CN-122022524-A - Power grid material emergency dispatching method based on machine learning

CN122022524ACN 122022524 ACN122022524 ACN 122022524ACN-122022524-A

Abstract

The invention discloses a power grid material emergency dispatching method based on machine learning, which comprises the steps of obtaining an updated material delivery sequence by training a neural network model based on geographical position information and disaster range boundary data of a power grid fault point and combining road traffic conditions, warehouse distribution and material types, obtaining a preliminary path option by adopting a reinforcement learning method to conduct road network planning simulation based on long-distance transportation data in the updated material delivery sequence, extracting an alternative starting point from the warehouse distribution when route interruption in the preliminary path option is detected, generating an adjusted long-distance transportation scheme by reinforcement learning, carrying out cross verification based on transportation cost indexes of the adjusted long-distance transportation scheme through a neural network model and a demand resource competition relationship, reallocating the material types when allocation conflict is detected, and carrying out feedback circulation processing by combining driver fatigue and vehicle load constraint to obtain an optimized material allocation and path scheme.

Inventors

XIE YUWEI
JIANG YUAN
WANG YANNI
Ji Xuanru
LI QILING
WANG MENGJUN
DENG YONG
DONG ZHAOLIN

Assignees

国网重庆市电力公司物资分公司
重庆码投科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (7)

1. The power grid material emergency dispatching method based on machine learning is characterized by comprising the following steps of: based on the geographical position information of the power grid fault point and disaster range boundary data, combining the road traffic condition, warehouse distribution and material types, training by adopting a neural network model to obtain an updated material delivery sequence; Based on the long-distance transportation data in the updated material delivery sequence, performing road network planning simulation by adopting a reinforcement learning method to obtain a preliminary path option for balancing transportation time and cost; when the route interruption in the preliminary path option is detected, extracting an alternative starting point from warehouse distribution, and generating an adjusted long-distance transportation scheme by adopting a reinforcement learning iterative optimization process; And based on the transport cost index of the adjusted long-distance transport scheme, cross verification is carried out through a neural network model and a competition relationship of required resources, material types are redistributed when distribution conflict is detected, and the optimized material distribution and path scheme is obtained by adopting reinforcement learning feedback circulation processing in combination with driver fatigue and vehicle load constraint.
2. The machine learning-based power grid material emergency dispatching method of claim 1, wherein the process of training the updated material arrival sequence by adopting the neural network model based on the geographical position information of the power grid fault point and disaster range boundary data and combining the road traffic condition, warehouse distribution and material category comprises the following steps: Acquiring initial disaster area description based on geographical position information of a power grid fault point and disaster range boundary data; integrating the road traffic real-time condition, the warehouse distribution position and the emergency response priority in the initial disaster area description by adopting a machine learning algorithm to obtain an integrated variable set; based on the integrated variable set, adding emergency material category variables and demand emergency degree indexes to form a complete input data set; Inputting the complete input data set into a neural network model for training, and obtaining a preliminary matching result after training convergence; Generating an initial optimal material matching scheme based on the initial matching result corresponding to a plurality of demand places; When the disaster range boundary expansion caused by the newly added power grid fault point is detected, recalculating a material distribution scheme by adopting a neural network model based on the priority distribution sequence and the demand resource competition relation data in the initial optimal material matching scheme, and obtaining an updated material delivery sequence.
3. The machine learning based power grid material emergency dispatch method of claim 2, wherein recalculating the material allocation scheme using a neural network model to obtain an updated material arrival sequence comprises: Acquiring priority allocation sequence and required resource competition relationship data based on the initial optimal material matching scheme to form an expansion judgment basis; When a newly added grid fault point is detected, based on the expansion judgment basis, integrating historical fault data from a preset historical database, and determining disaster range boundary expansion description; activating a real-time adjustment mechanism based on the disaster range boundary extension description, adopting a multi-layer neural network model to process warehouse distribution positions, road traffic conditions and required emergency indexes, and recalculating a material distribution scheme; Based on the recalculated material distribution scheme, emergency response priority and dynamic distribution adjustment information are obtained to form updated distribution details; and based on the updated allocation details, determining an updated material delivery sequence by combining the required resource competition relationship data.
4. The machine learning-based power grid material emergency dispatching method of claim 1, wherein the process of obtaining preliminary path options for balancing transportation time and cost by performing road network planning simulation by adopting a reinforcement learning method based on long-distance transportation data in the updated material arrival sequence comprises the following steps: based on the updated material delivery sequence, extracting long-distance transportation data, and combining a preset resource priority list to form a preliminary transportation data set; Processing the preliminary transportation data set by adopting a reinforcement learning method, and constructing a road network planning simulation scene among the dispatching nodes; evaluating a path selection strategy through an environmental rewarding function, and simulating the interactive relation between the transportation time cost and the logistics transportation cost by combining dynamic road condition data; Based on the simulation result of the interaction relation, generating a path simulation result by iteratively calculating the weight of the balance time cost and the logistics cost; when road network congestion is detected, adjusting node priority based on the path simulation result, and generating an optimized path set after reordering; and carrying out cost balance evaluation on the optimized path set to obtain a preliminary path option for balancing transportation time and cost.
5. The machine learning based power grid material emergency dispatch method of claim 1, wherein when route disruption in the preliminary path option is detected, extracting an alternative starting point from a warehouse distribution, generating an adjusted long distance transportation scheme using a reinforcement learning iterative optimization process comprises: extracting alternative starting point positions from warehouse distribution based on the detected route interruption information to form an interruption response data set; integrating standby resource variables and matching resource priorities based on the interrupt response data set to obtain a resource integration description; processing the resource integration description by adopting a node load evaluation method to obtain a load balancing index; based on the load balancing index, a dynamic adjustment path is determined by inputting a node state and an action space output adjustment strategy; and based on the dynamic adjustment path, combining with cost weight iterative computation to obtain an adjusted long-distance transportation scheme.
6. The machine learning based power grid material emergency dispatching method according to claim 1, wherein the process of performing cross-validation with a demand resource competition relationship through a neural network model based on the transportation cost index of the adjusted long-distance transportation scheme, and redistributing material types when the distribution conflict is detected comprises: based on the transportation cost index of the long-distance transportation scheme, a neural network model is adopted to process the competition relationship of the demand resources, and a competition evaluation result is obtained; cross-verifying the competition evaluation result and the priority allocation sequence, triggering standby path evaluation when a conflict is detected, and generating path adjustment description; Integrating the resource inventory verification variables based on the path adjustment description, and obtaining inventory availability indexes through threshold comparison; and based on the inventory availability indexes, matching and sorting are carried out according to the material category priority, and the distribution proportion is adjusted when the materials collide, so that a final matching scheme after the conflict resolution is obtained.
7. The machine learning based power grid material emergency dispatch method of claim 6, wherein the process of deriving an optimized material distribution and path scheme based on the final matching scheme after conflict resolution in combination with driver fatigue and vehicle load constraints comprises: Acquiring constraint data of fatigue limit of a driver and load limit of a vehicle based on road network planning details in the final matching scheme; processing the constraint data by adopting reinforcement learning feedback loop, and generating a constraint optimization result through a state-action-rewarding iterative optimization process; Based on the constraint optimization result, combining the inventory verification indexes to perform threshold comparison, and triggering priority ordering adjustment when the indexes are lower than the threshold; and based on the adjusted sorting result, scanning the standby route nodes in the road network planning, and integrating emergency allocation optimization to obtain an executable dynamic adjustment path.

Description

Power grid material emergency dispatching method based on machine learning Technical Field The invention belongs to the technical field of resource allocation optimization, and particularly relates to a power grid material emergency dispatching method based on machine learning. Background The emergency dispatching of the power grid supplies is a key link in emergency guarantee of the power system, and particularly after natural disasters or sudden faults occur, whether emergency supplies can be rapidly and accurately dispatched to disaster areas or not is directly related to the recovery speed of the power grid and the stable social operation. At present, the field mainly depends on manual experience and preset rules to make scheduling decisions, and lacks dynamic response capability to complex and changeable environments. When the traditional method is used for processing multiple constraints such as multipoint faults, road network congestion, resource competition and the like, global optimization is difficult to achieve in a limited time, and problems such as partial area material backlog, insufficient supply of other areas and the like are easily caused, so that the overall rescue efficiency is affected. However, the existing scheduling method still has significant disadvantages when facing the dynamically extended disaster range, the real-time changing road condition and the multi-target resource allocation conflict. The method has the advantages that when the fault point is newly increased or the range is enlarged, an effective real-time rescheduling mechanism is lacked, the material distribution sequence is difficult to quickly adjust, the coordination between path planning and resource distribution is lacked in a long-distance transportation scene, the scheme is often invalid due to path interruption or resource competition, and dynamic constraints in actual transportation, such as fatigue of a driver, load of a vehicle and the like, are not fully considered, so that the performability and robustness of the scheduling scheme are affected. These limitations make the existing system have low scheduling accuracy and delayed response when dealing with complex emergency scenes, and it is difficult to realize overall optimization of material distribution and path planning. Disclosure of Invention In order to solve the technical problems, the invention provides a power grid material emergency dispatching method based on machine learning, which aims to solve the problems existing in the prior art. In order to achieve the above purpose, the invention provides a power grid material emergency dispatching method based on machine learning, comprising the following steps: Based on the geographical position information of the power grid fault point and disaster range boundary data, combining the road traffic condition, warehouse distribution and material types, training an updated material delivery sequence by adopting a neural network model; Based on the long-distance transportation data in the updated material delivery sequence, performing road network planning simulation by adopting a reinforcement learning method to obtain a preliminary path option for balancing transportation time and cost; when the route interruption in the preliminary path option is detected, extracting an alternative starting point from warehouse distribution, and generating an adjusted long-distance transportation scheme by adopting a reinforcement learning iterative optimization process; And based on the transport cost index of the adjusted long-distance transport scheme, cross verification is carried out through a neural network model and a competition relationship of required resources, material types are redistributed when distribution conflict is detected, and the optimized material distribution and path scheme is obtained by adopting reinforcement learning feedback circulation processing in combination with driver fatigue and vehicle load constraint. Optionally, based on the geographical location information of the power grid fault point and the boundary data of the disaster range, combining the road traffic condition, warehouse distribution and material types, the process of training the updated material delivery sequence by using the neural network model comprises the following steps: Acquiring initial disaster area description based on geographical position information of a power grid fault point and disaster range boundary data; integrating the road traffic real-time condition, the warehouse distribution position and the emergency response priority in the initial disaster area description by adopting a machine learning algorithm to obtain an integrated variable set; based on the integrated variable set, adding emergency material category variables and demand emergency degree indexes to form a complete input data set; Inputting the complete input data set into a neural network model for training, and obtaining a preliminary matchi