CN-121618693-B - Emergency power supply system multi-source collaborative scheduling optimization method based on deep learning

CN121618693BCN 121618693 BCN121618693 BCN 121618693BCN-121618693-B

Abstract

The invention provides a deep learning-based multi-source collaborative scheduling optimization method of an emergency power supply system, which relates to the technical field of power system scheduling and comprises the steps of obtaining real-time running states and load demand data of a plurality of power supplies, obtaining an initial scheduling sequence through calculation of supply and demand power constraint and running constraint, carrying out wavelet transformation decomposition on the real-time running states according to time scales, constructing an energy-time characteristic spectrum, calculating a comprehensive complementary coefficient through complementary discrimination space mapping, generating an optimized weight matrix, correcting the initial scheduling sequence according to the optimized weight matrix to form a collaborative scheduling instruction, constructing a fault evolution path spectrum based on a fault sample set, calculating a risk assessment value, generating a priority order, constructing an execution time gradient sequence, generating a switching control sequence and executing. The invention improves the response speed of emergency power supply and the precision of multi-source collaborative scheduling, and reduces the risk of system faults.

Inventors

LIAO LIN
XU MENGYANG
WANG XIN

Assignees

北京西普霍斯科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251202

Claims (10)

1. The multi-source collaborative scheduling optimization method for the emergency power supply system based on deep learning is characterized by comprising the following steps of: Acquiring real-time running states and load demand data of a plurality of power supplies; Based on the real-time running state and the load demand data, obtaining an initial scheduling sequence by calculating supply and demand power constraint and running constraint; Performing wavelet transformation decomposition on the real-time running state according to time scale, constructing an energy-time characteristic map based on the decomposition result, and calculating the comprehensive complementary coefficient between power supplies through complementary discrimination space mapping to generate an optimized weight matrix; correcting the initial scheduling sequence according to the optimized weight matrix to form a cooperative scheduling instruction; Constructing a characteristic propagation link based on a fault sample set in historical operation data, establishing a fault evolution path map, calculating a risk evaluation value based on the fault evolution path map, and generating a priority order of the scheduling instructions according to the risk evaluation value; Dividing the collaborative scheduling instruction into a plurality of instruction gradient groups according to the priority order, constructing an execution time gradient sequence, forming a self-adaptive response time sequence frame, and generating a switching control sequence of a power supply; and executing the switching control sequence, recording an execution result, and performing iterative optimization through the execution result.
2. The method of claim 1, wherein decomposing the real-time operation state by wavelet transformation on a time scale, constructing an energy-time feature map based on the decomposition result, calculating a comprehensive complementary coefficient between power supplies by complementary discrimination space mapping, and generating an optimized weight matrix comprises: Dividing the real-time running state into a plurality of time scale levels according to the length of a time window, carrying out wavelet transformation decomposition on the running state data in each time scale level, and distinguishing a state fluctuation component and a state trend component through a preset frequency threshold value to obtain a state decomposition result of each time scale level; based on the state decomposition result, constructing an energy-time two-dimensional characteristic map of the power supply under each time scale level, calculating the gradient change rate of adjacent data points in the time axis direction, calculating the distribution entropy value of each energy interval in the energy axis direction, and combining to generate a dynamic change characteristic vector; Mapping the dynamic change feature vector of each pair of power supplies under the same time scale level to a complementary discrimination space, and calculating the cosine value of the included angle of the mapped vector to obtain a complementary degree quantization index; Determining fusion weights based on the load fluctuation intensity and the historical scheduling deviation of the corresponding time period of each time scale level, and multiplying and accumulating the complementation degree quantization indexes of each pair of power supplies under all the time scale levels and the fusion weights to obtain comprehensive complementation coefficients; and forming an optimized weight matrix by the comprehensive complementary coefficients between every two power supplies.
3. The method of claim 1, wherein constructing a feature propagation link based on the set of fault samples in the historical operating data, and wherein constructing a fault evolution path graph comprises: Extracting a fault sample set marking the fault occurrence time from the historical operation data, tracing back a preset time window forward by taking the fault occurrence time as the termination time to obtain an operation parameter sequence in a corresponding time window, and identifying upstream and downstream equipment directly connected with the fault equipment according to the operation parameter sequence to obtain an associated equipment set; Aiming at the operation parameter sequence of each device in the associated device set, calculating a correlation coefficient of the operation parameter sequence of the fault device, determining a time delay parameter based on a time difference corresponding to the maximum value of the correlation coefficient, constructing a feature tuple by utilizing the operation parameter sequence, the time delay parameter and the correlation coefficient, and incrementally sequencing the feature tuple according to the time delay parameter to obtain a feature propagation link; And extracting commonly contained equipment parameter combinations to determine key nodes by utilizing the characteristic propagation links of all the fault samples, connecting all the characteristic propagation links through the key nodes to form a directed graph, taking the equipment state as a node, transferring the state as an edge, and taking the occurrence frequency of the characteristic propagation links in the fault sample set as an edge weight to form a fault evolution path graph.
4. The method of claim 1, wherein calculating a risk assessment value based on the fault evolution path graph, generating a priority order of the scheduling instructions according to the risk assessment value comprises: Acquiring a node set and directed edge weights in a fault evolution path map, constructing a failure propagation matrix, and calculating to obtain node importance based on the degree centrality of the nodes and the number of paths among the nodes; Collecting real-time operation parameters, calculating similarity values of node characteristics in the real-time operation parameters and a fault evolution path map, extracting nodes with the similarity values exceeding a preset similarity threshold value to form a primary activation node set, carrying out propagation calculation on the primary activation node set by using a failure propagation matrix to obtain a secondary activation node set, and forming an activation node sequence by the primary activation node set and the secondary activation node set; Multiplying the node importance, the corresponding value of the failure propagation matrix and the similarity value by aiming at each node in the activated node sequence to obtain node failure probability, establishing a corresponding relation between node failure and the influence degree of power supply load, calculating load influence loss caused by each node failure, and calculating to obtain a power supply area risk assessment value based on the node failure probability and the load influence loss; and generating the priority order of the dispatching instructions according to the power supply area risk assessment value.
5. The method of claim 1, wherein the adaptive response timing framework comprises: Constructing a priority distribution curve according to the priority values of each scheduling instruction in the priority sequence, extracting the inflection point position with the largest curvature change based on the priority distribution curve, determining a time gradient dividing limit, and dividing the scheduling instructions into a plurality of instruction gradient groups; For each instruction gradient group, calculating the state stability and the switching preparation time of a power supply in the group, and carrying out weighted summation on the switching preparation time by taking the state stability as a weight to obtain group execution time delay, and constructing an execution time gradient sequence according to each group execution time delay; establishing a coupling relation matrix between adjacent gradients in the execution time gradient sequence, recording the mutual influence coefficient of power supply switching between the execution time gradients, calculating the minimum safety interval between the adjacent execution time gradients based on the coupling relation matrix, and performing verification adjustment on the execution time gradient sequence; when the deviation degree of the load demand data relative to the historical mean value exceeds a preset deviation boundary value, mapping the deviation degree into a frame adjustment factor, updating the group execution time delay according to the frame adjustment factor, and reconstructing an execution time sequence to form an adaptive response time sequence frame.
6. The method of claim 5, wherein mapping the deviation of the load demand data from the historical mean to a frame adjustment factor when the deviation exceeds a preset deviation boundary value, updating the group execution delay and reconstructing the execution time sequence according to the frame adjustment factor, forming the adaptive response timing framework comprises: Acquiring statistical distribution characteristics of historical load demand data, calculating deviation degree of the load demand data relative to a historical mean value, and mapping the deviation degree into a frame adjustment factor according to a preset interval when the deviation degree exceeds a preset deviation boundary value; Calculating the product of the frame adjustment factors and the state stability of the power supply in each instruction gradient group to obtain the execution time delay of the adjustment group, and reconstructing an execution time gradient sequence according to the execution time delay of the adjustment group; based on the reconstructed execution time gradient sequence, extracting a coupling relation matrix between adjacent execution time gradients, and calculating the minimum safety interval between the adjacent execution time gradients according to the mutual influence coefficient of power supply switching in the coupling relation matrix; and checking and adjusting the execution time gradient sequence by taking the minimum safety interval as a constraint condition to form an adaptive response time sequence frame considering the dynamic change of the load.
7. The method of claim 5, wherein generating a switching control sequence for a power supply comprises: Establishing a power supply dynamic response characteristic curve in each gradient interval of the execution time gradient sequence, and calculating the climbing rate and the stabilizing time of the power supply based on the dynamic response characteristic curve to obtain dynamic switching constraint of each power supply; aiming at the scheduling instruction combination in each time gradient, constructing a power supply group switching topological graph, calculating the transmission delay and the power fluctuation amplitude of each switching path in the topological graph, and selecting the optimal switching path with the minimum transmission delay and limited power fluctuation; And generating a switching control sequence comprising a switching time sequence, a switching object and power distribution according to the dynamic switching constraint and the optimal switching path.
8. Deep learning-based emergency power supply system multi-source collaborative scheduling optimization system for implementing the method of any of the preceding claims 1-7, comprising: the first unit is used for acquiring real-time running states and load demand data of a plurality of power supplies; The second unit is used for obtaining an initial scheduling sequence by calculating supply and demand power constraint and operation constraint based on the real-time operation state and load demand data; The third unit is used for carrying out wavelet transformation decomposition on the real-time running state according to the time scale, constructing an energy-time characteristic spectrum based on the decomposition result, calculating the comprehensive complementary coefficient between power supplies through complementary discrimination space mapping, and generating an optimized weight matrix; the fourth unit is used for correcting the initial scheduling sequence according to the optimized weight matrix to form a cooperative scheduling instruction; the fifth unit is used for constructing a characteristic propagation link based on a fault sample set in the historical operation data, establishing a fault evolution path map, calculating a risk evaluation value based on the fault evolution path map, and generating a priority order of the scheduling instruction according to the risk evaluation value; A sixth unit, configured to divide the co-scheduling instruction into a plurality of instruction gradient groups according to the priority order, construct an execution time gradient sequence, form an adaptive response timing frame, and generate a switching control sequence of the power supply; and a seventh unit, configured to execute the switching control sequence and record an execution result, and perform iterative optimization according to the execution result.
9. An electronic device, comprising: A processor; A memory for storing processor-executable instructions; Wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any of claims 1 to 7.
10. A computer readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the method of any of claims 1 to 7.

Description

Emergency power supply system multi-source collaborative scheduling optimization method based on deep learning Technical Field The invention relates to the technical field of power system dispatching, in particular to an emergency power supply system multi-source collaborative dispatching optimization method based on deep learning. Background With the rapid development of socioeconomic and the continuous increase of power demand, emergency power supply systems play an increasingly important role in protecting critical infrastructure, important sites and power supply in emergency events. The emergency power supply system is generally composed of a diesel generator, an energy storage battery, photovoltaic power generation, wind power generation and other power supply sources, and how to realize efficient collaborative scheduling of the multi-source power supply equipment, so that quick response and stable power supply under emergency conditions are ensured, and the emergency power supply system is a key technical problem to be solved in the current power system field. The traditional emergency power supply scheduling method mainly relies on preset rules and experience models, and the switching and power distribution of the power supply are performed in a manual or semi-automatic mode. These methods can basically meet the demands when facing simple scenes, but the conventional methods gradually expose many limitations as the scale of emergency power supply systems is enlarged and the types of power supplies are diversified. In recent years, with the development of artificial intelligence and deep learning technologies, researchers have begun to attempt to apply intelligent algorithms to the dispatch optimization of emergency power supply systems to improve the level of intelligence and response efficiency of the systems. However, the prior art still has some defects and shortcomings, the prior emergency power supply scheduling method focuses on the power balance problem under a single time scale, lacks deep analysis on the operation characteristics of different power supplies under a plurality of time scales, is difficult to fully mine complementary characteristics among the power supplies, lacks prediction and evaluation capability of potential fault risks of a system when scheduling decisions, lacks a dynamic tracking and analysis mechanism of a fault evolution process, cannot identify key risk nodes possibly causing system failure in advance, affects power supply reliability and system safety, and lacks self-adaptive adjustment capability, fails to dynamically adjust the execution sequence and response time of instructions according to actual operation states and risk levels, so that key instruction execution delay is caused under a high risk scene. Disclosure of Invention The embodiment of the invention provides a multi-source collaborative scheduling optimization method for an emergency power supply system based on deep learning, which can solve the problems in the prior art. In a first aspect of the embodiment of the present invention, a method for optimizing multi-source collaborative scheduling of an emergency power supply system based on deep learning is provided, including: Acquiring real-time running states and load demand data of a plurality of power supplies; Based on the real-time running state and the load demand data, obtaining an initial scheduling sequence by calculating supply and demand power constraint and running constraint; Performing wavelet transformation decomposition on the real-time running state according to time scale, constructing an energy-time characteristic map based on the decomposition result, and calculating the comprehensive complementary coefficient between power supplies through complementary discrimination space mapping to generate an optimized weight matrix; correcting the initial scheduling sequence according to the optimized weight matrix to form a cooperative scheduling instruction; Constructing a characteristic propagation link based on a fault sample set in historical operation data, establishing a fault evolution path map, calculating a risk evaluation value based on the fault evolution path map, and generating a priority order of the scheduling instructions according to the risk evaluation value; Dividing the collaborative scheduling instruction into a plurality of instruction gradient groups according to the priority order, constructing an execution time gradient sequence, forming a self-adaptive response time sequence frame, and generating a switching control sequence of a power supply; and executing the switching control sequence, recording an execution result, and performing iterative optimization through the execution result. In an alternative embodiment, performing wavelet transform decomposition on the real-time running state according to time scale, constructing an energy-time characteristic spectrum based on the decomposition result, calculating a compr