CN-121984031-A - AI-based micro-grid multi-source collaborative optimization scheduling method

CN121984031ACN 121984031 ACN121984031 ACN 121984031ACN-121984031-A

Abstract

The invention relates to the technical field of micro-grid operation control, in particular to an AI-based micro-grid multi-source collaborative optimization scheduling method. Inputting the feature vector into a pre-trained multi-source collaborative characterization neural network model to generate a collaborative state vector set fusing the multi-source complementary characteristics and the time sequence dependency relationship. On the basis, a scheduling decision generation network model is called to conduct multi-step prospective optimization deduction, and a coordinated scheduling instruction set of a plurality of time periods in the future is generated. And carrying out safety constraint verification on the instruction set to form a final executable scheduling control instruction sequence and transmitting the scheduling control instruction sequence to controllable equipment. According to the invention, the deep learning model is utilized to deeply mine the multi-source cooperative characteristic and make a look-ahead decision, so that the intelligent level and the overall optimization effect of the operation of the micro-grid are improved.

Inventors

ZHOU XIAOWEI
WANG FENG
DING WUJUN
LI JIANG

Assignees

杭州鸿途智慧能源技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260211

Claims (10)

1. The micro-grid multi-source collaborative optimization scheduling method based on the AI is characterized by comprising the following steps of: Acquiring power generation data of a distributed power supply, state data of an energy storage device, operation data of an adjustable load and interactive constraint data of an upper power grid in a micro-grid system, and generating a multi-source heterogeneous micro-grid original operation data set; Performing data cleaning and format unification processing on the original operation data set of the micro-grid to form a standardized micro-grid operation data set, and extracting key feature vectors representing multi-source cooperative relationships from the standardized micro-grid operation data set; inputting the key feature vector into a pre-trained multi-source collaborative characterization neural network model, and carrying out joint state coding on the distributed power supply, the energy storage device and the adjustable load to generate a collaborative state vector set containing multi-source complementary characteristics and time sequence dependency relations; based on the collaborative state vector set, calling a scheduling decision generation network model to carry out multi-step prospective optimization deduction, and generating a coordinated scheduling instruction set of a plurality of scheduling periods in the future; and carrying out safety constraint verification on the coordinated scheduling instruction set, generating a final executable scheduling control instruction sequence according to a verification result, and issuing the final executable scheduling control instruction sequence to various controllable devices of the micro-grid system for execution.
2. The AI-based microgrid multi-source collaborative optimization scheduling method according to claim 1, wherein performing data cleaning and format unification processing on the original operation data set of the microgrid to form a standardized microgrid operation data set comprises: the standardized micro-grid operation data set comprises a time stamp, a power generation power sequence of each distributed power supply, a state of charge sequence of an energy storage device, an adjustable load power sequence and an interactive constraint upper limit sequence; Identifying and correcting abnormal values in the power generation data, the state data, the operation data and the interactive constraint data, wherein the identification of the abnormal values is based on the statistical distribution of historical data to set a deviation threshold value, and the correction is completed by adopting adjacent effective data interpolation; clock synchronization and resampling are carried out on the corrected power generation data, state data, operation data and interaction constraint data according to a unified time reference, so that all data sequences have the same time resolution; And respectively carrying out normalization processing on the generated power data, the state of charge data, the power data and the constraint upper and lower limit data which are subjected to clock synchronization, and mapping the normalized data to the same numerical interval to form the standardized micro-grid operation data set.
3. The AI-based microgrid multi-source collaborative optimization scheduling method according to claim 2, wherein extracting key feature vectors characterizing multi-source collaborative relationships from the standardized microgrid operational dataset comprises: the key feature vector comprises a source load power matching degree feature, an energy storage stabilizing wave capability feature and a power trend feature interacted with a power grid under a multi-time scale; Calculating real-time difference values between the generated power sequence and the adjustable load power sequence under a plurality of preset time window scales to form a source load power deviation sequence with multiple time scales; Calculating a quantization index representing the energy storage system absorbing or releasing power in unit time to smooth source charge fluctuation capacity based on the charge state sequence and rated parameters of the energy storage device, and obtaining the energy storage stabilizing fluctuation capacity characteristic; Analyzing the relation between the generated power sequence, the adjustable load power sequence and the interactive constraint upper and lower limit sequences, and extracting a power change trend mode of the micro-grid for purchasing or selling electricity to the power grid in a dispatching period to obtain the power trend characteristics interacted with the power grid; and combining the source load power deviation sequence with multiple time scales, the energy storage stabilizing wave capability characteristic and the power trend characteristic interacted with the power grid to construct the key characteristic vector.
4. The AI-based microgrid multi-source collaborative optimization scheduling method of claim 3, wherein inputting the key feature vector into a pre-trained multi-source collaborative characterization neural network model, performing joint state encoding on the distributed power source, the energy storage device and the adjustable load, generating a collaborative state vector set comprising multi-source complementary characteristics and timing dependencies, comprises: The multi-source collaborative characterization neural network model comprises a characteristic interaction coding layer and a time sequence dependent coding layer; the feature interaction coding layer receives the key feature vectors, calculates dynamic association weights among different distributed power supplies, between the distributed power supplies and the stored energy and among source stores through an internal multi-head attention mechanism, and generates primary fusion features; The time sequence dependent coding layer receives the primary fusion characteristics, captures the evolution rule of the primary fusion characteristics in the time dimension through a circulating neural network structure in the primary fusion characteristics, and outputs a characteristic sequence containing time sequence context information; And pooling and compressing the characteristic sequence containing the time sequence context information, generating a state vector with a fixed dimension for each scheduling moment, and forming the collaborative state vector set by the state vectors at all scheduling moments.
5. The AI-based microgrid multi-source collaborative optimization scheduling method according to claim 4, wherein invoking a scheduling decision-making network model for multi-step prospective optimization deduction based on the collaborative state vector set generates a coordinated scheduling instruction set of a plurality of scheduling periods in the future, comprising: The coordination scheduling instruction set comprises a planned output instruction of each distributed power supply, a charge and discharge power instruction of the energy storage device and a regulation and control instruction of an adjustable load; Inputting the collaborative state vector set into the scheduling decision generation network model, wherein the scheduling decision generation network model is constructed based on a deep reinforcement learning framework; the scheduling decision generation network model takes current and historical collaborative state vectors as input to simulate the dynamic evolution process of the micro-grid system in a preset multi-step scheduling time domain; In each step of evolution simulation, the scheduling decision generation network model outputs temporary scheduling actions corresponding to scheduling moments, wherein the temporary scheduling actions comprise the power drawn out of various distributed power supplies, the power to be charged and discharged for an energy storage device and the amount to be adjusted for adjustable loads; And collecting temporary scheduling actions at all scheduling moments in the simulation process, and carrying out optimization adjustment on the temporary scheduling action sequence by combining with evaluation of the overall running state of the system at the end of the simulation to form a final coordinated scheduling instruction set.
6. The AI-based microgrid multi-source collaborative optimization scheduling method according to claim 5, wherein the scheduling decision generating network model is constructed based on a deep reinforcement learning framework, comprising: Defining a state space of the scheduling decision generating network model as a space formed by the collaborative state vector set; Defining an action space of the scheduling decision generation network model as an allowable adjustment quantity range of each controllable device in a single scheduling period; Defining a reward function, wherein the reward function is used for carrying out instant evaluation on the temporary scheduling action in the simulation process, and calculating out-of-limit penalty based on source load matching degree, energy storage state health degree and interaction with a power grid in a simulation state; and performing offline training and online fine tuning on the scheduling decision generating network model through a large amount of historical data and simulation data, so that the scheduling decision generating network model learns to select a scheduling action sequence capable of maximizing cumulative rewards in the state space.
7. The AI-based microgrid multi-source co-optimal scheduling method of claim 6, wherein said performing a security constraint check on said coordinated scheduling instruction set comprises: The method comprises the steps of obtaining the upper limit and the lower limit of output and the climbing rate limit of each distributed power supply from the standardized micro-grid operation data set, the charge state safety range of an energy storage device, the charge-discharge power limit, the adjustable load adjusting range and the adjustable rate limit, and the power exchange limit of a micro-grid and an upper-level grid connection point as a physical safety constraint set; Comparing the planned output instruction, the charge-discharge power instruction and the regulation instruction in the coordinated scheduling instruction set with the physical safety constraint set time-period-by-time period; Checking whether the planned output instruction meets the upper and lower limits of output and the climbing speed limit, checking whether the charge-discharge power instruction causes the energy storage charge state to exceed a safe range or the charge-discharge power to exceed the limit, checking whether the regulation instruction exceeds the regulation range and the speed limit of an adjustable load, and checking whether the net exchange power with a power grid exceeds the exchange limit of a connection point; And recording all instructions which do not meet the physical security constraint set and specific constraint contents of the violations of the instructions, and generating a constraint violation report.
8. The AI-based microgrid multi-source collaborative optimization scheduling method according to claim 7, wherein the generating a final executable scheduling control instruction sequence according to the verification result comprises: if the constraint violation report is empty, indicating that the coordinated scheduling instruction set completely meets all physical security constraints, directly arranging the coordinated scheduling instruction set into the scheduling control instruction sequence according to the time sequence; If the constraint violation report is not empty, starting an instruction correction flow, namely correcting each instruction violating the constraint in the report to meet the corresponding physical security constraint by using a minimum adjustment principle on the premise of meeting the original optimization target of the instruction violating the constraint; Re-integrating the corrected instruction with the original instruction which does not violate the constraint to form a corrected dispatching instruction set which meets all safety constraints; And carrying out sequential logic consistency check on the instructions in the corrected dispatching instruction set to ensure that no time or logic conflict exists among the instructions, and generating the final executable dispatching control instruction sequence through time sequence arrangement after checking.
9. The AI-based microgrid multi-source co-optimization scheduling method of claim 8, wherein the various controllable devices issued to the microgrid system perform comprising: Packaging the scheduling control instruction sequence into a data packet with a specified format according to a communication protocol; The packaged data packets are respectively sent to a corresponding distributed power supply controller, an energy storage system manager and a load aggregation controller through a communication network of the micro-grid energy management system; the distributed power supply controller, the energy storage system manager and the load aggregation controller receive and analyze the data packet, extract respective corresponding control instructions, convert the control instructions into driving signals or switching control signals of the bottom-layer power electronic equipment, and execute scheduling instructions.
10. The AI-based microgrid multi-source collaborative optimization scheduling method according to claim 9, wherein the analyzing the relationship between the generated power sequence, the adjustable load power sequence and the interactive constraint upper and lower limit sequences, extracting a power change trend pattern of the microgrid for purchasing or selling electricity to a power grid in a scheduling period, comprises: establishing a real-time net power calculation model of the generated power sequence and the adjustable load power sequence, and generating a net load power sequence; The net load power sequence is compared with the interactive constraint upper and lower limit sequences point by point, and a time interval in which the net power exceeds a constraint boundary is identified; Calculating the deviation degree and the deviation duration time of the net power relative to the constraint boundary in the identified time interval; Dividing the severity level of the power out-of-limit event based on the combined characteristics of the degree of deviation and the duration of the deviation; extracting the occurrence frequency and distribution rule of power out-of-limit events of different severity levels in each scheduling period; And integrating the overall fluctuation form and the power out-of-limit event characteristics of the payload power sequence, and constructing a multidimensional characteristic vector representing the power change trend mode.

Description

AI-based micro-grid multi-source collaborative optimization scheduling method Technical Field The invention relates to the technical field of micro-grid operation control, in particular to an AI-based micro-grid multi-source collaborative optimization scheduling method. Background The micro-grid is an autonomous system integrating various heterogeneous resources such as distributed photovoltaic, fans, energy storage, adjustable loads and the like, and the efficient and stable operation of the micro-grid depends on the cooperative scheduling of internal multi-type devices. The existing scheduling method is mainly divided into two types of centralized optimization and decomposition coordination. The centralized optimization method generally builds a unified mixed integer programming model, and integrates and solves the operation constraint and the system security constraint of various devices. However, the micro-grid operation data has various sources, different formats and different space-time scales, and the centralized model is difficult to effectively process the multi-source heterogeneity, so that the description accuracy of the real operation state of the system is insufficient. The decomposition coordination method decomposes the problem into a plurality of sub-problems according to the device type or the region, and coordinates the problems through alternate iteration. The method reduces the complexity of the problem to a certain extent, but a coordination mechanism among the sub-problems often depends on fixed boundary conditions or simple price signals, and a deep dynamic complementary relation between the time sequence coupling of the distributed power supply output and the energy storage charging and discharging energy and the load adjustable potential cannot be fully described. In the prior art, an effective joint characterization means is lacking for complex, nonlinear and time-sequence-dependent cooperative relations among multi-source equipment, scheduling instructions are mostly generated based on current or short-term states, prospective overall deduction for evolution of operation states in a plurality of subsequent time periods is lacking, when renewable energy source severely fluctuates, scheduling instructions are easy to look short, and operation risks such as frequent charging and discharging of an energy storage device or power overrun interaction with an upper power grid are caused. Therefore, how to accurately sense and encode the overall collaborative operation state of the system from the multi-source heterogeneous data and generate an optimal scheduling instruction with prospective and coordination based on the collaborative operation state is a key problem to be solved in the current micro-grid operation control field. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides an AI-based micro-grid multi-source collaborative optimization scheduling method. In order to achieve the purpose, the invention adopts the following technical scheme that the micro-grid multi-source collaborative optimization scheduling method based on AI comprises the following steps: Acquiring power generation data of a distributed power supply, state data of an energy storage device, operation data of an adjustable load and interactive constraint data of an upper power grid in a micro-grid system, and generating a multi-source heterogeneous micro-grid original operation data set; Performing data cleaning and format unification processing on the original operation data set of the micro-grid to form a standardized micro-grid operation data set, and extracting key feature vectors representing multi-source cooperative relationships from the standardized micro-grid operation data set; inputting the key feature vector into a pre-trained multi-source collaborative characterization neural network model, and carrying out joint state coding on the distributed power supply, the energy storage device and the adjustable load to generate a collaborative state vector set containing multi-source complementary characteristics and time sequence dependency relations; based on the collaborative state vector set, calling a scheduling decision generation network model to carry out multi-step prospective optimization deduction, and generating a coordinated scheduling instruction set of a plurality of scheduling periods in the future; and carrying out safety constraint verification on the coordinated scheduling instruction set, generating a final executable scheduling control instruction sequence according to a verification result, and issuing the final executable scheduling control instruction sequence to various controllable devices of the micro-grid system for execution. As a further aspect of the present invention, the performing data cleaning and format unification processing on the original operation data set of the micro-grid to form a standardized micro-grid operation data set inclu