Emergency power equipment group power coordination optimization method based on deep learning
By using deep learning-based load prediction and particle swarm optimization algorithms, the frequency and voltage of emergency power generation equipment are dynamically adjusted, solving the power coordination problem of emergency power generation equipment groups in multi-objective and highly dynamic scenarios. This achieves forward-looking load prediction and real-time power allocation, improving power supply stability and equipment efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID FUJIAN ELECTRIC POWER RES INST
- Filing Date
- 2026-04-09
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies for power coordination control of emergency power generation equipment groups suffer from problems such as excessively long calculation times, difficulty in adapting to sudden load changes and equipment anomalies, and unbalanced power distribution, leading to unstable power supply and increased equipment losses.
A deep learning-based approach is adopted to construct a load prediction model through a long short-term memory network, combine it with a particle swarm optimization algorithm to generate a power allocation strategy, and dynamically adjust the output frequency and voltage of the emergency power generation equipment through a group power coordination controller.
It enables accurate prediction and rapid response to future loads, improves the real-time performance and adaptability of power allocation, solves the "curse of dimensionality" problem of traditional methods in multi-objective and highly dynamic scenarios, and ensures the stability of power supply and the efficient operation of equipment.
Smart Images
Figure CN122437141A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of emergency power supply technology for power systems, specifically relating to a deep learning-based method for power coordination optimization of emergency power generation equipment groups. Background Technology
[0002] Emergency power generation equipment is crucial for ensuring power supply to critical loads during natural disasters or emergencies. In actual emergency scenarios, multiple emergency power generation devices (such as low-voltage power supply vehicles, medium-voltage power supply vehicles, and mobile energy storage equipment) are usually connected in parallel to form a power generation equipment group to meet larger power demands and improve power supply reliability.
[0003] In group power coordination control, the core issue to be addressed is how to achieve rapid, economical, and stable allocation of active and reactive power among various equipment while meeting load demands. Existing technologies include schemes using global optimization algorithms such as particle swarm optimization for power allocation optimization. However, these traditional methods have significant shortcomings in practical applications: First, as global optimization algorithms, in multi-objective, highly dynamic emergency power supply scenarios, when the number of equipment increases or constraints become complex, they are prone to the "curse of dimensionality," leading to excessively long computation times and an inability to adapt in real-time to sudden load changes or abnormal equipment conditions. Second, traditional methods lack the ability to predict future load changes. When multiple generating units are operating in parallel, sudden load switching can cause power surges, and existing methods struggle to achieve rapid, adaptive power pre-allocation, easily causing drastic fluctuations in power supply parameters, triggering equipment protection actions, and even leading to power outage risks. Furthermore, power allocation imbalances can exacerbate equipment wear and tear, shortening equipment lifespan.
[0004] Therefore, improving the real-time performance, foresight, and adaptability of power coordination control for emergency power generation equipment groups is a technical problem that urgently needs to be solved in this field. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a deep learning-based method for power coordination optimization of emergency power generation equipment groups, so as to achieve accurate prediction of future load, rapid generation of power allocation strategies, and dynamic synchronous adjustment of output power.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: a power coordination optimization method for emergency power generation equipment groups based on deep learning, comprising the following steps:
[0007] Step S101: Collect real-time operating data of the emergency power generation equipment group to form a real-time operating dataset;
[0008] Step S102: Obtain historical operating data of the emergency power generation equipment group to form a historical operating dataset; construct a load prediction model based on the historical operating dataset, the output of which is the load prediction value for a future preset time period;
[0009] Step S103: Input the real-time running dataset into the load prediction model to obtain the load prediction value for a future preset time period; use an optimization algorithm to generate a power allocation strategy based on the load prediction value, wherein the power allocation strategy includes the active power setpoint and reactive power setpoint of each emergency power generation equipment.
[0010] Step S104: According to the power allocation strategy, dynamically adjust the output frequency and voltage of each emergency power generation device through the group power coordination controller.
[0011] Furthermore, in step S101, the emergency power generation equipment group includes multiple emergency power generation equipment operating in parallel; the real-time operation dataset includes the voltage value, current value, active power output value, reactive power output value, load demand value, ambient temperature value, and grid connection point frequency value of each emergency power generation equipment.
[0012] Further, in step S101, the real-time operation data of the emergency power generation equipment group is collected to form a real-time operation dataset, specifically: real-time operation data of the emergency power generation equipment group is collected at a sampling frequency of at least 10 times per second, and the real-time operation data is preprocessed to form a real-time operation dataset; the preprocessing includes normalization processing and outlier removal of the real-time operation data.
[0013] Furthermore, in step S101, the emergency power generation equipment includes a low-voltage power vehicle, a medium-voltage power vehicle, and mobile energy storage equipment, and the emergency power generation equipment group includes at least 10 emergency power generation equipment units.
[0014] Further, in step S102, the construction of the load prediction model based on the historical operating dataset specifically includes:
[0015] Historical running datasets are used as training samples for a deep learning model, which is a long short-term memory network.
[0016] The training samples are input into the Long Short-Term Memory network. The number of nodes in the input layer of the Long Short-Term Memory network corresponds to the number of features in the historical running data, the number of nodes in the hidden layer is set to 128, and the number of nodes in the output layer corresponds to the load prediction value.
[0017] The load prediction model is trained by minimizing the mean squared error loss function using the backpropagation algorithm.
[0018] Further, in step S102, the nodes of the input layer include emergency event type encoding feature nodes and key load priority parameter nodes; the emergency load mutation detection submodule of the hidden layer establishes a data interaction channel with the load prediction value calculation unit of the output layer. When a load mutation is detected, the output layer triggers a prediction value correction mechanism with a correction coefficient of 1.05-1.1, which is dynamically adjusted according to the mutation magnitude.
[0019] Further, in step S103, the optimization algorithm is a particle swarm optimization algorithm; the step of using the optimization algorithm to generate a power allocation strategy based on the load prediction value specifically includes:
[0020] The load prediction value is input into the particle swarm optimization algorithm, and the power allocation strategy is obtained by minimizing the total fuel consumption and voltage deviation of the group as the objective function. The power allocation strategy includes the active power setpoint and reactive power setpoint of each emergency power generation equipment. The strategy execution instruction of the power allocation strategy is generated according to the power allocation strategy.
[0021] Furthermore, the objective function of the particle swarm optimization algorithm also includes an emergency response delay constraint term and a critical load priority weighting factor; the expression of the objective function is:
[0022] min (F=α×(C×W)+β×ΔU+γ×k×max(0,t-500));
[0023] Where F represents the total optimization target value, α, β, and γ are the weighting coefficients of the group's total fuel consumption C, grid connection point voltage deviation ΔU, and emergency response delay penalty, respectively, and α+β+γ=1, α∈[0.3,0.5], β∈[0.3,0.5], γ∈[0.1,0.2]; k is the adjustment delay penalty coefficient, with a value range of 0.8-1.2; t is the estimated delay for power adjustment of a single piece of equipment, and the delay penalty term is triggered when t>500ms; W is the priority weighting factor for critical loads, with a value range of 0.8-1.2.
[0024] Further, in step S104, the step of dynamically adjusting the output frequency and voltage of each emergency power generation device according to the power allocation strategy through the group power coordination controller specifically includes:
[0025] The group power coordination controller issues policy execution instructions to the local controllers of each emergency power generation device.
[0026] The local controller converts the active power setpoint into a frequency reference value adjustment amount and the reactive power setpoint into a voltage reference value adjustment amount according to the droop control; and adjusts the frequency and voltage of the emergency power generation equipment according to the frequency reference value adjustment amount and the voltage reference value adjustment amount.
[0027] The present invention also provides a power coordination and optimization system for emergency power generation equipment groups based on deep learning, including a memory, a processor, and computer program instructions stored in the memory and executable by the processor. When the processor executes the computer program instructions, it can implement the above-mentioned method.
[0028] Compared with existing technologies, this invention has the following advantages: Based on real-time and historical operating data of emergency power generation equipment groups, this invention constructs a load prediction model through a long short-term memory network, effectively alleviating the lag problem caused by the lack of prediction in traditional particle swarm optimization algorithms. By using the load prediction value as input to the particle swarm optimization algorithm, it achieves forward-looking adaptation to the load over a preset time period in the future. Based on this, a power allocation strategy is generated, and the output frequency and voltage of the emergency power generation equipment are dynamically adjusted through a group power coordination controller and local droop control. This invention realizes a complete technical closed loop of future load prediction, active / reactive power allocation, and power output synchronization, achieving the dual goals of group power balance and energy efficiency optimization, while significantly improving the real-time performance and adaptability of power allocation. Therefore, this invention effectively alleviates the "curse of dimensionality" problem that traditional particle swarm optimization algorithms are prone to in multi-objective, highly dynamic scenarios, and overcomes the shortcomings of traditional methods in adapting to sudden load changes or equipment anomalies. Attached Figure Description
[0029] Figure 1 This is a flowchart of a deep learning-based emergency power generation equipment group power coordination optimization method provided in an embodiment of the present invention. Detailed Implementation
[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0031] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0032] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0033] like Figure 1 As shown in the figure, this embodiment provides a power coordination optimization method for emergency power generation equipment groups based on deep learning, and its implementation steps are as follows.
[0034] Step S101: Collect real-time operating data of the emergency power generation equipment group to form a real-time operating dataset.
[0035] The emergency power generation equipment group includes multiple emergency power generation equipment operating in parallel; the real-time operation dataset includes the voltage value, current value, active power output value, reactive power output value, load demand value, ambient temperature value, and grid connection point frequency value of each emergency power generation equipment.
[0036] The emergency power generation equipment includes a low-voltage power vehicle, a medium-voltage power vehicle, and mobile energy storage equipment. The emergency power generation equipment group includes at least 10 emergency power generation devices.
[0037] The process of collecting real-time operational data from the emergency power generation equipment group to form a real-time operational dataset involves: collecting real-time operational data from the emergency power generation equipment group at a sampling frequency of at least 10 times per second, preprocessing the real-time operational data, and forming a real-time operational dataset. The preprocessing includes normalizing the real-time operational data and removing outliers.
[0038] Step S102: Obtain historical operating data of the emergency power generation equipment group to form a historical operating dataset; construct a load prediction model based on the historical operating dataset, and the output of the load prediction model is the load prediction value for a future preset time period.
[0039] Specifically, the construction of the load prediction model based on historical operational datasets includes:
[0040] 1) Use historical running datasets as training samples for the deep learning model, which is a long short-term memory network;
[0041] 2) Input the training samples into the Long Short-Term Memory network. The number of input layer nodes of the Long Short-Term Memory network corresponds to the number of features of the historical running data, the number of hidden layer nodes is set to 128, and the number of output layer nodes corresponds to the load prediction value.
[0042] The historical operating data has a total of 7 features, namely voltage value, current value, active power output value, reactive power output value, load demand value, ambient temperature value, and grid connection point frequency value.
[0043] 3) The load prediction model is trained by minimizing the mean squared error loss function through the backpropagation algorithm.
[0044] In this embodiment, the nodes in the input layer include three emergency event type encoding feature nodes and two key load priority parameter nodes. The emergency load mutation detection submodule in the hidden layer establishes a data interaction channel with the load prediction calculation unit in the output layer. When a load mutation is detected, the output layer triggers a prediction correction mechanism with a correction coefficient of 1.05-1.1, which is dynamically adjusted according to the mutation magnitude.
[0045] In constructing the load prediction model, a multi-dimensional design was adopted: First, historical operating data covering voltage, current, active / reactive power output, load demand, ambient temperature, and grid connection frequency were used as training samples. This comprehensively incorporates factors affecting emergency power generation load, avoiding the bias of predictions caused by single features and providing rich data support for load prediction. Second, a Long Short-Term Memory (LSTM) network, a deep learning model adapted to time-series data, was selected. By matching the number of input layer nodes with the number of historical data features, the dynamic patterns of load changes over time in emergency power generation scenarios can be effectively captured, solving the problem that traditional models are difficult to adapt to changes in time-series load. Third, the number of hidden layer nodes was set to 128. This ensures that the model has sufficient learning ability to explore deep data correlations while avoiding the risk of overfitting due to too many nodes, balancing the model's learning depth and stability. Fourth, the mean squared error loss function was minimized through backpropagation, which continuously iteratively optimizes model parameters, significantly reducing load prediction errors and ensuring the reliability of the output load prediction values for the future preset time period. This provides a basis for power allocation strategies based on prediction values in subsequent steps, improving the rationality and foresight of power coordination of emergency power generation equipment groups from the source.
[0046] Step S103: Input the real-time running dataset into the load prediction model to obtain the load prediction value for a future preset time period. Using an optimization algorithm, generate a power allocation strategy based on the load prediction value. The power allocation strategy includes the active power setpoint and reactive power setpoint for each emergency power generation device.
[0047] The optimization algorithm is a particle swarm optimization algorithm. The process of generating a power allocation strategy based on the load prediction using this algorithm is as follows:
[0048] The load prediction value is input into the particle swarm optimization algorithm, and the power allocation strategy is obtained by minimizing the total fuel consumption and voltage deviation of the group as the objective function. The power allocation strategy includes the active power setpoint and reactive power setpoint of each emergency power generation equipment. The strategy execution instruction of the power allocation strategy is generated according to the power allocation strategy.
[0049] The objective function of the particle swarm optimization algorithm also includes an emergency response delay constraint and a critical load priority weighting factor; the expression of the objective function is:
[0050] min (F=α×(C×W)+β×ΔU+γ×k×max(0,t-500));
[0051] Where F represents the total optimization target value, α, β, and γ are the weighting coefficients of the total fuel consumption of the group C, the voltage deviation at the grid connection point ΔU, and the emergency response delay penalty, respectively, and α+β+γ=1, α∈[0.3,0.5], β∈[0.3,0.5], γ∈[0.1,0.2].
[0052] k is the adjustment delay penalty coefficient, which ranges from 0.8 to 1.2; t is the estimated delay for power adjustment of a single piece of equipment, in milliseconds. When t > 500 ms, the delay penalty is calculated.
[0053] W is the priority weighting factor for critical loads, with a value ranging from 0.8 to 1.2. For example, the weighting factor for equipment supplying power to medical loads is set to 1.2; for equipment supplying power to emergency communication loads, it is set to 1.1; and for equipment supplying power to general lighting loads, it is set to 0.8. By multiplying by W, the fuel consumption cost corresponding to high-priority loads is amplified, thereby guiding the algorithm to prioritize critical loads.
[0054] In generating the power allocation strategy, the following effects were achieved by combining particle swarm optimization (PSO) with load forecasts and setting a dual objective function: First, by using load forecasts as input, the power allocation strategy can adapt to the load demand of a preset time period in advance, avoiding the allocation lag problem caused by the lack of prediction in traditional methods, and providing forward-looking guidance for emergency power generation equipment groups to cope with potential load changes; Second, by using "minimizing the total fuel consumption of the group" as the objective function, the power output ratio of each piece of equipment can be optimized while meeting load demand, reducing unnecessary fuel consumption, significantly improving energy utilization efficiency and economy in emergency power supply scenarios, and reducing operating costs; Third, simultaneously minimizing voltage deviation... With the goal of controlling the reactive power output of each piece of equipment, effectively suppressing voltage fluctuations, ensuring that the grid connection point voltage remains stable within a reasonable range, and avoiding problems such as equipment protection actions or sensitive loads failing to operate normally due to excessive voltage deviation; fourth, clearly outputting the active power setpoint and reactive power setpoint of each emergency power generation equipment, making power allocation more targeted and precise, avoiding power distribution imbalances such as some equipment being overloaded and others being lightly loaded when multiple pieces of equipment are connected in parallel, and reducing equipment losses; at the same time, the generated strategy execution instructions provide a clear and practical operational basis for the subsequent dynamic adjustment of the group power coordination controller, further ensuring the effective execution of the power allocation strategy and helping to improve the stability and reliability of the emergency power generation equipment group operation.
[0055] Step S104: According to the power allocation strategy, dynamically adjust the output frequency and voltage of each emergency power generation device through the group power coordination controller.
[0056] Specifically, the step of dynamically adjusting the output frequency and voltage of each emergency power generation device through a group power coordination controller according to a power allocation strategy includes:
[0057] 1) The group power coordination controller issues policy execution instructions to the local controllers of each emergency power generation device;
[0058] 2) The local controller converts the active power setpoint into a frequency reference value adjustment amount and the reactive power setpoint into a voltage reference value adjustment amount according to the droop control; it adjusts the frequency and voltage of the emergency power generation equipment according to the frequency reference value adjustment amount and the voltage reference value adjustment amount to achieve power output synchronization.
[0059] This invention, based on real-time and historical operating data of emergency power generation equipment groups, constructs a load prediction model through a long short-term memory network, mitigating the lag of traditional particle swarm optimization (PSO) algorithms. It combines PSO optimization to generate power allocation strategies and dynamically adjusts the output frequency and voltage of emergency power generation equipment through a group power coordination controller and local droop control. This achieves load prediction for a preset time period, allocation of active / reactive power among emergency power generation equipment, and power output synchronization, realizing group power balance and energy efficiency optimization while improving the real-time performance and adaptability of power allocation. It effectively alleviates the "curse of dimensionality" problem that traditional PSO algorithms easily encounter in multi-objective, highly dynamic scenarios, and addresses the difficulty of adapting to sudden load changes or equipment anomalies.
[0060] This embodiment also provides a power coordination and optimization system for emergency power generation equipment groups based on deep learning, including a memory, a processor, and computer program instructions stored in the memory and executable by the processor. When the processor executes the computer program instructions, it can implement the above-mentioned method.
[0061] This embodiment also provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the above-described method.
[0062] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0063] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A deep learning-based method for power coordination optimization of emergency power generation equipment groups, characterized in that, Includes the following steps: Step S101: Collect real-time operating data of the emergency power generation equipment group to form a real-time operating dataset; Step S102: Obtain historical operating data of the emergency power generation equipment group to form a historical operating dataset; construct a load prediction model based on the historical operating dataset, the output of which is the load prediction value for a future preset time period; Step S103: Input the real-time running dataset into the load prediction model to obtain the load prediction value for a future preset time period; use an optimization algorithm to generate a power allocation strategy based on the load prediction value, wherein the power allocation strategy includes the active power setpoint and reactive power setpoint of each emergency power generation equipment. Step S104: According to the power allocation strategy, dynamically adjust the output frequency and voltage of each emergency power generation device through the group power coordination controller.
2. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 1, characterized in that, In step S101, the emergency power generation equipment group includes multiple emergency power generation equipment operating in parallel; the real-time operation dataset includes the voltage value, current value, active power output value, reactive power output value, load demand value, ambient temperature value, and grid connection point frequency value of each emergency power generation equipment.
3. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 1, characterized in that, In step S101, the real-time operation data of the emergency power generation equipment group is collected to form a real-time operation dataset. Specifically, the real-time operation data of the emergency power generation equipment group is collected at a sampling frequency of at least 10 times per second, and the real-time operation data is preprocessed to form a real-time operation dataset. The preprocessing includes normalization processing and outlier removal of the real-time operation data.
4. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 1, characterized in that, In step S101, the emergency power generation equipment includes a low-voltage power vehicle, a medium-voltage power vehicle, and mobile energy storage equipment, and the emergency power generation equipment group includes at least 10 emergency power generation equipment units.
5. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 1, characterized in that, In step S102, the construction of the load prediction model based on the historical operating dataset specifically includes: Historical running datasets are used as training samples for a deep learning model, which is a long short-term memory network. The training samples are input into the Long Short-Term Memory network. The number of nodes in the input layer of the Long Short-Term Memory network corresponds to the number of features in the historical running data, the number of nodes in the hidden layer is set to 128, and the number of nodes in the output layer corresponds to the load prediction value. The load prediction model is trained by minimizing the mean squared error loss function using the backpropagation algorithm.
6. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 5, characterized in that, In step S102, the nodes of the input layer include emergency event type encoding feature nodes and key load priority parameter nodes; the emergency load mutation detection submodule of the hidden layer establishes a data interaction channel with the load prediction value calculation unit of the output layer. When a load mutation is detected, the output layer triggers a prediction value correction mechanism with a correction coefficient of 1.05-1.1, which is dynamically adjusted according to the mutation magnitude.
7. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 1, characterized in that, In step S103, the optimization algorithm is a particle swarm optimization algorithm; the step of using the optimization algorithm to generate a power allocation strategy based on the load prediction value specifically involves: The load prediction value is input into the particle swarm optimization algorithm, and the power allocation strategy is obtained by minimizing the total fuel consumption and voltage deviation of the group as the objective function. The power allocation strategy includes the active power setpoint and reactive power setpoint of each emergency power generation equipment. The strategy execution instruction of the power allocation strategy is generated according to the power allocation strategy.
8. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 7, characterized in that, The objective function of the particle swarm optimization algorithm also includes an emergency response delay constraint and a critical load priority weighting factor; the expression of the objective function is: min (F=α×(C×W)+β×ΔU+γ×k×max(0,t-500)); Where F represents the total optimization target value, α, β, and γ are the weighting coefficients of the group's total fuel consumption C, grid connection point voltage deviation ΔU, and emergency response delay penalty, respectively, and α+β+γ=1, α∈[0.3,0.5], β∈[0.3,0.5], γ∈[0.1,0.2]; k is the adjustment delay penalty coefficient, with a value range of 0.8-1.2; t is the estimated delay for power adjustment of a single piece of equipment, and the delay penalty term is triggered when t>500ms; W is the priority weighting factor for critical loads, with a value range of 0.8-1.
2.
9. The method for power coordination optimization of emergency power generation equipment groups based on deep learning according to claim 1, characterized in that, In step S104, the step of dynamically adjusting the output frequency and voltage of each emergency power generation device according to the power allocation strategy through the group power coordination controller specifically includes: The group power coordination controller issues policy execution instructions to the local controllers of each emergency power generation device. The local controller converts the active power setpoint into a frequency reference value adjustment amount and the reactive power setpoint into a voltage reference value adjustment amount according to the droop control; and adjusts the frequency and voltage of the emergency power generation equipment according to the frequency reference value adjustment amount and the voltage reference value adjustment amount.
10. A power coordination optimization system for emergency power generation equipment groups based on deep learning, characterized in that, It includes a memory, a processor, and computer program instructions stored in the memory and executable by the processor, wherein when the processor executes the computer program instructions, it can implement the method as described in any one of claims 1-9.