal management and equipment operation and maintenance, promoting the system's upgrade from passive thermal protection to data-driven proactive predictive operation and maintenance.

[0014] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of an AC power supply system according to an embodiment of the present invention. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. Example 1:

[0018] This invention provides an AC power supply system, referenced Figure 1 ,include: Evolution Module: Acquires historical operating data of the AC power system for each natural day within a specified historical period, constructs a long-term trend prediction model, and determines the load prediction sequence and multiple possible power evolution sequences of the AC power system within the specified prediction time period. Module construction: Acquire thermal response data of AC power system, establish three-node thermal resistance network model structure for each power change device, determine thermal parameter data of each power change device and mutual thermal resistance parameter data between power change devices, and construct multi-module coupled thermal impedance field model. Prediction module: acquires the devices to be decommissioned in the AC power system, determines the candidate decommissioning timing set based on the current time and the predicted specified time period, and determines the predicted junction temperature curve of each power change device within the decommissioning time window based on the candidate decommissioning timing set and the multi-module coupled thermal impedance field model. Exit Module: Based on all possible power evolution sequences and the predicted junction temperature curves of all power change devices within the exit time window, construct the power supply thermal margin field, determine the set of feasible exit opportunities, and execute the exit time.

[0019] In this embodiment, the evolution module first acquires historical operating data of the AC power system for each natural day within a specified historical period. This historical operating data includes a historical daily power time series reflecting the daily variation pattern of the system's total power, a cyclical operation pattern (records of the alternating start-up and shutdown cycles of multiple power-changing devices), a historical load rate time series for each power-changing device, and historical shutdown data including overheat alarm tags and load reduction protection tags. Based on this data, a long-term trend prediction model is constructed. This model extracts structured features from the cyclical operation pattern, load rate sequence, and historical shutdown data using a feature extraction module. A time-series decomposition and extrapolation module performs seasonal decomposition, trend extraction, and residual modeling on the historical daily power series, and extrapolates the decomposition results to the specified prediction time period. The long-term trend prediction model outputs a load prediction sequence, i.e., the estimated total system load at various future moments, and multiple possible power evolution sequences, i.e., multiple power trajectories under different operating scenarios to cover future operational uncertainties.

[0020] In this embodiment, the construction module acquires the thermal response data of the AC power system. This thermal response data is complete information describing the dynamic temperature response of each device when the power changes, including time series of junction temperature, case temperature, radiator temperature, and ambient temperature for each power-changing device, as well as thermal coupling data between devices. Based on the thermal response data, a three-node thermal resistance network model structure is established for each power-changing device. The three nodes include a junction node (power chip heat source), a case node (package casing), and a radiator node (heat dissipation device). The nodes are connected by thermal resistance and thermal capacitance to form heat flow conduction paths. Thermal parameter data for each device is identified through step power perturbation, including junction-to-case thermal resistance, case-to-radiator thermal resistance, radiator-to-ambient thermal resistance, and thermal capacitance at each node. Mutual thermal resistance parameters between devices are determined by combining pseudo-random binary power perturbation with cross-correlation and transfer function fitting, i.e., the degree of influence of one device's heating on the temperature rise of each node in another device. The thermal parameters are embedded into the three-node thermal resistance network of each device, and the mutual thermal resistance parameters are connected as lateral coupling branches between corresponding nodes to construct a multi-module coupled thermal impedance field model.

[0021] In this embodiment, the prediction module acquires the devices to be decommissioned in the AC power system, i.e., power-changing devices planned for maintenance or repair, and determines a set of candidate decommissioning opportunities based on the current time and a predicted specified time period. For each candidate decommissioning time, the power share of the device to be decommissioned is reset to zero, and its load is proportionally redistributed to the remaining devices to obtain the updated power share of each device. Combining the expected power values ​​at each candidate time, the electrical power is converted into thermal loss power through a power-loss mapping model, forming a power loss allocation sequence for each device. The power loss allocation sequences of all devices are input into a multi-module coupled thermal impedance field model, and the model outputs the predicted junction temperature curve for each power-changing device within the decommissioning time window.

[0022] In this embodiment, the exit module extracts statistical features for each predicted moment from all power evolution sequences: the median as the expected power value, the upper quantile of the set confidence level as the upper power envelope, and the lower quantile as the lower power envelope. The difference between the upper power envelope and the expected value is calculated as the maximum power increment, and allocated to each device according to the updated power share of each device to obtain the maximum power sub-increment. Based on the mutual thermal resistance parameter, the maximum power sub-increment is mapped to the equivalent junction temperature increment, that is, the additional rise in junction temperature of each device under the most unfavorable power scenario. The safe junction temperature threshold of each device is obtained, and the thermal margin of each candidate moment is calculated by combining the predicted junction temperature curve and the equivalent junction temperature increment. After constructing the power thermal margin field, candidate moments in which the thermal margin of all operating devices is greater than zero and the minimum thermal margin is greater than the set safe threshold are screened to form a set of feasible exit opportunities. If the set is empty, an early warning is triggered and an external load reduction or activation of a backup cold source is requested; if it is not empty, the exit moment is determined by combining the thermal margin field and the lower power envelope. The purpose of this step is to ensure that exiting the operation under any possible power fluctuation conditions will not lead to system thermal safety failure, thus achieving safe and reliable active device exit.

[0023] The beneficial effects of the above technologies are as follows: First, they acquire historical operating data of the AC power system for each natural day within a specified historical period, construct a long-term trend prediction model, and determine the load prediction sequence and multiple possible power evolution sequences of the AC power system within the specified prediction time period. Second, they acquire thermal response data of the AC power system and construct a multi-module coupled thermal impedance field model. Third, they acquire the devices to be decommissioned from the AC power system and determine the predicted junction temperature curve of each power-changing device within the decommissioning time window. Fourth, they construct a power supply thermal margin field to determine the set of feasible decommissioning opportunities and the execution decommissioning time. By deeply coupling the prediction of the thermal state evolution of the power system with the equipment decommissioning decision, they achieve dynamic optimization of the AC power system equipment decommissioning timing based on the joint drive of long-term load trend prediction and multi-module coupled thermal impedance field, thereby improving the operational reliability and thermal safety redundancy of the AC power system during the equipment decommissioning process. Example 2:

[0024] Based on Example 1, an AC power supply system, an evolution module, includes: Historical Operation Data Unit: Acquires historical operation data of the AC power system for each natural day within a specified historical period. The historical operation data includes historical daily power time series, cyclic operation patterns, historical load rate time series of each power change device, and historical exit data. Cyclic operation patterns include start-up and shutdown time series of multiple power change devices, rotation cycle data, and priority data. Historical exit data includes multiple exit events of power change devices, the time of occurrence of each exit, the reason for exit, the thermal state before exit, the number of remaining devices after exit, load redistribution data, overheat alarm tags, and load reduction protection tags. Long-term trend prediction model unit: Based on historical operating data of all natural days within a specified historical period, a long-term trend prediction model is constructed. The long-term trend prediction model includes at least a feature extraction module and a time series decomposition extrapolation module. The feature extraction module extracts features from the cyclical operating patterns, historical load rate time series of each power change device, and historical exit data in the historical operating data of all natural days within the specified historical period. The time series decomposition extrapolation module performs seasonal decomposition, trend extraction, and residual modeling on the historical daily power series, and extrapolates the decomposition results to the specified prediction time period.

[0025] In this embodiment, historical operating data of the AC power system for each natural day within a specified historical period is acquired. The specified historical period refers to a relatively long past time, such as the past year or three years, to accumulate sufficient statistical patterns. Each natural day refers to a time granularity measured in days; the load of the power system often exhibits obvious daily and weekly periodicity. The historical operating data contains four types of information. The first type is the historical daily power time series, which is a continuous record of the total system power changing over time each day, reflecting the daily variation pattern of the system load. The second type is the cyclical operation pattern, which refers to the operating mode in which multiple power changing devices in the AC power system take turns being put into operation and taken out of operation according to a preset strategy, used to balance the operating time and aging degree of each device. The cyclical operation pattern specifically includes the start-up and stop time series of multiple power changing devices, i.e., the records of the start and stop times of each device in each historical day; rotation cycle data, i.e., the time interval between the device's first and second starts; and priority data, i.e., the priority level of each device in load allocation. The third type is the historical load rate time series of each power changing device, i.e., a record of the proportion of load borne by each device in historical operation changing over time. The fourth category is historical exit data, which is a detailed record of all past exit events. This includes the device name and time of each exit event, the reason for exit (e.g., maintenance or fault alarm), the thermal state of the device before exit (e.g., junction or case temperature), the number of devices remaining in the system after exit, the load redistribution data among the remaining devices, and whether the system triggered overheat alarms or load derating protection tags after exit. Overheat alarm tags indicate whether any device temperature exceeded the alarm threshold after the exit event, while load derating protection tags indicate whether the system actively reduced the total load due to insufficient thermal margin.

[0026] In this embodiment, the power conversion device refers to a modular unit in an AC power system that performs power conversion functions, such as a rectifier module, an inverter module, or a DC / DC converter module.

[0027] In this embodiment, the system extracts information such as the time of occurrence of daily peak power, daily average power level, and peak-to-valley ratio from the historical load rate time series. It also extracts historical load rate distribution, peak load rate and duration, and load rate mutation rate from the historical load rate data of each power-changing device. Furthermore, it extracts the rotation cycle length, rotation time distribution, and cumulative operating time differences among devices from the cyclical operation pattern. Finally, it extracts the correlation pattern between exit events and power surges, the total system load level distribution before exit, and the over-temperature occurrence rate of remaining devices after exit from the historical exit data of each power-changing device.

[0028] In this embodiment, seasonal decomposition aggregates power values ​​for all natural days by hourly position, extracts typical intraday fluctuation patterns, and forms a 24-hour seasonal baseline curve. Trend extraction uses sliding window averaging or low-pass filtering to eliminate seasonal fluctuations and random noise, obtaining a trend component reflecting the evolution of the long-term baseline. Residual modeling performs statistical characteristic analysis on the remaining sequence after removing seasonality and trends, establishing an autoregressive model or Gaussian process model to quantify the probability distribution of random fluctuations. Extrapolation adds the seasonal component to a table based on the hourly position of the prediction time, and extrapolates the trend component to the prediction window based on the time difference; the sum of the two forms a long-term system-level load power prediction baseline. The statistical parameters of the residual components are retained for subsequent uncertainty envelope generation but do not directly participate in the numerical calculation of the long-term prediction baseline.

[0029] The beneficial effects of the above technologies are: acquiring historical operating data of AC power systems for each natural day within a specified historical period, constructing a long-term trend prediction model, and realizing long-term trend prediction of AC power system load through multi-layer heterogeneous historical data fusion and time-series decomposition extrapolation. Example 3:

[0030] Based on Embodiment 1, an AC power supply system, including an evolution module, further includes: Load forecast sequence unit: Based on a long-term trend forecast model, determine the load forecast sequence of the AC power system within a specified forecast time period; Short-term dispatch data unit: acquires short-term dispatch data from multiple information sources of the AC power system, and labels each information source with information tags, including definite information and uncertain information; First power increment sequence unit: Transforms short-term mobilization data of each information source with information labeled as definite information into a power increment sequence for predicting a specified time period; The second power increment sequence unit: Monte Carlo sampling is used to construct a deviation probability distribution model based on the short-term mobilization data of each information source with uncertain information label, and generate multiple possible power increment sequences for each information source with uncertain information label within the specified prediction time period; Power Evolution Sequence Unit: Based on the load prediction sequence of the AC power system, the power increment sequence of all information sources with definite information labels, and multiple possible power increment sequences of all information sources with uncertain information labels, multiple possible power evolution sequences of the AC power system are generated within the specified prediction time period.

[0031] In this embodiment, based on the established long-term trend forecasting model, the load forecast sequence of the AC power system within a specified forecasting time period is determined. The specified forecasting time period is a pre-set future time period, such as the next 24 hours or the next 48 hours. The load forecast sequence is the predicted value of the total system load at each point in time within this time period, derived from the output of the time-series decomposition extrapolation module in the long-term trend forecasting model. This sequence reflects the baseline load change under the condition of no additional short-term disturbances.

[0032] In this embodiment, short-term dispatch data from multiple information sources in the AC power system are acquired, and each information source is labeled with an information tag. An information source refers to a source that can provide information affecting future system power changes, such as load adjustment notifications from the power grid dispatch center, maintenance plans from the equipment management system, charging and discharging plans from the energy storage system, and meteorological data from the weather forecasting system. Short-term dispatch data refers to planned or predictive data provided by these information sources regarding the near future. Each information source is labeled with an information tag, which includes both definite and indefinite information.

[0033] In this embodiment, the information source for the information tag to determine the information can be the grid-side demand response signal. After the grid-side demand response signal is officially issued by the grid dispatching agency, the required power adjustment range, duration, and response rate parameters are mandatory and deterministic.

[0034] In this embodiment, the information source with uncertain information labels can be the future task queue and expected resource requirements, the data center IT load scheduling plan, or the ambient temperature forecast curve.

[0035] In this embodiment, for each information source with a definite information tag, its short-term mobilization data is directly converted into a power increment sequence for a predicted specified time period. The power increment sequence refers to the sequence of changes over time in the predicted increase or decrease in the total system power relative to the baseline load caused by that information source. Since the information is definite, the conversion process does not need to introduce randomness; the power change at each time point can be calculated directly based on the information content. For example, a device that is determined to start operating at 10:00 AM will generate a power increment occurring at a definite time during its startup process, which is directly mapped to the power increase at the corresponding moment within the predicted time period.

[0036] In this embodiment, Monte Carlo sampling is employed to probabilistically model each information source labeled as uncertain. Specifically, based on historical deviation data or theoretical distribution assumptions for each uncertain information source, a deviation probability distribution model is constructed. This model describes the probabilistic distribution characteristics of the difference between the predicted and actual values ​​of the information source. Then, multiple random samples are taken from this model to generate multiple possible power increment sequences for each uncertain information source. Each sampling result represents a possible scenario for that information source within the future prediction time period. The core idea of ​​Monte Carlo sampling is to approximate the true probability distribution through extensive random sampling, thereby covering the various consequences that uncertain information may bring.

[0037] In this embodiment, using the load prediction sequence as a benchmark, all power increment sequences from determined information sources and any one possible power increment sequence from all uncertain information sources are superimposed on the benchmark to obtain multiple complete power evolution sequences. Since each uncertain information source generates multiple possible increment sequences through Monte Carlo sampling, these increment sequences are combined fully or randomly to ultimately obtain a set of multiple power evolution sequences covering different future scenarios.

[0038] The beneficial effects of the above technologies are: determining the load prediction sequence and multiple possible power evolution sequences of the AC power system within a specified time period, enabling the generation of multiple power evolution scenarios based on direct conversion of deterministic information and Monte Carlo sampling collaboration of uncertain information. Example 4:

[0039] Based on Embodiment 1, an AC power supply system, comprising a module, includes: Thermal response data unit: Acquires thermal response data of AC power system, including historical power disturbance sequence, junction temperature time series, case temperature time series, radiator temperature time series, ambient temperature time series, and inter-device thermal coupling data, including heat dissipation coupling data, air duct coupling data and radiation coupling data of each radiator. Model structure unit: Establish a three-node thermal resistance network model structure for each power change device. The three nodes include junction nodes, shell nodes, and heat sink nodes. The nodes are connected by thermal resistance and thermal capacity to form a heat flow conduction path.

[0040] In this embodiment, thermal response data of the AC power system is acquired. Thermal response data is a complete dataset describing the dynamic temperature response of the system under power changes, and it includes the following specific components: Historical power disturbance sequence refers to a record of power changes during the system's historical operation, including the time points and magnitudes of power increases, decreases, or abrupt changes. Junction temperature time series for each power change device is data on the change in chip junction temperature over time, directly measured by internal temperature sensors such as thermocouples or thermistors. Junction temperature is the most critical temperature indicator for power semiconductor devices, directly determining the device's safety margin and lifespan. Case temperature time series is a record of the change in surface temperature of the device's package over time, measured by temperature sensors attached to the package surface. Case temperature is the intermediate stage for heat transfer from the junction region to the external environment. Heat sink temperature time series is a record of the change in temperature of the heat sink substrate or fins on which the device is installed over time. The heat sink is the key channel for the final dissipation of heat to the environment. Ambient temperature time series is a record of the change in air temperature within the system's cabinet or server room over time. Ambient temperature serves as the final heat sink temperature for heat dissipation. The thermal coupling data between devices includes data from three coupling methods: heat dissipation coupling data, air duct coupling data, and radiation coupling data.

[0041] In this embodiment, the junction temperature time series of each power converter is obtained through temperature-sensitive electrical parameters integrated inside the power semiconductor device or infrared thermometry, with a sampling frequency no less than one-tenth of the switching frequency. The case temperature time series is obtained through thermocouples or thermistors mounted on the surface of the power converter package, with a sampling frequency consistent with the junction temperature sampling. The heat sink temperature time series is obtained through temperature sensors arranged at key locations on the heat sink substrate and fins, recording the spatial distribution of the heat sink temperature field. The ambient temperature time series records the cabinet intake air temperature and the computer room ambient temperature, serving as the external boundary conditions for the thermal resistance network.

[0042] In this embodiment, the three-node thermal resistance network model is a lumped-parameter thermal model that simplifies the main thermal nodes inside and outside the device into three nodes. The first is the junction node, representing the junction region of the power semiconductor chip, which is the source of heat generation. The second is the shell node, representing the inner or outer surface of the device's packaging shell, which is the intermediate node for heat transfer from the chip to the outside. The third is the heat sink node, representing the substrate or fins of the heat sink, which is the node where heat is finally dissipated into the ambient air. The three nodes are connected by thermal resistance and thermal capacitance to form a heat flow conduction path: the junction node and the shell node are connected by the thermal resistance and thermal capacitance from the chip to the shell, reflecting the process of heat from the chip being conducted to the shell through the packaging material; the shell node and the heat sink node are connected by the interface thermal resistance and thermal capacitance from the shell to the heat sink, reflecting the process of heat being conducted to the heat sink through the contact interface or thermally conductive medium; the heat sink node is usually also connected to the ambient temperature node through a convection thermal resistance, reflecting the process of heat being dissipated from the heat sink into the surrounding air.

[0043] The beneficial effects of the above technologies are: acquiring thermal response data of AC power systems, establishing a three-node thermal resistance network model structure for each power change device, enabling the thermal state prediction of multi-module power systems to take into account both the internal thermal paths and inter-device thermal interactions in the spatial dimension, and providing complete physical model support for the prediction of junction temperature curves. Example 5:

[0044] Based on Embodiment 4, an AC power supply system, including a construction module, further includes: Coupling Path Unit: Based on the heat dissipation coupling data in the inter-device thermal coupling data of the thermal response data, a heat dissipation conduction coupling path is generated for each radiator; based on the air duct coupling data in the inter-device thermal coupling data of the thermal response data, an airflow thermal coupling path is generated; and simultaneously, based on the radiation coupling data in the inter-device thermal coupling data of the thermal response data, a thermal radiation coupling path is generated. Thermal parameter data unit: A step power perturbation sequence is sequentially injected into each power change device. Except for the power change device, each power change device maintains a constant power. The junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of the power change device are collected. The step power perturbation sequence is used as input, and the junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of the power change device are used as output. The least squares algorithm is used to determine the thermal parameter data of each power change device. The thermal parameter data includes junction-to-case thermal resistance, case-to-radiator thermal resistance, radiator-to-ambient thermal resistance, junction heat capacity, case heat capacity, and radiator heat capacity. Mutual thermal resistance parameter data unit: A pseudo-random binary power perturbation sequence is sequentially injected into each power change device. Except for the power change device, each power change device maintains a constant power and does not perform additional active temperature intervention. The junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of each power change device are collected. The pseudo-random binary power perturbation sequence is cross-correlated with the junction temperature response curve, case temperature response curve, and radiator temperature response curve of each power change device. Transfer function fitting is performed to determine the mutual thermal resistance parameter vector from the power change device to each power change device. The mutual thermal resistance parameter data is determined. The mutual thermal resistance parameter vector includes junction-to-junction mutual thermal resistance parameters, junction-to-case mutual thermal resistance parameters, and junction-to-radiator mutual thermal resistance parameters. Multi-module coupled thermal impedance field model unit: The thermal parameter data of each power change device is embedded into the corresponding three-node thermal resistance network model structure, and the mutual thermal resistance parameter data is connected to the nodes of the corresponding power change device as a lateral coupling branch to construct a multi-module coupled thermal impedance field model.

[0045] In this embodiment, the heat dissipation coupling data includes the device set for each heat sink, the first local coordinates of each power changing device in the device set, the thermal conductivity of the heat sink substrate material, and the substrate thickness. Based on the first local coordinates of the power changing devices on the heat sink, the heat conduction distance between each pair of power changing devices is calculated, i.e., the Euclidean distance between the center points of the power changing devices. Based on the thermal conductivity of the heat sink substrate material, the substrate thickness, and the heat conduction distance, the thermal resistance value of the heat sink nodes between the power changing devices is calculated. The thermal resistance value is directly proportional to the heat conduction distance and inversely proportional to the thermal conductivity of the substrate material and the substrate thickness. The calculated thermal resistance between the power changing devices is connected between the heat sink nodes of the corresponding power changing devices to construct a shared heat sink heat conduction coupling path.

[0046] In this embodiment, the duct coupling data includes the duct inlet location, duct cross-sectional area, airflow direction, second local coordinates of each power converter in the duct, and airflow velocity. Based on the airflow direction and velocity distribution, the upstream and downstream order of each power converter in the duct is determined. The power converter through which the airflow first passes is upstream, and the power converters through which it passes subsequently are downstream. An airflow temperature node is established based on the rack inlet air temperature, with the initial value being the rack inlet air temperature. The heat dissipation power of the upstream power converter is the heat flux value on the path from the heat sink of the power converter to the ambient thermal resistance. This heat flux value contributes to the airflow temperature rise, which is directly proportional to the heat dissipation power of the upstream power converter and inversely proportional to the airflow velocity and duct cross-sectional area. The ambient temperature end on the path from the heat sink of the downstream power converter to the ambient thermal resistance is connected to the airflow temperature node after being heated by the upstream power converter. The effective ambient temperature of each downstream power converter is the inlet air temperature plus the cumulative airflow temperature rise contributed by all upstream power converters, thus constructing an airflow thermal coupling path in the forced air cooling duct.

[0047] In this embodiment, the radiation coupling data includes the shielding relationship of the internal structure of the cabinet to thermal radiation, the emissivity of the surface of the encapsulation shell of each power change device and the surface of the heat sink. Based on the physical layout coordinates of the power change devices on the heat sink and the shielding relationship of the internal structure of the cabinet to thermal radiation, the viewing angle factor between the surfaces of each power change device is calculated. The viewing angle factor depends on the relative position, area, and shielding relationship between the surfaces. Based on the emissivity of the surface of the encapsulation shell of each power change device and the surface of the heat sink, a radiation thermal resistance network is established. The radiation thermal resistance is related to the viewing angle factor, surface emissivity, and surface area. Radiative heat flux is injected as an additional heat source term into the heat sink node or shell node of the corresponding power change device to construct the internal thermal radiation coupling path of the cabinet.

[0048] In this embodiment, a step power perturbation sequence is sequentially injected into each power-changing device, while other devices maintain a constant power. Simultaneously with the application of the step perturbation, the junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of the perturbed device are acquired—that is, the dynamic process of temperature rising over time and gradually approaching a new equilibrium. Using the applied step power perturbation sequence as the input excitation and the acquired three-node temperature response sequence as the output response, a least squares algorithm is used for parameter identification to determine the thermal parameter data of the device. This step perturbation identification process is performed sequentially for each device to obtain the individual thermal parameters of all devices.

[0049] In this embodiment, a step power disturbance refers to instantaneously changing the power reference value of a specific module from one steady-state value to another, maintaining this value constant, and then observing the temperature response until a new thermal steady state is established. This disturbance is simple and clear: the input is a power step signal with a fixed amplitude, and the output is a response curve showing a monotonically increasing or decreasing temperature over time. Step disturbances are mainly used to excite the first-order or higher-order thermal inertial responses of the system. The thermal resistance parameter is determined by the steady-state rise of the temperature response curve, and the heat capacity parameter is determined by the time it takes for the temperature response curve to reach 63% of the steady state.

[0050] In this embodiment, the junction-to-case thermal resistance characterizes the thermal conductivity from the power device chip to the package casing, and is determined by the ratio of the steady-state response amplitudes of the junction temperature and the case temperature to power disturbances. The case-to-heatsink thermal resistance characterizes the thermal conductivity of the thermal interface material, and is determined by the ratio of the steady-state response amplitudes of the case temperature and the heatsink temperature to power disturbances. The heatsink-to-ambient thermal resistance characterizes the heat dissipation characteristics of the heatsink to the cooling medium, and is determined by the ratio of the steady-state response amplitudes of the heatsink temperature and the ambient temperature to power disturbances.

[0051] In this embodiment, each thermal capacity parameter is determined by the transient time constant of the temperature response. Junction thermal capacity reflects the chip's heat storage capacity and is determined by the product of the junction temperature rise time constant and the junction-to-case thermal resistance. Case thermal capacity reflects the package's heat storage capacity and is determined by the product of the case temperature rise time constant and the case-to-heat sink thermal resistance. Heat sink thermal capacity reflects the heat storage capacity of the heat sink itself and is determined by the product of the heat sink temperature rise time constant and the heat sink-to-ambient thermal resistance.

[0052] In this embodiment, a pseudo-random binary power perturbation sequence is sequentially injected into each power-changing device, while other devices maintain a constant power and do not undergo additional active temperature intervention. Simultaneously with the perturbation, the junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of each of the other devices (excluding the perturbed device) are acquired, observing how heat is transferred to the nodes of the other devices when the perturbed device generates heat. The pseudo-random perturbation identification process described above is sequentially performed on each device to obtain the mutual thermal resistance parameters in all directions between all devices.

[0053] In this embodiment, pseudo-random binary sequence perturbation refers to rapidly switching the power reference value of a specific module between high and low levels according to a pre-generated pseudo-random sequence. The switching timing of the sequence is determined by the pseudo-random code generated by the shift register, exhibiting wide-spectrum characteristics similar to white noise. This perturbation is complex yet information-rich, with the input being a power signal whose amplitude jumps randomly, and the output being a response signal whose temperature fluctuates around its mean. Pseudo-random binary sequence perturbation is mainly used to simultaneously excite the thermal dynamic response of the system at multiple time scales. By analyzing the correlation between the input power spectrum and the output temperature spectrum, the contributions of different thermal resistance and thermal capacity elements can be separated, making it particularly suitable for identifying weakly coupled thermal resistance parameters between modules.

[0054] In this embodiment, the cross-correlation operation refers to using the pseudo-random binary power disturbance sequence of the disturbed power change device as the input signal, and the junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of another power change device as output signals, respectively, and performing cross-correlation operations on the three output signals sequentially. Specifically, the input signal is multiplied point-by-point with the junction temperature response sequence and then accumulated, and a sliding time delay is used to obtain the junction temperature cross-correlation function sequence; the input signal is multiplied point-by-point with the case temperature response sequence and then accumulated, and a sliding time delay is used to obtain the case temperature cross-correlation function sequence; the input signal is multiplied point-by-point with the radiator temperature response sequence and then accumulated, and a sliding time delay is used to obtain the radiator temperature cross-correlation function sequence. The peak position of the junction temperature cross-correlation function corresponds to the propagation delay from the power disturbance to the junction temperature change, and the peak amplitude reflects the junction temperature coupling strength; the peak position of the case temperature cross-correlation function corresponds to the propagation delay from the power disturbance to the case temperature change, and the peak amplitude reflects the case temperature coupling strength; the peak position of the radiator temperature cross-correlation function corresponds to the propagation delay from the power disturbance to the radiator temperature change, and the peak amplitude reflects the radiator temperature coupling strength. By identifying the peak characteristics of the three cross-correlation functions, the time delay and gain of inter-module thermal coupling at the three hot nodes are estimated respectively.

[0055] In this embodiment, transfer function fitting refers to establishing three mathematical transfer relationships based on the time delay and gain estimates obtained from three cross-correlation operations: from power perturbation to junction temperature change, from power perturbation to case temperature change, and from power perturbation to radiator temperature change. The thermal coupling path is approximated as a first- or second-order thermal resistance-capacity network, and its transfer function is a rational fraction composed of thermal resistance and thermal capacity. A pseudo-random binary power perturbation sequence is used as the input to the transfer function, and the junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence are used as the desired outputs. The coefficients of the three transfer functions are iteratively optimized by minimizing the sum of squared errors between the model output and the actual measured output. During the optimization process, the thermal resistance parameter and thermal capacity parameter are treated as variables to be estimated and are subject to physical constraints. When the sum of squared errors between the model output and the actual measurement of the three transfer functions converges to a minimum, the thermal resistance coefficients in the three transfer functions are extracted as the junction-to-junction thermal resistance parameter, the junction-to-case thermal resistance parameter, and the junction-to-radiator thermal resistance parameter, respectively. Repeat the above cross-correlation calculation and transfer function fitting for each other power change device, and arrange the three mutual thermal resistance parameters according to the module pair relationship to form a mutual thermal resistance parameter matrix.

[0056] In this embodiment, the junction-to-junction thermal resistance is connected between the junction nodes of the two power change devices, the junction-to-shell thermal resistance is connected between the junction node of the disturbance module and the shell node of the target module, and the junction-to-heat sink thermal resistance is connected between the junction node of the disturbance module and the heat sink node of the target module.

[0057] In this embodiment, the thermal resistance between power changing devices in the shared radiator heat conduction coupling path is connected in parallel between the radiator nodes of each power changing device. The airflow temperature node in the forced air cooling duct's airflow-to-air thermal coupling path is connected in series to the ambient temperature end of the thermal resistance path from the radiator to the environment for each power changing device. The radiative heat flow in the rack's internal thermal radiation coupling path is injected as an additional heat source term in the form of a current source into the heat capacity of the corresponding radiator node or shell node of the power changing device. The three types of paths share the same set of node temperature variables and are coupled through the node heat flow balance equation, forming a complete multi-power changing device coupled thermal impedance field model.

[0058] In this embodiment, the thermal parameter data of each power changing device is embedded into the corresponding three-node thermal resistance network model structure, giving each device's internal three-node network specific quantitative parameter values. Then, the mutual thermal resistance parameter data is used as lateral coupling branches to connect the nodes of the corresponding power changing devices. These lateral coupling branches connect the independent three-node thermal resistance networks into an interconnected overall network, constructing a multi-module coupled thermal impedance field model. This multi-module coupled thermal impedance field model simultaneously considers the internal heat conduction paths of the devices and the three types of lateral coupling paths between devices, ensuring that the power change of any device can be accurately reflected at all temperature nodes of all devices.

[0059] The beneficial effects of the above technologies are as follows: determining the thermal parameter data of each power change device and the mutual thermal resistance parameter data between power change devices, constructing a multi-module coupled thermal impedance field model, decomposing the complex multi-physics thermal interaction between devices in the power system into quantifiable coupling paths and mutual thermal resistance parameters, so that the multi-module coupled thermal impedance field model can accurately reflect the comprehensive impact of the heating of any device on the temperature of each node of other devices, and providing high-fidelity thermal coupling simulation capability for predicting the exit time. Example 6:

[0060] Based on Embodiment 1, an AC power supply system includes a prediction module, comprising: Candidate Exit Timing Set Unit: Based on the current time and the predicted termination time of a specified time period, determine the exit time window, and based on the exit time window and the set time step, determine the candidate exit timing set; Acquisition Unit: Acquires the devices to be decommissioned in the AC power system, and acquires the current power share of each power-changing device in the AC power system at the current moment; Update power share unit: Determine the updated power share of the device to be withdrawn as 0. Based on the power share of the device to be withdrawn in the AC power system and all power change devices, determine the normalized updated power share of all power change devices after the withdrawal of the device to be withdrawn. Power loss allocation value unit: For each candidate time in the candidate exit timing set, based on the updated power share undertaken by each power change device and the expected power value at the predicted time corresponding to the candidate time, the power loss allocation value of each power change device is input into the power-loss mapping model of the power change device to determine the power loss allocation value of each power change device at each candidate time. Power loss allocation sequence unit: Based on the power loss allocation value of each power change device at all candidate times, determine the power loss allocation sequence of each power change device; Predicted junction temperature curve unit: Input the power loss allocation sequence of all power change devices into the multi-module coupled thermal impedance field model to determine the predicted junction temperature curve of each power change device within the exit time window.

[0061] In this embodiment, the exit time window is determined based on the current time and the predicted end time of a specified time period. The current time is directly obtained from the real-time clock of the AC power system. The entire time interval between the current time and the predicted end time of the specified time period constitutes the exit time window, which covers all possible time ranges for performing the exit operation. Then, a set of candidate exit opportunities is determined based on this exit time window and a set time step. The set time step is a predefined time interval used to discretize the continuous exit time window into a series of equally spaced time points. Specifically, starting from the current time, one time point is taken at every set time step until the predicted end time is reached or exceeded, and all taken time points are aggregated to form a set of candidate exit opportunities.

[0062] In this embodiment, if the hot exit decision has extremely high timeliness requirements, such as an emergency exit due to a sudden module failure, the set time step can be ten to thirty seconds. In this case, candidate exit opportunities are dense, and the thermal margin field calculation accuracy is high, but the computational overhead increases significantly. This is suitable for scenarios with a small number of modules or abundant computing resources. If the system redundancy is sufficient, and the hot exit decision allows for a certain margin, the set time step can be one to five minutes. In this case, candidate exit opportunities are sparse, and computational efficiency is improved, but the optimal exit opportunity may be missed. This is suitable for scenarios with a large number of modules or limited computing resources. The set time step can be dynamically adjusted according to the system operating status. During periods of high load rate or high temperature stress, the step step is automatically shortened to the ten-second level to improve the granularity of thermal safety verification; during periods of low load rate or abundant thermal margin, the step step can be widened to the two-minute level to reduce the computational burden. The dynamic adjustment mechanism is triggered by dual thresholds: system load rate and minimum thermal margin.

[0063] In this embodiment, the device to be decommissioned refers to a power conversion device planned to switch from the operating state to the decommissioned state. The system continuously monitors the operating status and health status of each power conversion device, including electrical parameters, thermal parameters, and fault diagnosis results. When a diagnosable abnormal state is detected in a power conversion device, such as a continuous abnormal increase in junction temperature, output current harmonic distortion rate exceeding the threshold, switching frequency loss of synchronization, or a decrease in insulation impedance, the device is marked as a device to be decommissioned by the fault diagnosis system. This type of scenario is a fault-driven determination. The system receives maintenance instructions issued by an external operation and maintenance platform, such as the arrival of a periodic preventive maintenance window, the execution of a spare parts replacement plan, or a software upgrade requirement. Devices with specified numbers are marked as devices to be decommissioned. This type of scenario is a maintenance plan determination. The goal of hot decommissioning timing calculation is to find the optimal decommissioning time within the maintenance window to minimize the impact on system operation. The system determines, based on digital twins or health status assessments, that the aging degree of a device is significantly higher than that of other devices, such as the equivalent series resistance of the capacitor increasing to more than 130% of the average value, or the IGBT on-resistance drift exceeding the threshold. This device is marked as a device to be decommissioned by the system's autonomous health management logic. In this type of scenario, which is determined by aging balance, the goal of thermal shutdown timing calculation is to shut it down before aging accelerates, so as to balance the remaining lifespan of each device.

[0064] In this embodiment, the current power share refers to the proportion or absolute value of the total system load undertaken by each device in the current operating state, expressed as a percentage.

[0065] In this embodiment, the updated power share of the device to be decommissioned is determined to be 0, meaning that the device to be decommissioned will no longer bear any load after decommissioning. Then, based on the current power share of the device to be decommissioned and all power-changing devices in the AC power system, a normalized redistribution is performed. Specifically, the power share originally borne by the device to be decommissioned is proportionally distributed to all remaining power-changing devices, so that the sum of the new power shares of all remaining devices is still equal to the total system load share. The normalization process ensures that after decommissioning, the total system load is redistributed by the remaining devices according to a certain proportional rule, such as according to their respective capacity ratios or according to the current redundancy margin ratio.

[0066] In this embodiment, for each candidate exit time in the candidate exit time set, the updated power share of each power-changing device and the expected power value at the predicted time corresponding to that candidate time are input into the device's power-loss mapping model to determine the power loss allocation value for each device at that candidate time. The candidate exit time set is a pre-determined set of future possible exit times, with each candidate time corresponding to a predicted time point. The expected power value refers to the predicted system load value at that predicted time, derived from the aforementioned load prediction sequence. The power-loss mapping model is a functional relationship established in advance through experiments or simulations, describing the power loss generated by the device under a given load power, including conduction losses and switching losses. The loss allocation value is the thermal loss power generated by the device under the new load conditions after exiting, serving as the input excitation for the thermal impedance field model.

[0067] In this embodiment, the power-loss mapping model quantifies the heat loss power generated by the power changing device under a given electrical load condition as the input excitation for predicting the junction temperature of the device. This model is built upon the physical loss mechanism of power semiconductor devices, whose losses mainly consist of conduction losses and switching losses. Conduction losses depend on the device's on-resistance and the effective value of the current flowing through it, and are proportional to the square of the load current. Switching losses depend on the device's switching frequency, bus voltage, and current change rate, and are proportional to the product of the switching frequency and voltage and current. In practical applications, the load current of each device is determined by its updated power share and the system voltage. Therefore, the model takes the device's power share as input and outputs the corresponding heat loss power value through a preset loss characteristic curve or loss lookup table. This loss characteristic curve is typically pre-calibrated using loss parameters provided in the device datasheet, double-pulse experimental measurements, or high-precision electrothermal co-simulation, and corrected for temperature coefficients under different junction temperature conditions. In the power system's exit decision framework, the power-loss mapping model is used to convert the expected power value or power increment at each candidate exit time into the power loss allocation value for each device.

[0068] In this embodiment, based on the power loss allocation value of each power change device at all candidate times, these loss values ​​are arranged in the time order of the candidate times to form a power loss allocation sequence of each device within the exit time window.

[0069] In this embodiment, the power loss distribution sequence of all power-changing devices is simultaneously input into the pre-constructed multi-module coupled thermal impedance field model. The multi-module coupled thermal impedance field model receives the time-varying power loss sequence of each device as excitation input, calculates the dynamic temperature response of the junction, case, and heat sink of each node of each device through the internal thermal resistance network and coupling branches, and outputs the predicted junction temperature curve of each device within the exit time window.

[0070] The beneficial effects of the above technology are as follows: It acquires the devices to be decommissioned from the AC power system; based on the current time and a predicted specified time period, it determines a set of candidate decommissioning opportunities; and based on the set of candidate decommissioning opportunities and a multi-module coupled thermal impedance field model, it determines the predicted junction temperature curve of each power-changing device within the decommissioning time window. By tightly integrating decommissioning decisions with thermal simulation through load redistribution logic and power-loss mapping, the thermal state assessment under each candidate decommissioning opportunity can accurately reflect the real thermal effects of load redistribution after decommissioning. Example 7:

[0071] Based on Embodiment 1, an AC power supply system, including an exit module, comprises: Selection Unit: Extract all power values ​​at each prediction time within the specified prediction time period from all power evolution sequences and sort them to determine the power evolution sequence at each prediction time within the specified prediction time period. Select the median of the power evolution sequence as the expected power value at the prediction time. Select the upper quantile of the set confidence level as the upper limit envelope of the power at the prediction time. Select the lower quantile of the set confidence level as the lower limit envelope of the power at the prediction time. Maximum power increment unit: For each candidate time in the candidate exit timing set, calculate the difference between the upper limit envelope of the power and the expected power value of the predicted time corresponding to the candidate time, and determine the maximum power increment of the candidate time; Maximum power sub-increment unit: Based on the normalized updated power share of all power change devices after the exit of the device to be exited, and the maximum power increment at each candidate moment in the candidate exit timing set, the maximum power sub-increment of each power change device is determined. Equivalent junction temperature increment unit: Based on mutual thermal resistance parameter data and the maximum power sub-increment and thermal parameter data of each power change device, calculate the equivalent junction temperature increment of each power change device; Thermal margin unit: Obtain the safe junction temperature threshold for each power change device, and determine the thermal margin of each power change device at each candidate moment within the exit time window based on the equivalent junction temperature increment, safe junction temperature threshold, and predicted junction temperature curve within the exit time window for each power change device. Power supply thermal margin field unit: Based on the thermal margin of all candidate moments within the exit time window of all power change devices with non-zero updated power share, a power supply thermal margin field is constructed. Exit tag unit: The thermal margin of all power change devices at each candidate time in the power thermal margin field is judged. If the thermal margin of all power change devices at the candidate time is greater than 0, and the minimum thermal margin of all power change devices at the candidate time is greater than the set safety threshold, the exit tag of the candidate time is determined to be feasible. Feasible exit timing set unit: Based on all exit tags within the exit time window being feasible candidate times, determine the set of feasible exit timings.

[0072] In this embodiment, all power values ​​for each prediction time within a specified time period are extracted from all power evolution sequences and sorted. The power evolution sequences are multiple power trajectories covering different future scenarios, with each prediction time corresponding to multiple power values ​​on these trajectories. After sorting all power values ​​for each prediction time, the median of the sorted sequence is selected as the expected power value for that time, representing the typical power level at that time. Simultaneously, the upper quantile of a set confidence level is selected as the upper power limit envelope, for example, the 95th percentile, representing the high-value boundary that the power might reach at that time; the lower quantile of a set confidence level is selected as the lower power limit envelope, for example, the 5th percentile, representing the low-value boundary. The upper and lower power limit envelopes together define the range of uncertainty in future power changes.

[0073] In this embodiment, for each candidate time point in the candidate exit timing set, the difference between the upper power envelope of the predicted time point corresponding to that candidate time point and the expected power value is calculated to determine the maximum power increment for that candidate time point. The maximum power increment represents the additional increase in the total system power relative to the typical expected level under the most unfavorable high-power scenario. This increment represents the worst power conditions that the exit timing needs to withstand.

[0074] In this embodiment, based on the normalized updated power share of all remaining power-changing devices after the device to be exited exits, and the maximum power increment at each candidate moment in the candidate exit timing set, the maximum power increment is allocated to each remaining device according to the updated power share ratio of each device, thus determining the maximum power sub-increment of each power-changing device. The maximum power sub-increment represents the additional power increment share undertaken by each device when the total power reaches the upper limit envelope.

[0075] In this embodiment, based on the mutual thermal resistance parameter data and the maximum power sub-increment and thermal parameter data of each power changing device, the equivalent junction temperature increment of each power changing device is calculated, and the calculation formula is expressed as follows: ; in, This represents the equivalent junction temperature increment of the i-th power conversion device. This represents the maximum power sub-increment of the i-th power changing device and the j-th power changing device. This represents the junction-to-case thermal resistance, case-to-heater thermal resistance, and heatsink-to-ambient thermal resistance in the thermal parameter data of the i-th power conversion device. The junction-to-junction thermal resistance parameter represents the thermal resistance between the j-th power change device and the i-th power change device in the mutual thermal resistance parameter data, where N1 represents the number of power change devices in the AC power system. This indicates the updated power share undertaken by the j-th power changing device; In this embodiment, express When, the value is .

[0076] In this embodiment, based on the equivalent junction temperature increment, safe junction temperature threshold, and predicted junction temperature curve within the exit time window for each power change device, the thermal margin of each power change device at each candidate moment within the exit time window is determined, and the calculation formula is expressed as follows: ; in, This represents the thermal margin of the i-th power conversion device at the k-th candidate time within the exit time window. This represents the safe junction temperature threshold of the i-th power conversion device. This represents the predicted junction temperature at the k-th candidate moment in the predicted junction temperature curve of the i-th power change device within the exit time window. This indicates the updated power share undertaken by the i-th power change device.

[0077] In this embodiment, a power supply thermal margin field is constructed based on the thermal margins of all power change devices that are still operating after exiting the exit time window and whose updated power share is not zero. The power supply thermal margin field is plotted with the candidate exit time as the horizontal axis and each operating device as the vertical axis, with each element representing the thermal margin value of a specific device at a specific candidate time. This field comprehensively describes the distribution of the system's thermal safety state within the exit time window.

[0078] In this embodiment, the thermal margin of all operating devices at each candidate time point within the exit time window in the power supply thermal margin field is assessed. If the thermal margin of all operating devices at that candidate time point is greater than zero (i.e., none of the devices exceeds the safe temperature), and the minimum thermal margin among all operating devices is greater than the set safety threshold (i.e., the thermal margin of the weakest device still has sufficient redundancy), then the exit label for that candidate time point is determined to be feasible. These dual conditions ensure that thermal safety meets design requirements.

[0079] In this embodiment, based on all feasible candidate times for exiting within the exit time window, a set of feasible exit opportunities is formed. This set lists all selectable time points that can safely execute the exit operation under all power evolution scenarios and after considering the thermal coupling effect between devices, for subsequent optimal selection of the final execution time.

[0080] The beneficial effects of the above technologies are as follows: Based on all possible power evolution sequences and the predicted junction temperature curves of all power-changing devices within the exit time window, a power supply thermal margin field is constructed, and a set of feasible exit opportunities is determined. This enables the construction of a power system thermal margin field and the screening of feasible exit opportunities, transforming the uncertainty of power prediction into an envelope boundary and quantifying thermal coupling effects into incremental compensation, thus ensuring that the thermal safety assessment of exit opportunities possesses both statistical robustness and physical accuracy. Example 8:

[0081] Based on Embodiment 7, an AC power supply system, including an exit module, further includes: Triggering Unit: If the feasible exit timing set is empty, trigger an early warning and request external load reduction or activation of a backup cooling source; Execution Exit Time Unit: If the set of feasible exit opportunities is not empty, the execution exit time is determined based on the set of feasible exit opportunities, the thermal margin field, and the lower limit envelope of the power at each predicted time within the specified time period. The device to be exited in the AC power system will exit at the execution exit time.

[0082] In this embodiment, the status of the feasible exit timing set is checked. If the feasible exit timing set is empty, it means that at least one power-changing device's thermal margin does not meet the safety conditions at any of the candidate exit times. That is, the system can not safely execute the exit operation of the device to be exited at any time within the currently predicted power evolution range. At this time, an early warning is triggered, sending an alarm signal to the operators or the upper control system, and simultaneously requesting external load reduction, i.e., requesting a reduction in the total system load to reduce the heat loss of each device, or requesting the activation of a backup cooling source, such as starting additional cooling fans or liquid cooling systems to enhance heat dissipation capacity. External load reduction and backup cooling sources are two proactive intervention methods to change the thermal state of the system. By reducing heat generation or enhancing heat dissipation capacity, the thermal margin is increased, making the originally infeasible exit timing feasible. The purpose of triggering the early warning and requesting external intervention is to create safe exit conditions through the intervention of external resources when a safe exit cannot be completed autonomously, avoiding forced exit that could lead to overheating and damage to the device.

[0083] In this embodiment, if the set of feasible exit opportunities is not empty, that is, if there are one or more candidate times that satisfy the thermal margin of all operating devices being greater than the set safety threshold, then the exit time is determined from the set of feasible exit opportunities.

[0084] In this embodiment, the execution exit time is determined based on the feasible exit opportunity set, the thermal margin field, and the lower limit envelope of the power at each prediction time within the specified time period. The calculation formula is expressed as follows: ; ; ; ; ; Where ET represents the execution exit time, and SE represents the safety assessment value of the b-th candidate time in the set of feasible exit opportunities. This represents the economic assessment value of the b-th candidate time in the set of feasible exit opportunities. N represents the aging assessment value of the b-th candidate moment in the set of feasible exit opportunities, and N3 represents the number of candidate moments in the set of feasible exit opportunities. Indicates economic weight. 2 indicates aging weight. This represents the thermal margin of the a-th power conversion device in the set of feasible exit opportunities at the b-th candidate time. This represents the average thermal margin of all power-changing devices in the power supply thermal margin field at the b-th candidate time in the set of feasible exit opportunities. This represents the minimum thermal margin of all power-changing devices in the power supply thermal margin field at the b-th candidate time in the set of feasible exit opportunities. This indicates the set safety threshold, and N2 represents the number of power changing devices in the power supply thermal margin field. Indicates the first indicator function, Indicates the margin discrete weights, 2 represents the margin evolution weight. This represents the minimum power lower bound envelope and the maximum power lower bound envelope for all prediction times within a specified time period. This represents the lower bound envelope of the power at the predicted time corresponding to the b-th candidate time in the set of feasible exit opportunities. CuM represents the predicted time corresponding to the b-th candidate time in the set of feasible exit opportunities, and CuM represents the current time. Indicates the length of the exit time window.

[0085] In this embodiment, The time step is the interval between two adjacent candidate moments within the exit time window.

[0086] In this embodiment, This represents the thermal margin of the i-th power change device at the k-th candidate moment within the exit time window.

[0087] In this embodiment, economic weight The value range is 0.3-0.7. When there is a large difference between peak and off-peak electricity prices, priority should be given to shutting down during off-peak hours. The value is relatively large.

[0088] In this embodiment, aging weight 2 When the device is nearing the end of its lifespan, it should be decommissioned as early as possible. The value is relatively large.

[0089] In this embodiment, the margin discrete weight control The penalty intensity, ranging from 0.5 to 2, should be prioritized when the AC power system operates in high-reliability scenarios or critical load applications to avoid local hot spots. A value of 1.5 to 2.0 is used to make the system more inclined to choose the time when the thermal margin distribution of each module is balanced during decision-making, ensuring that thermal stress is evenly distributed in space; when the AC power system operates in an economy-first scenario, a certain difference in the thermal margin of each module is allowed in order to obtain a better load exit window, in which case a value of 1.5 to 2.0 is used. The value is 0.5 to 0.7, which reduces the constraint strength of spatial homogeneity on decision-making. The value of is only related to the physical layout of the AC power system. When all modules share a heat sink, the coupling is strong and the uniformity naturally tends to be the same. The smaller value should be taken; when each module has independent heat dissipation, the coupling is weak, and spatial differences fully reflect uneven load. The larger value should be taken. This parameter should be determined based on the heatsink layout during the AC power system calibration phase and should remain unchanged during operation.

[0090] In this embodiment, the margin evolution weight 2 control The penalty intensity, ranging from 5 to 20, should be prioritized when the AC power system operates in a high-temperature environment or under frequent load fluctuations, taking into account the rapid deterioration trend of the thermal margin. A value of 2, between 15 and 20, makes the AC power system more inclined to choose a time when the thermal state is improving or stabilizing, avoiding termination when deterioration accelerates; when the AC power system is operating in a scenario with stable load and slow temperature changes, a value of 2 is chosen. A value of 2 is between 5 and 8, which reduces sensitivity to the rate of change and avoids invalid penalties due to measurement noise or minor fluctuations. The value of 2 is determined by the system's thermal time constant—systems with small heat capacity and fast response have a relatively large rate of change of heat margin. 2 should be a smaller value; systems with large heat capacity and slow response naturally have smaller rates of change. 2 should be set to a large value to produce an effective penalty. This parameter should be determined based on the thermal response characteristics during the AC power system calibration phase and should remain unchanged during operation.

[0091] In this embodiment, after determining the exit time, the device to be exited in the AC power system performs an exit operation at that time, switching the device from the running state to the exit state, thus completing the entire safe exit process.

[0092] The beneficial effects of the above technologies are: determining and executing the exit time; realizing an adaptive exit execution and safety fallback mechanism for AC power systems; expanding equipment exit decisions from a single thermal margin feasibility judgment to a two-stage closed loop of empty set emergency response and non-empty optimal selection; enabling the power system to maintain safety boundaries through external collaborative means even under thermal safety critical conditions; and improving the integrity of thermal safety assurance and engineering practicality of the exit process.

[0093] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An AC power supply system, characterized in that, include: Evolution Module: Acquires historical operating data of the AC power system for each natural day within a specified historical period, constructs a long-term trend prediction model, and determines the load prediction sequence and multiple possible power evolution sequences of the AC power system within the specified prediction time period. Module construction: Acquire thermal response data of AC power system, establish three-node thermal resistance network model structure for each power change device, determine thermal parameter data of each power change device and mutual thermal resistance parameter data between power change devices, and construct multi-module coupled thermal impedance field model. Prediction module: acquires the devices to be decommissioned in the AC power system, determines the candidate decommissioning timing set based on the current time and the predicted specified time period, and determines the predicted junction temperature curve of each power change device within the decommissioning time window based on the candidate decommissioning timing set and the multi-module coupled thermal impedance field model. Exit Module: Based on all possible power evolution sequences and the predicted junction temperature curves of all power change devices within the exit time window, construct the power supply thermal margin field, determine the set of feasible exit opportunities, and execute the exit time.

2. The AC power supply system according to claim 1, characterized in that, Evolutionary modules, including: Historical Operation Data Unit: Acquires historical operation data of the AC power system for each natural day within a specified historical period. The historical operation data includes historical daily power time series, cyclic operation patterns, historical load rate time series of each power change device, and historical exit data. Cyclic operation patterns include start-up and shutdown time series of multiple power change devices, rotation cycle data, and priority data. Historical exit data includes multiple exit events of power change devices, the time of occurrence of each exit, the reason for exit, the thermal state before exit, the number of remaining devices after exit, load redistribution data, overheat alarm tags, and load reduction protection tags. Long-term trend prediction model unit: Based on historical operating data of all natural days within a specified historical period, a long-term trend prediction model is constructed. The long-term trend prediction model includes at least a feature extraction module and a time series decomposition extrapolation module. The feature extraction module extracts features from the cyclical operating patterns, historical load rate time series of each power change device, and historical exit data in the historical operating data of all natural days within the specified historical period. The time series decomposition extrapolation module performs seasonal decomposition, trend extraction, and residual modeling on the historical daily power series, and extrapolates the decomposition results to the specified prediction time period.

3. The AC power supply system according to claim 1, characterized in that, The evolution module also includes: Load forecast sequence unit: Based on a long-term trend forecast model, determine the load forecast sequence of the AC power system within a specified forecast time period; Short-term dispatch data unit: acquires short-term dispatch data from multiple information sources of the AC power system, and labels each information source with information tags, including definite information and uncertain information; First power increment sequence unit: Transforms short-term mobilization data of each information source with information labeled as definite information into a power increment sequence for predicting a specified time period; The second power increment sequence unit: Monte Carlo sampling is used to construct a deviation probability distribution model based on the short-term mobilization data of each information source with uncertain information label, and generate multiple possible power increment sequences for each information source with uncertain information label within the specified prediction time period; Power Evolution Sequence Unit: Based on the load prediction sequence of the AC power system, the power increment sequence of all information sources with definite information labels, and multiple possible power increment sequences of all information sources with uncertain information labels, multiple possible power evolution sequences of the AC power system are generated within the specified prediction time period.

4. An AC power supply system according to claim 1, characterized in that, Build modules, including: Thermal response data unit: Acquires thermal response data of AC power system, including historical power disturbance sequence, junction temperature time series, case temperature time series, radiator temperature time series, ambient temperature time series, and inter-device thermal coupling data, including heat dissipation coupling data, air duct coupling data and radiation coupling data of each radiator. Model structure unit: Establish a three-node thermal resistance network model structure for each power change device. The three nodes include junction nodes, shell nodes, and heat sink nodes. The nodes are connected by thermal resistance and thermal capacity to form a heat flow conduction path.

5. An AC power supply system according to claim 4, characterized in that, The building blocks also include: Coupling Path Unit: Based on the heat dissipation coupling data in the inter-device thermal coupling data of the thermal response data, a heat dissipation conduction coupling path is generated for each radiator; based on the air duct coupling data in the inter-device thermal coupling data of the thermal response data, an airflow thermal coupling path is generated; and simultaneously, based on the radiation coupling data in the inter-device thermal coupling data of the thermal response data, a thermal radiation coupling path is generated. Thermal parameter data unit: A step power perturbation sequence is sequentially injected into each power change device. Except for the power change device, each power change device maintains a constant power. The junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of the power change device are collected. The step power perturbation sequence is used as input, and the junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of the power change device are used as output. The least squares algorithm is used to determine the thermal parameter data of each power change device. The thermal parameter data includes junction-to-case thermal resistance, case-to-radiator thermal resistance, radiator-to-ambient thermal resistance, junction heat capacity, case heat capacity, and radiator heat capacity. Mutual thermal resistance parameter data unit: A pseudo-random binary power perturbation sequence is sequentially injected into each power change device. Except for the power change device, each power change device maintains a constant power and does not perform additional active temperature intervention. The junction temperature response sequence, case temperature response sequence, and radiator temperature response sequence of each power change device are collected. The pseudo-random binary power perturbation sequence is cross-correlated with the junction temperature response curve, case temperature response curve, and radiator temperature response curve of each power change device. Transfer function fitting is performed to determine the mutual thermal resistance parameter vector from the power change device to each power change device. The mutual thermal resistance parameter data is determined. The mutual thermal resistance parameter vector includes junction-to-junction mutual thermal resistance parameters, junction-to-case mutual thermal resistance parameters, and junction-to-radiator mutual thermal resistance parameters. Multi-module coupled thermal impedance field model unit: The thermal parameter data of each power change device is embedded into the corresponding three-node thermal resistance network model structure, and the mutual thermal resistance parameter data is connected to the nodes of the corresponding power change device as a lateral coupling branch to construct a multi-module coupled thermal impedance field model.

6. An AC power supply system according to claim 1, characterized in that, The prediction module includes: Candidate Exit Timing Set Unit: Based on the current time and the predicted termination time of a specified time period, determine the exit time window, and based on the exit time window and the set time step, determine the candidate exit timing set; Acquisition Unit: Acquires the devices to be decommissioned in the AC power system, and acquires the current power share of each power-changing device in the AC power system at the current moment; Update power share unit: Determine the updated power share of the device to be withdrawn as 0. Based on the power share of the device to be withdrawn in the AC power system and all power change devices, determine the normalized updated power share of all power change devices after the withdrawal of the device to be withdrawn. Power loss allocation value unit: For each candidate time in the candidate exit timing set, based on the updated power share undertaken by each power change device and the expected power value at the predicted time corresponding to the candidate time, the power loss allocation value of each power change device is input into the power-loss mapping model of the power change device to determine the power loss allocation value of each power change device at each candidate time. Power loss allocation sequence unit: Based on the power loss allocation value of each power change device at all candidate times, determine the power loss allocation sequence of each power change device; Predicted junction temperature curve unit: Input the power loss allocation sequence of all power change devices into the multi-module coupled thermal impedance field model to determine the predicted junction temperature curve of each power change device within the exit time window.

7. An AC power supply system according to claim 1, characterized in that, Exit module, including: Selection Unit: Extract all power values ​​at each prediction time within the specified prediction time period from all power evolution sequences and sort them to determine the power evolution sequence at each prediction time within the specified prediction time period. Select the median of the power evolution sequence as the expected power value at the prediction time. Select the upper quantile of the set confidence level as the upper limit envelope of the power at the prediction time. Select the lower quantile of the set confidence level as the lower limit envelope of the power at the prediction time. Maximum power increment unit: For each candidate time in the candidate exit timing set, calculate the difference between the upper limit envelope of the power and the expected power value of the predicted time corresponding to the candidate time, and determine the maximum power increment of the candidate time; Maximum power sub-increment unit: Based on the normalized updated power share of all power change devices after the exit of the device to be exited, and the maximum power increment at each candidate moment in the candidate exit timing set, the maximum power sub-increment of each power change device is determined. Equivalent junction temperature increment unit: Based on mutual thermal resistance parameter data and the maximum power sub-increment and thermal parameter data of each power change device, calculate the equivalent junction temperature increment of each power change device; Thermal margin unit: Obtain the safe junction temperature threshold for each power change device, and determine the thermal margin of each power change device at each candidate moment within the exit time window based on the equivalent junction temperature increment, safe junction temperature threshold, and predicted junction temperature curve within the exit time window for each power change device. Power supply thermal margin field unit: Based on the thermal margin of all candidate moments within the exit time window of all power change devices with non-zero updated power share, a power supply thermal margin field is constructed. Exit tag unit: The thermal margin of all power change devices at each candidate time in the power thermal margin field is judged. If the thermal margin of all power change devices at the candidate time is greater than 0, and the minimum thermal margin of all power change devices at the candidate time is greater than the set safety threshold, the exit tag of the candidate time is determined to be feasible. Feasible exit timing set unit: Based on all exit tags within the exit time window being feasible candidate times, determine the set of feasible exit timings.

8. An AC power supply system according to claim 7, characterized in that, The exit module also includes: Triggering Unit: If the feasible exit timing set is empty, trigger an early warning and request external load reduction or activation of a backup cooling source; Execution Exit Time Unit: If the set of feasible exit opportunities is not empty, the execution exit time is determined based on the set of feasible exit opportunities, the thermal margin field, and the lower limit envelope of the power at each predicted time within the specified time period. The device to be exited in the AC power system will exit at the execution exit time.

Images