CN116432968A - Energy storage planning method and equipment for power system - Google Patents

Energy storage planning method and equipment for power system Download PDF

Info

Publication number
CN116432968A
CN116432968A CN202310420259.0A CN202310420259A CN116432968A CN 116432968 A CN116432968 A CN 116432968A CN 202310420259 A CN202310420259 A CN 202310420259A CN 116432968 A CN116432968 A CN 116432968A
Authority
CN
China
Prior art keywords
energy storage
scene
model
power system
scene set
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310420259.0A
Other languages
Chinese (zh)
Inventor
陈玥
李力
杨银国
骆晓明
陆秋瑜
林旭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Power Grid Co Ltd
Electric Power Dispatch Control Center of Guangdong Power Grid Co Ltd
Original Assignee
Guangdong Power Grid Co Ltd
Electric Power Dispatch Control Center of Guangdong Power Grid Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Power Grid Co Ltd, Electric Power Dispatch Control Center of Guangdong Power Grid Co Ltd filed Critical Guangdong Power Grid Co Ltd
Priority to CN202310420259.0A priority Critical patent/CN116432968A/en
Publication of CN116432968A publication Critical patent/CN116432968A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06312—Adjustment or analysis of established resource schedule, e.g. resource or task levelling, or dynamic rescheduling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/04—Power grid distribution networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)

Abstract

The invention discloses an energy storage planning method and equipment of a power system, wherein the method comprises the steps of generating an extreme event scene set and a source load fluctuation scene set according to scene data of the power system; according to different key type operation scenes, driving physical elements of the power system by deep neural network data to obtain an element fault probability model, and performing Monte Carlo scene simulation on the element fault probability model and an extreme event scene set to generate a complex fault scene set; performing cluster analysis on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set; and constructing an energy storage planning double-layer model according to the energy storage configuration investment and the energy storage scheduling operation, and solving the energy storage planning double-layer model according to a typical key scene set to generate an energy storage planning scheme. According to the embodiment, an energy storage double-layer intelligent planning scheme is accurately formulated, energy storage is efficiently scheduled, the recovery efficiency of the power system is improved, and the elasticity level of the power system is improved.

Description

Energy storage planning method and equipment for power system
Technical Field
The invention relates to the field of safety planning operation of power systems, in particular to an energy storage planning method and equipment of a power system.
Background
With the continuous rising of the new energy duty ratio, the power system has higher requirements on friendly access of new energy, and the novel power system construction should be capable of realizing large-scale new energy grid connection and ensuring safe and stable operation of the system. However, the pulling-out limit caused by the wind power dip in the Sichuan high Wen Xiandian and northeast region gives a warning to the construction of the novel power system, which is not only limited to the absorption of new energy and the related stability improvement, but also has the capability of actively coping with extreme events with small probability and high loss, resists, responds and adapts to extreme natural disasters to a certain extent, and realizes quick recovery therefrom. The method ensures that the novel power system can effectively resist extreme events before disaster and coordinate the rapid recovery of the energy storage resource acceleration system after disaster, and is urgent to develop the research on the intelligent energy storage planning method facing the elastic lifting of the system.
Due to the access of new energy, demand response and load fluctuation caused by flexible load, the system cannot fully consider the fluctuation of the system under the conventional scene and the uncertainty of the extreme scene in the process of elastic lifting construction, and cannot pertinently plan and deploy energy storage, so that the influence of the extreme event is reduced while the system fluctuation is dealt with. Meanwhile, because the system power-energy fluctuation operation working conditions are various in extreme event scenes and conventional scenes, the scene number is continuously increased along with the system scale, typical scenes are difficult to obtain effectively, and efficient scheduling energy storage cannot be realized, so that the recovery efficiency of the power system is low.
Disclosure of Invention
The invention provides an energy storage planning method and equipment for an electric power system, which can accurately formulate an energy storage double-layer intelligent planning scheme, efficiently schedule energy storage, improve the recovery efficiency of the electric power system and improve the elasticity level of the electric power system.
In order to solve the above technical problems, an embodiment of the present invention provides an energy storage planning method for an electric power system, including:
generating an extreme event scene set and a source load fluctuation scene set according to scene data of the power system; the scene data comprise extreme event data, new energy fluctuation data and load fluctuation data in a conventional scene;
according to different key type operation scenes, driving physical elements of the power system by deep neural network data to obtain an element fault probability model, and performing Monte Carlo scene simulation on the element fault probability model and an extreme event scene set to generate a complex fault scene set;
performing cluster analysis on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set;
and constructing an energy storage planning double-layer model according to the energy storage configuration investment and the energy storage scheduling operation, and solving the energy storage planning double-layer model according to a typical key scene set to generate an energy storage planning scheme.
By implementing the embodiment of the invention, the initial scene generation of new energy fluctuation, load fluctuation and extreme event uncertainty under the conventional scene is considered; accurately describing action mechanisms of event scenes of different poles on element faults by adopting a data-driven method, acquiring an element fault rate model under an extreme event, acquiring a complex fault scene set under the extreme event by utilizing Monte Carlo simulation, acquiring key operation scenes from a large number of complex fault scene sets and source load fluctuation scene sets by adopting cluster analysis, and forming a typical key scene set; an energy storage planning double-layer model is built, an upper-layer model is built to be an energy storage configuration investment master problem, a lower-layer model is set to be an energy storage scheduling operation sub-problem, and corresponding energy storage configuration and scheduling operation strategies are formulated for different typical key scene sets. The energy storage double-layer intelligent planning scheme is accurately formulated for elastic lifting of the novel power system, the energy storage configuration is guided to strengthen weak links, the energy storage is efficiently scheduled to improve the recovery efficiency of the power system, and the elasticity level of the power system is improved from two aspects of energy storage planning configuration and regulation and control operation.
As a preferred scheme, carrying out Monte Carlo scene simulation on the element fault probability model and the extreme event scene set to generate a complex fault scene set, wherein the complex fault scene set specifically comprises:
Inputting the extreme event scene into an element fault probability model to obtain the fault probability of each element;
according to a preset numerical range, simulating sampling by a Monte Carlo method to obtain a plurality of random numbers;
comparing each random number with each element fault probability respectively to obtain the state of each device;
and generating a complex fault scene set according to the states and the positions of the devices.
As a preferred scheme, according to different key type operation scenes, the physical elements of the power system are subjected to deep neural network data driving to obtain an element fault probability model, which is specifically as follows:
according to the operation scenes of different key types, obtaining corresponding meteorological factors under the operation scenes of different key types; the different key types of operation scenes comprise an extreme operation scene and a complex operation scene, and meteorological factors comprise wind speed, wind direction, temperature, humidity, illumination included angle and light intensity;
building a cyclic convolutional neural network model, inputting an extreme event data set into the cyclic convolutional neural network model for training, and obtaining an element fault probability model; the extreme event data set comprises corresponding historical equipment fault conditions and meteorological factors under different key type operation scenes.
By implementing the embodiment of the invention, the data driving method is utilized, the association relation between different types of scenes and element faults is characterized by adopting the deep neural network, the element fault probability model is established, and complex display expression of the different types of scenes and the element fault rate is avoided.
As a preferred scheme, the extreme event data set is input into a cyclic convolutional neural network model for training, and an element fault probability model is obtained, specifically:
inputting the extreme event data set into a cyclic convolutional neural network model for training, adjusting model parameters, drawing a learning curve in combination with a training process, setting a model loss function threshold value to determine a training termination condition, updating model parameters and optimal weights, and describing association relations between different complex extreme scenes and equipment element faults to obtain a trained element fault probability model; the model parameters comprise a learning rate, a small-batch sample scale parameter and a neural network scale parameter.
As a preferable scheme, a cyclic convolution neural network model is built, specifically:
designing a circulating layer of a circulating convolutional neural network model as a bidirectional circulating neural network, forward circulating to access the previous environmental factor to obtain forward circulating output, backward circulating to access the latter environmental factor to obtain backward circulating output, and carrying out transverse cascade connection on the forward circulating output and the backward circulating output to obtain circulating layer output;
The method comprises the steps of designing a convolutional layer of a cyclic convolutional neural network model as a CNN convolutional neural network, selecting a preset number of convolutional kernels with different sizes, and respectively carrying out convolutional operation on the output of the cyclic layer to obtain various convolutional output matrixes, wherein the number of columns of the convolutional kernels is the same as the number of columns of the output of the cyclic layer, and each step of convolutional operation comprises the following formula:
h c (i)＝f(W c ·h 2 (i:i+l-1)+b c )
hr c (i)＝relu(h c (i))
wherein h is c (i) For convolving the output matrix, hr c (i) For the i-th convolution vector obtained after convolution, l is the number of rows of the convolution kernel, i=1, 2,.. 2 (i: i+l-1) is a matrix formed by the i-th to i+l-1-th row vectors outputted from the cyclic layer, b c As a convolution operation bias term, relu is an activation function of the CNN convolution neural network;
designing a pooling layer of the cyclic convolutional neural network model as a maximum element pooling layer, selecting the maximum element in the output of the cyclic layer and the convolutional vector as a characteristic value, and cascading all the characteristic values to form a characteristic vector.
By implementing the embodiment of the invention, different feature vectors are obtained by setting convolution kernels with different sizes, and then the feature vectors are longitudinally cascaded to enrich scene environment features. The pooling layer further extracts text features, so that feature vector dimensions are reduced, calculated amount is reduced, and classification efficiency is improved.
As a preferred scheme, the complex fault scene set and the source load fluctuation scene set are subjected to cluster analysis to generate a typical key scene set, wherein the typical key scene set is specifically as follows:
Taking the complex fault scene set and the source load fluctuation scene set as training sets of a scene aggregation algorithm, selecting a plurality of typical fault scenes as scene clustering centers, solving the spatial distances between all the complex fault scenes and the initial scenes, classifying the complex fault scenes and the initial scenes into the nearest initial scenes, and obtaining scene sets of all categories; the initial scene comprises an extreme event scene set and a source load fluctuation scene set;
and updating each scene clustering center according to each category of scene set, and performing loop iteration until each scene clustering center meets the clustering requirement, thereby obtaining a typical key scene set.
As a preferred scheme, according to the scene data of the power system, an extreme event scene set and a source load fluctuation scene set are generated, specifically:
acquiring a source load fluctuation scene set according to new energy fluctuation data and load fluctuation data in a conventional scene in historical data of the power system;
and carrying out extreme event key parameter simulation on the extreme event data of the power system to generate an extreme event scene set.
As a preferred scheme, an energy storage planning double-layer model is built according to energy storage configuration investment and energy storage scheduling operation, and the method specifically comprises the following steps:
taking energy storage investment, load reduction and weak link reinforcement cost as a first objective function, taking allowable installation maximum power and capacity as first constraint conditions, taking energy storage power energy configuration and key equipment reinforcement measures as first decision variables, and building an upper model of energy storage configuration investment;
Constructing a lower model of energy storage scheduling operation by taking load reduction cost as a second objective function, energy storage scheduling operation constraint as a second constraint condition and energy storage scheduling strategy, line and unit recovery strategy as second decision variables; the energy storage scheduling operation constraint comprises an energy storage scheduling constraint, a system power balance constraint, a voltage constraint, an energy storage electric quantity balance and an energy storage charge state;
the upper model provides a planning scheme for the lower model, the lower model provides an energy storage operation simulation strategy for the upper model, and an energy storage planning double-layer model is built.
By implementing the embodiment of the invention, the energy storage operation simulation strategy is provided for the upper model by the lower model, the upper model provides a planning scheme for the lower model, the correction planning configuration and the operation strategy are continuously updated, and finally the optimal energy storage configuration and the related scheduling operation strategy are obtained. When solving the energy storage planning double-layer model, the energy storage configuration in a typical fault scene is used as research content, and aiming at a certain typical fault scene, how to deploy the energy storage and strengthen key links is researched, and meanwhile, the recovery efficiency of the system is improved. By adopting a planning double-layer model, the economy of system planning of the reinforcement key links is considered, the recovery efficiency of the energy storage dispatching operation support system is also considered, the energy storage configuration and dispatching operation strategy under different typical fault scenes are used for reinforcing the key links, and the elasticity level of the system is improved.
In order to solve the same technical problem, an embodiment of the present invention further provides an energy storage planning device of a power system, including: the system comprises a scene data module, a fault probability module, a cluster analysis module and an energy storage planning module;
the scene data module is used for generating an extreme event scene set and a source load fluctuation scene set according to scene data of the power system; the scene data comprise extreme event data, new energy fluctuation data and load fluctuation data in a conventional scene;
the fault probability module is used for driving physical elements of the power system to carry out deep neural network data according to different key types of operation scenes to obtain an element fault probability model, carrying out Monte Carlo scene simulation on the element fault probability model and the extreme event scene set, and generating a complex fault scene set;
the cluster analysis module is used for carrying out cluster analysis on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set;
the energy storage planning module is used for constructing an energy storage planning double-layer model according to energy storage configuration investment and energy storage scheduling operation, solving the energy storage planning double-layer model according to a typical key scene set and generating an energy storage planning scheme.
In order to solve the same technical problems, the embodiment of the invention also provides a regulating and controlling method of the electric power system, which comprises the steps of regulating the energy storage configuration of the electric power system according to an energy storage planning scheme, and regulating and controlling the operation of the electric power system; the energy storage planning scheme is obtained according to an energy storage planning method of the power system.
Drawings
Fig. 1: a schematic flow chart of an embodiment of an energy storage planning method of an electric power system is provided by the invention;
fig. 2: the invention provides a power system performance schematic diagram before and after the action of an extreme event of one embodiment of an energy storage planning method of a power system;
fig. 3: the invention provides a cyclic convolution neural network model diagram of an embodiment of an energy storage planning method of a power system;
fig. 4: the invention provides a cluster analysis model diagram of one embodiment of an energy storage planning method of a power system;
fig. 5: an elastic lifting-oriented energy storage double-layer planning model diagram in a typical fault scene of one embodiment of an energy storage planning method of a power system is provided by the invention;
fig. 6: the invention provides a structural schematic diagram of an embodiment of energy storage planning equipment of a power system.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1, a flow chart of an energy storage planning method of an electric power system according to an embodiment of the invention is shown. The energy storage planning method is suitable for energy storage double-layer intelligent planning for elastic lifting of a novel power system, the embodiment generates a complex fault scene set through deep neural network data driving and Monte Carlo scene simulation, a typical key scene set is obtained through cluster analysis, an energy storage planning double-layer model is solved, an energy storage double-layer intelligent planning scheme is accurately formulated, energy storage is efficiently scheduled, recovery efficiency of the power system is improved, and the elasticity level of the power system is improved.
It should be noted that, as shown in fig. 2, the overall process of resisting and recovering the elastic power grid can be divided into two phases: an elastic receiving stage (t 0-t 3) and an elastic restoring stage (t 3-t 6). The elastic endurance phase is divided into elastic preparation process (from t0 to t 1) and elastic resistance, absorption and response process (from t1 to t 3). From t0 to t1, the power system is in an elastic state with robustness/resistance capability during operation; when an extreme event starts to attack from t1, the power grid needs to mobilize elastic resources to start self-adaptive adjustment; entering a degradation state after an event, and starting to recover the system after t2 and t 3;
In the process of operating the power system from t0 to t1, fig. 2 shows a schematic diagram of the power system from before the disaster occurs to finally restoring to the initial state. In connection with the disaster process, the response of the system to extreme events is mainly represented by: before the disaster action (before t 1), the system is not affected to maintain the initial running state and the system performance, and the system can be prepared and prevented in a targeted way before the initial running state and the system performance; in the action process (t 1-t 2) of the disaster on the system, the system response adapts to the disaster influence, the related equipment is damaged to different degrees, and the performance of the system is degraded; after the extreme disasters, the system performance drops to a certain degree, and partial important loads (t 2-t 3) are maintained through local power supply; and then, emergency power supply is carried out on the power failure link through elastic resources in the effective dispatching system, the damaged facilities are maintained to gradually recover the system net frame, normal power supply and system performance recovery are realized (t 3-t 4), and finally, after the damaged facilities of the power grid are integrally repaired and the power supply is recovered, the system performance reaches an expected state and safe and stable operation is kept (t 4).
The energy storage planning method of the power system comprises steps 101 to 104, wherein the steps are as follows:
step 101: generating an extreme event scene set and a source load fluctuation scene set according to scene data of the power system; the scene data comprises extreme event data, new energy fluctuation data and load fluctuation data in a conventional scene.
In this embodiment, an initial scenario is generated in consideration of new energy fluctuation, load fluctuation, and extreme event uncertainty in a conventional scenario. The meteorological factors of the conventional scene are full cloud-free illumination, breeze, normal humidity and temperature.
Optionally, step 101 specifically includes steps 1011 to 1012, each of which specifically includes the following steps:
step 1011: and obtaining a source load fluctuation scene set according to the new energy fluctuation data and the load fluctuation data in the conventional scene in the historical data of the power system.
In the embodiment, new energy fluctuation and load fluctuation conditions in a conventional scene are extracted through historical data, and a source load fluctuation scene set is established.
Step 1012: and carrying out extreme event key parameter simulation on the extreme event data of the power system to generate an extreme event scene set.
In this embodiment, the extreme event data is combined with the extreme event parameters to generate an extreme event scenario set. Taking typhoons as an example, firstly, grid dividing a power grid area at a geographic level; secondly, calculating the vertical wind speed of a line central point at each moment according to typhoon wind field position data, wind speed data, typhoon 10-level wind circle radius and other data; and finally, forming wind speed values suffered by the line at different moments in different grid areas under the typhoon path.
Step 102: and driving physical elements of the power system by deep neural network data according to different key types of operation scenes to obtain an element fault probability model, and performing Monte Carlo scene simulation on the element fault probability model and the extreme event scene set to generate a complex fault scene set.
In this embodiment, by combining the key scenario and the physical element (the current power grid physical structure), the action mechanism of different pole end event scenarios on element faults is accurately described by adopting a data driving method, an element fault rate model under an extreme event is obtained, a critical scenario of a large number of faults of the power system, namely a complex fault scenario set, is generated by adopting Monte Carlo simulation, and fragile elements and fault conditions thereof in the power system after a disaster with certain intensity are obtained.
Optionally, step 102 specifically includes steps 1021 through 1022, where each step specifically includes:
step 1021: according to the operation scenes of different key types, obtaining corresponding meteorological factors under the operation scenes of different key types; the different key types of operation scenes comprise an extreme operation scene and a complex operation scene, and meteorological factors comprise wind speed, wind direction, temperature, humidity, illumination included angle and light intensity; building a cyclic convolutional neural network model, inputting an extreme event data set into the cyclic convolutional neural network model for training, and obtaining an element fault probability model; the extreme event data set comprises corresponding historical equipment fault conditions and meteorological factors under different key type operation scenes.
In the embodiment, a mechanism of action of a typical event on a novel power system is described through a neural network data driving method, a system fault scene and probability distribution thereof are obtained, and an element fault probability model is obtained. Constructing a cyclic convolutional neural network, and describing the fault conditions and probability distribution of different elements in the system under different scenes through model training to obtain an element fault probability model, wherein the method specifically comprises the following steps of: input external environmental factors (extreme event data set): such as historical equipment fault conditions under meteorological factors such as historical wind speed, wind direction, temperature, humidity, illumination included angle, light intensity and the like and corresponding different key type operation scenes (complex operation scenes and extreme operation scenes), wherein the complex operation scenes are temperature and humidity abnormal cloudy weather under cloudy weather and new energy fluctuation under wind instability; the extreme scenes are complex operation scenes of the operation scenes under disaster weather such as typhoons, heavy rain and the like. And (3) carrying out cyclic convolution neural network model training on the input extreme event data set, then adopting more unknown environmental factors in different scenes as model input, acquiring fault probability distribution of different equipment, obtaining a trained element fault probability model, and generating a fault scene set through Monte Carlo simulation.
Optionally, the extreme event data set is input into a cyclic convolutional neural network model for training, and an element fault probability model is obtained, specifically: inputting the extreme event data set into a cyclic convolutional neural network model for training, adjusting model parameters, drawing a learning curve in combination with a training process, setting a model loss function threshold value to determine a training termination condition, updating model parameters and optimal weights, and describing association relations between different complex extreme scenes and equipment element faults to obtain a trained element fault probability model; the model parameters comprise a learning rate, a small-batch sample scale parameter and a neural network scale parameter.
In this embodiment, based on the critical parameter information of the extreme event, an extreme event data set is generated through simulation, and the extreme event data set includes corresponding historical equipment fault conditions and meteorological factors under different critical type operation scenes. And inputting the extreme event data set as a training set into a cyclic convolutional neural network model for training, continuously adjusting model parameters such as a learning rate, a small batch of sample scale, a neural network scale and the like, drawing a learning curve in combination with a training process, setting a model loss function threshold value to determine a termination condition, updating model related parameters and optimal weights, effectively describing association relations between different polar end scenes and element faults, and establishing an element fault rate model and probability distribution.
Optionally, a cyclic convolutional neural network model is built, as shown in fig. 3, and specifically includes steps 301 to 303, where each step is specifically as follows:
in this embodiment, the main construction process includes respectively designing an input layer, a circulation layer, a convolution layer and a pooling layer of the cyclic convolutional neural network model. The input layer takes meteorological factors such as wind speed, wind direction, temperature, humidity, illumination included angle, light intensity and the like as external environment factors to input the neural network, and takes the fault probability of equipment in corresponding scenes (complex operation scenes and extreme operation scenes) as the output of the circular convolution neural network model.
Step 301: the method comprises the steps of designing a circulating layer of a circulating convolutional neural network model to be a bidirectional circulating neural network, forward circulating to access the previous environmental factor to obtain forward circulating output, backward circulating to access the latter environmental factor to obtain backward circulating output, and carrying out transverse cascade connection on the forward circulating output and the backward circulating output to obtain circulating layer output.
In this embodiment, the cyclic layer is a Bi-directional cyclic neural network (Bi-directional Recurrent Neural Network, BIRNN) for analyzing the environmental factor h 1 Wherein the forward loop accesses the previous environmental factor, the forward loop outputs, as follows:
h f (i)＝δ(W f ·h 1 (i)+W (fc) ·c(i-1))
The backward circulation accesses the latter environment factor, and outputs the backward circulation, as follows:
h b (i)＝δ(W b ·h 1 (i)+W (bc) ·c(i+1))
the cyclic layer output is formed by forward cyclic output and backward cyclic output which are transversely cascaded, and the expression is as follows:
h 2 (i)＝[h f (i),h b (i)]
h 2 ＝[h 2 (1)；h 2 (2)；...h 2 (n)]
wherein W is f And W is b The current input to output conversion matrices corresponding to the forward sequence and the reverse sequence, respectively. W (W) (fc) And W is (bc) Transfer matrix from the forward sequence environment factor and the reverse sequence environment factor to the current output respectively, c (i-1) and c (i+1) correspond to the forward sequence environment factor and the reverse sequence environment factor respectively, h f (i) For the forward output of the ith environmental factor, h b (i) Is the reverse output of the ith environmental factor, h 1 (i) Is the ith value, h in the environmental factor sequence 2 As an output of the recurrent neural network. In the transverse cascade, "is a transverse cascade,"; "is a vertical concatenation,".
Step 302: the method comprises the steps of designing a convolutional layer of a cyclic convolutional neural network model as a CNN convolutional neural network, selecting a preset number of convolutional kernels with different sizes, and respectively carrying out convolutional operation on the output of the cyclic layer to obtain various convolutional output matrixes, wherein the number of columns of the convolutional kernels is the same as the number of columns of the output of the cyclic layer, and each step of convolutional operation comprises the following formula:
h c (i)＝f(W c ·h 2 (i:i+l-1)+b c )
hr c (i)＝relu(h c (i))
Wherein h is c (i) For convolving the output matrix, hr c (i) For the i-th convolution vector obtained after convolution, l is the number of rows of the convolution kernel, i=1, 2,.. 2 (i: i+l-1) is a moment formed by the i-th to i+l-1 th row vectors outputted from the loop layerArray, b c For the convolution operation bias term, relu is the activation function of the CNN convolution neural network.
In the present embodiment, the convolutional layer is a CNN convolutional neural network (Convolutional Neural Network, CNN), and a convolutional kernel matrix W is first adopted c And carrying out convolution operation on the output of the circulating layer, wherein the number of the convolution kernel columns is the same as that of the output columns of the circulating layer, and the number of the rows is l, so that the size of the convolution kernel can be adjusted. Each step of convolution operation in the CNN convolution neural network is as follows:
h c (i)＝f(W c ·h 2 (i:i+l-1)+b c )
hr c (i)＝relu(h c (i))
wherein i=1, 2,..n-l+1; h is a 2 (i: i+l-1) is a cyclic layer output matrix h 2 Matrix formed by the i-th to i+l-1 th row vectors, b c For the convolution operation bias term, relu is the activation function of the neural network, hr c (i) Is the ith convolution vector obtained after convolution.
Preferably, the preset number is 3, 3 convolution kernels with different sizes are selected to respectively operate with the output matrixes of the circulating layer, and 3 convolution output matrixes are obtained.
Step 303: designing a pooling layer of the cyclic convolutional neural network model as a maximum element pooling layer, selecting the maximum element in the output of the cyclic layer and the convolutional vector as a characteristic value, and cascading all the characteristic values to form a characteristic vector.
In this embodiment, the pooling layer extracts important features using the maximum pooling layer, and selects a convolution vector hr obtained by the calculation of the cyclic layer matrix and the convolution kernel c (i) The largest element of (2) is taken as a characteristic value p c (i) Cascading all feature values to form feature vectors, and representing environmental factors in the scene, wherein the following formula is as follows:
p c (i)＝max[hr c (i)]
p c ＝[p c (1)；p c (2)；...；p c (i-l+1)]
three groups of different feature vectors are obtained by the selected three groups of convolution kernels with different sizes, and then the three groups of different feature vectors are longitudinally cascaded to enrich scene environment features. The pooling layer further extracts text features, so that feature vector dimensions are reduced, calculated amount is reduced, and classification efficiency is improved. And finally outputting the model as the equipment failure probability under different types of scenes.
Step 1022: inputting the extreme event scene into an element fault probability model to obtain the fault probability of each element; according to a preset numerical range, simulating sampling by a Monte Carlo method to obtain a plurality of random numbers; comparing each random number with each element fault probability respectively to obtain the state of each device; and generating a complex fault scene set according to the states and the positions of the devices.
In this embodiment, the critical scene of the system is generated by simulating each element fault probability through the trained element fault probability model by using the monte carlo, that is, the fault scene under the extreme event is obtained by using the monte carlo simulation, so as to obtain the fragile element and the possible fault condition thereof in the power system after the disaster with certain intensity occurs, and obtain the complex fault scene set. The Monte Carlo simulation is specifically: adopting Monte Carlo analog sampling according to a preset numerical range to obtain a plurality of random numbers; comparing each random number with each element fault probability respectively to obtain the equipment state; and forming a candidate fault scene set according to the state and the position of each device judged by each random number, and generating a complex fault scene set.
Step 103: and carrying out cluster analysis on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set.
In this embodiment, a cluster analysis method is adopted to obtain key operation scenes from a large number of complex fault scene sets and source load fluctuation scene sets, the source load fluctuation scene sets and the large number of complex fault scene sets are subjected to scene aggregation, specific representative key scenes are extracted, and source load fluctuation scenes with high probability-small loss and typical extreme scenes with low probability-high loss are obtained to form typical key scene sets.
Optionally, step 103 specifically includes taking a complex fault scene set and a source load fluctuation scene set as training sets of a scene aggregation algorithm, selecting a plurality of typical fault scenes as scene clustering centers, solving spatial distances between all the complex fault scenes and initial scenes, classifying the complex fault scenes and the initial scenes into the nearest initial scenes, and obtaining scene sets of all categories; the initial scene comprises an extreme event scene set and a source load fluctuation scene set; and updating each scene clustering center according to each category of scene set, and performing loop iteration until each scene clustering center meets the clustering requirement, thereby obtaining a typical key scene set.
In this embodiment, the fault scenes of different types caused by the conventional scene and the extreme scene are used as training sets, that is, a complex fault scene set and a source load fluctuation scene set are used as training sets of a scene aggregation algorithm, and a cluster analysis model is used, as shown in fig. 4, k typical fault scenes are selected as clustering centers by combining experience, all fault scenes are separated from an initial scene by solving spatial distances, and are classified into the nearest initial scene, k category scene sets are finally obtained, k scene centers are updated according to the k scene sets, and the clustering requirements are satisfied by the k scene centers through recycling iteration.
It should be noted that, the implementation flow of the clustering algorithm is as follows: scene aggregation means that in a given original scene set S, a deleted scene set Q and a key scene set J are obtained through a scene aggregation algorithm, so that the key scene set J replaces the scene set S to the greatest extent, and under the general condition, the objective function of the scene aggregation optimization algorithm is as follows:
Figure BDA0004186490230000121
in p i For scene u i Probability of d (u) i ,u j ) For scene u i And scene u j The distance between the two data sets is similar to a clustering center set in a clustering algorithm, and the clustering is one of the common methods in the fields of data mining and artificial intelligence, is an unsupervised learning method, and groups the data sets without prior knowledge guidance, so that the same type of data is similar as much as possible, and the difference between different types of data sets is as large as possible.
The K-medoids algorithm is a clustering method for discrete mixed data, and is applied to a field Jing Juge, firstly, K category scenes are determined and divided, and according to the K category scenesExperience gets K initial scenes u k (k=1, 2, …, K), calculating the distances from the rest of the scenes to the K initial scenes, classifying the rest of the scenes into the initial scenes closest to the rest of the scenes, and obtaining K category scene sets Mk (k=1, 2, …, K), and updating the new center of each scene set using the following equation;
Figure BDA0004186490230000131
Repeating the above process until
Figure BDA0004186490230000132
And if the scene aggregation classification value is smaller than the set value or various key scenes are not changed any more, ending the scene aggregation classification process.
The scene clustering center is obtained through experience, and is adjusted by combining with the actual clustering effect to obtain the optimal category number and the corresponding scene set.
Step 104: and constructing an energy storage planning double-layer model according to the energy storage configuration investment and the energy storage scheduling operation, and solving the energy storage planning double-layer model according to a typical key scene set to generate an energy storage planning scheme.
In this embodiment, based on a typical fault scenario, an elastic lifting-oriented energy storage planning dual-layer model is built, and as shown in fig. 5, the energy storage configuration in the typical fault scenario is first used as research content, and for a certain typical fault scenario, how to deploy energy storage and strengthen key links is researched, and at the same time, the recovery efficiency of the system is improved.
Optionally, step 104 specifically includes steps 1041 to 1044, where each step specifically includes the following steps:
step 1041: and constructing an upper model of the energy storage configuration investment by taking the energy storage investment, load reduction and weak link reinforcement cost as a first objective function, taking allowable installation maximum power and capacity as first constraint conditions and taking energy storage power energy configuration and key equipment reinforcement measures as first decision variables.
In this embodiment, first, an upper model is set up for configuration planning of energy storage, taking the main problem of energy storage configuration investment into consideration, wherein the upper model is set up by taking energy storage investment (energy storage investment), load reduction under a typical fault scene and weak link reinforcement cost (key link reinforcement cost) as objective functions, taking the installation positions and the number of the energy storage, and the maximum allowable installation power and capacity as constraints, and taking the energy storage power capacity configuration (energy storage power energy configuration) and key equipment reinforcement measures as decision functions.
Step 1042: constructing a lower model of energy storage scheduling operation by taking load reduction cost as a second objective function, energy storage scheduling operation constraint as a second constraint condition and energy storage scheduling strategy, line and unit recovery strategy as second decision variables; the energy storage scheduling operation constraint comprises an energy storage scheduling constraint, a system power balance constraint, a voltage constraint, an energy storage electric quantity balance and an energy storage charge state.
In this embodiment, the lower model is an energy storage scheduling operation, and the energy storage scheduling operation sub-problem is considered to be the lower model, wherein the load reduction cost (load reduction amount) is taken as an objective function, the energy storage scheduling policy (energy storage regulation policy) and the line and unit recovery policy are taken as decision variables, the line and unit recovery policy includes a line recovery policy and a unit recovery policy in system recovery, and the energy storage scheduling constraint conditions, the system power balance constraint conditions, the voltage constraint conditions, the energy storage electric quantity balance conditions, the energy storage charge state and the like are taken as energy storage scheduling operation constraint conditions, so that the lower model of the energy storage scheduling operation is built.
Step 1043: the upper model provides a planning scheme for the lower model, the lower model provides an energy storage operation simulation strategy for the upper model, and an energy storage planning double-layer model is built.
In the embodiment, in the overall collaborative planning, an upper model provides a planning scheme for a lower model, the lower model provides an energy storage operation simulation strategy for the upper model, and finally an intelligent energy storage planning double-layer model is established. And the energy storage optimization configuration is realized, an energy storage operation simulation strategy is provided for the upper model through the lower model, the upper model provides a planning scheme for the lower model, the correction planning configuration and the operation strategy are continuously updated, and finally the optimal energy storage configuration and the related scheduling operation strategy are obtained.
Step 1044: and solving the energy storage planning double-layer model according to the typical key scene set to generate an energy storage planning scheme.
In this embodiment, according to a typical key scene set, an energy storage planning dual-layer model is solved, corresponding energy storage planning schemes are formulated for different typical key scenes, and energy storage configuration and scheduling operation strategies under different typical scenes are generated, wherein the energy storage planning schemes comprise the energy storage configuration and the scheduling operation strategies. Based on typical fault scenes, an energy storage intelligent planning double-layer model oriented to elastic lifting is built, the energy storage configuration is guided to strengthen weak links, the energy storage is strengthened key links, the recovery efficiency of the energy storage lifting system is efficiently scheduled, and the elasticity level of the system is improved from two aspects of energy storage planning configuration and regulation and control operation.
By implementing the embodiment of the invention, the initial scene generation of new energy fluctuation, load fluctuation and extreme event uncertainty under the conventional scene is considered; accurately describing action mechanisms of event scenes of different poles on element faults by adopting a data-driven method, acquiring an element fault rate model under an extreme event, acquiring a complex fault scene set under the extreme event by utilizing Monte Carlo simulation, acquiring key operation scenes from a large number of complex fault scene sets and source load fluctuation scene sets by adopting cluster analysis, and forming a typical key scene set; an energy storage planning double-layer model is built, an upper-layer model is built to be an energy storage configuration investment master problem, a lower-layer model is set to be an energy storage scheduling operation sub-problem, and corresponding energy storage configuration and scheduling operation strategies are formulated for different typical key scene sets. The energy storage double-layer intelligent planning scheme is accurately formulated for elastic lifting of the novel power system, the energy storage configuration is guided to strengthen weak links, the energy storage is efficiently scheduled to improve the recovery efficiency of the power system, and the elasticity level of the power system is improved from two aspects of energy storage planning configuration and regulation and control operation.
Example two
Accordingly, referring to fig. 6, fig. 6 is a schematic structural diagram of a second embodiment of an energy storage planning device of an electric power system according to the present invention. As shown in fig. 6, the energy storage planning device of the power system includes a scene data module 601, a fault probability module 602, a cluster analysis module 603, and an energy storage planning module 604;
The scene data module 601 is configured to generate an extreme event scene set and a source load fluctuation scene set according to scene data of the power system; the scene data comprises extreme event data, new energy fluctuation data and load fluctuation data in a conventional scene.
Optionally, according to the scene data of the power system, an extreme event scene set and a source load fluctuation scene set are generated, specifically:
acquiring a source load fluctuation scene set according to new energy fluctuation data and load fluctuation data in a conventional scene in historical data of the power system;
and carrying out extreme event key parameter simulation on the extreme event data of the power system to generate an extreme event scene set.
The fault probability module 602 is configured to perform deep neural network data driving on physical elements of the power system according to different key types of operation scenarios, obtain an element fault probability model, and perform Monte Carlo scenario simulation on the element fault probability model and the extreme event scenario set, so as to generate a complex fault scenario set.
Optionally, according to different key type operation scenarios, driving the physical elements of the power system by the deep neural network data to obtain an element fault probability model, which specifically includes:
According to the operation scenes of different key types, obtaining corresponding meteorological factors under the operation scenes of different key types; the different key types of operation scenes comprise an extreme operation scene and a complex operation scene, and meteorological factors comprise wind speed, wind direction, temperature, humidity, illumination included angle and light intensity;
building a cyclic convolutional neural network model, inputting an extreme event data set into the cyclic convolutional neural network model for training, and obtaining an element fault probability model; the extreme event data set comprises corresponding historical equipment fault conditions and meteorological factors under different key type operation scenes.
Optionally, building a cyclic convolutional neural network model, which specifically comprises:
designing a circulating layer of a circulating convolutional neural network model as a bidirectional circulating neural network, forward circulating to access the previous environmental factor to obtain forward circulating output, backward circulating to access the latter environmental factor to obtain backward circulating output, and carrying out transverse cascade connection on the forward circulating output and the backward circulating output to obtain circulating layer output;
the method comprises the steps of designing a convolutional layer of a cyclic convolutional neural network model as a CNN convolutional neural network, selecting a preset number of convolutional kernels with different sizes, and respectively carrying out convolutional operation on the output of the cyclic layer to obtain various convolutional output matrixes, wherein the number of columns of the convolutional kernels is the same as the number of columns of the output of the cyclic layer, and each step of convolutional operation comprises the following formula:
h c (i)＝f(W c ·h 2 (i:i+l-1)+b c )
hr c (i)＝relu(h c (i))
Wherein h is c (i) For convolving the output matrix, hr c (i) For the i-th convolution vector obtained after convolution, l is the number of rows of the convolution kernel, i=1, 2,.. 2 (i: i+l-1) is a matrix formed by the i-th to i+l-1-th row vectors outputted from the cyclic layer, b c As a convolution operation bias term, relu is an activation function of the CNN convolution neural network;
designing a pooling layer of the cyclic convolutional neural network model as a maximum element pooling layer, selecting the maximum element in the output of the cyclic layer and the convolutional vector as a characteristic value, and cascading all the characteristic values to form a characteristic vector.
Optionally, the extreme event data set is input into a cyclic convolutional neural network model for training, and an element fault probability model is obtained, specifically:
inputting the extreme event data set into a cyclic convolutional neural network model for training, adjusting model parameters, drawing a learning curve in combination with a training process, setting a model loss function threshold value to determine a training termination condition, updating model parameters and optimal weights, and describing association relations between different complex extreme scenes and equipment element faults to obtain a trained element fault probability model; the model parameters comprise a learning rate, a small-batch sample scale parameter and a neural network scale parameter.
Optionally, performing Monte Carlo scene simulation on the element fault probability model and the extreme event scene set to generate a complex fault scene set, which specifically includes:
inputting the extreme event scene into an element fault probability model to obtain the fault probability of each element;
according to a preset numerical range, simulating sampling by a Monte Carlo method to obtain a plurality of random numbers;
comparing each random number with each element fault probability respectively to obtain the state of each device;
and generating a complex fault scene set according to the states and the positions of the devices.
The cluster analysis module 603 is configured to perform cluster analysis on the complex fault scene set and the source load fluctuation scene set, and generate a typical key scene set.
Optionally, performing cluster analysis on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set, which specifically includes:
taking the complex fault scene set and the source load fluctuation scene set as training sets of a scene aggregation algorithm, selecting a plurality of typical fault scenes as scene clustering centers, solving the spatial distances between all the complex fault scenes and the initial scenes, classifying the complex fault scenes and the initial scenes into the nearest initial scenes, and obtaining scene sets of all categories; the initial scene comprises an extreme event scene set and a source load fluctuation scene set;
And updating each scene clustering center according to each category of scene set, and performing loop iteration until each scene clustering center meets the clustering requirement, thereby obtaining a typical key scene set.
The energy storage planning module 604 is configured to operate according to the energy storage configuration investment and the energy storage schedule, build an energy storage planning dual-layer model, and solve the energy storage planning dual-layer model according to a typical key scene set to generate an energy storage planning scheme.
Optionally, building an energy storage planning double-layer model according to energy storage configuration investment and energy storage scheduling operation, and specifically:
taking energy storage investment, load reduction and weak link reinforcement cost as a first objective function, taking allowable installation maximum power and capacity as first constraint conditions, taking energy storage power energy configuration and key equipment reinforcement measures as first decision variables, and building an upper model of energy storage configuration investment;
constructing a lower model of energy storage scheduling operation by taking load reduction cost as a second objective function, energy storage scheduling operation constraint as a second constraint condition and energy storage scheduling strategy, line and unit recovery strategy as second decision variables; the energy storage scheduling operation constraint comprises an energy storage scheduling constraint, a system power balance constraint, a voltage constraint, an energy storage electric quantity balance and an energy storage charge state;
The upper model provides a planning scheme for the lower model, the lower model provides an energy storage operation simulation strategy for the upper model, and an energy storage planning double-layer model is built.
By implementing the embodiment of the invention, the purpose of developing disaster prevention and reinforcement technology of an elastic power system is to provide an intelligent energy storage planning technology for system elastic lifting. Considering the condition of new energy fluctuation and load fluctuation under the conventional scene of a novel power system and an extreme scene caused by the uncertainty of an extreme event; describing the action mechanism of an extreme event scene on a system device element by adopting a data driving method, and acquiring the element fault condition and probability distribution of the element under different extreme event scenes; combining the research, adopting Monte Carlo simulation to generate system fault scenes with different fault scales, introducing a source load fluctuation scene set, adopting a cluster analysis method to obtain the most representative key operation scene, and constructing an energy storage double-layer intelligent planning model by using an upper model with energy storage configuration leading and a lower model with energy storage scheduling operation leading based on the different key scenes; and finally, reinforcing key links by using energy storage configuration and scheduling operation strategies under different scenes, and improving the elasticity level of the system. The invention adopts a data driving method, adopts a deep neural network to describe the association relation between different types of scenes and element faults, avoids complex display expression of the different types of scenes and element fault rates, and simultaneously adopts a planning double-layer model, thereby not only considering the economy of system planning in a reinforcement key link, but also considering the recovery efficiency of an energy storage scheduling operation support system.
Example III
Correspondingly, the embodiment of the application also provides a regulating and controlling method of the power system, which comprises the following steps: and according to the energy storage planning scheme, adjusting the energy storage configuration of the power system, and regulating and controlling the operation of the power system.
The energy storage planning scheme is obtained according to the steps in the embodiment of the energy storage planning method of the power system.
According to the embodiment of the invention, the energy storage planning scheme, namely the energy storage planning configuration and the regulation strategy, in the corresponding scene are obtained, the reinforcement weak links of the energy storage configuration are guided, the recovery efficiency of the energy storage system is efficiently scheduled, the elasticity level of the system is improved from the two aspects of the energy storage planning configuration and the regulation operation, and the pre-disaster resistance and the post-disaster recovery efficiency of the system are improved.
The regulation and control method of the electric power system can implement the energy storage planning method of the electric power system in the energy storage planning method embodiment. The options in the above embodiments of the energy storage planning method are also applicable to this embodiment, and are not described in detail here. The rest of the embodiments of the present application may refer to the content of the embodiments of the energy storage planning method, and in this embodiment, no further description is given.
The foregoing embodiments have been provided for the purpose of illustrating the general principles of the present invention, and are not to be construed as limiting the scope of the invention. It should be noted that any modifications, equivalent substitutions, improvements, etc. made by those skilled in the art without departing from the spirit and principles of the present invention are intended to be included in the scope of the present invention.

Claims (10)

1. An energy storage planning method for an electric power system, comprising:
generating an extreme event scene set and a source load fluctuation scene set according to scene data of the power system; the scene data comprise extreme event data, new energy fluctuation data and load fluctuation data in a conventional scene;
according to different key types of operation scenes, driving physical elements of the power system by deep neural network data to obtain an element fault probability model, and performing Monte Carlo scene simulation on the element fault probability model and the extreme event scene set to generate a complex fault scene set;
performing cluster analysis on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set;
and constructing an energy storage planning double-layer model according to energy storage configuration investment and energy storage scheduling operation, and solving the energy storage planning double-layer model according to the typical key scene set to generate an energy storage planning scheme.
2. The energy storage planning method of the power system according to claim 1, wherein the element fault probability model and the extreme event scene set are subjected to monte carlo scene simulation to generate a complex fault scene set, specifically:
Inputting the extreme event scene into the element fault probability model to obtain the fault probability of each element;
according to a preset numerical range, simulating sampling by a Monte Carlo method to obtain a plurality of random numbers;
comparing each random number with each element fault probability respectively to obtain the state of each device;
and generating the complex fault scene set according to the state and the position of each device.
3. The energy storage planning method of the power system according to claim 1, wherein the step of performing deep neural network data driving on physical elements of the power system according to different key type operation scenarios to obtain an element fault probability model comprises the following steps:
acquiring corresponding meteorological factors under the different key type operation scenes according to the different key type operation scenes; the different key types of operation scenes comprise extreme operation scenes and complex operation scenes, and the meteorological factors comprise wind speed, wind direction, temperature, humidity, illumination included angle and light intensity;
building a cyclic convolutional neural network model, inputting an extreme event data set into the cyclic convolutional neural network model for training, and obtaining an element fault probability model; wherein the extreme event data set includes corresponding historical equipment fault conditions and meteorological factors under the different key type operation scenes.
4. The energy storage planning method of the power system according to claim 3, wherein the step of inputting the extreme event data set into the cyclic convolutional neural network model for training to obtain an element fault probability model comprises the following steps:
inputting the extreme event data set into the cyclic convolutional neural network model for training, adjusting model parameters, drawing a learning curve in combination with a training process, setting a model loss function threshold value to determine a training termination condition, updating the model parameters and the optimal weight, and describing association relations between different complex extreme scenes and equipment element faults to obtain a trained element fault probability model; the model parameters comprise a learning rate, a small-batch sample scale parameter and a neural network scale parameter.
5. The energy storage planning method of the power system according to claim 3, wherein the building of the cyclic convolutional neural network model is specifically as follows:
designing a circulating layer of the circulating convolutional neural network model as a bidirectional circulating neural network, forward circulating to access a previous environmental factor to obtain forward circulating output, backward circulating to access a latter environmental factor to obtain backward circulating output, and performing transverse cascade connection on the forward circulating output and the backward circulating output to obtain circulating layer output;
The convolution layer of the cyclic convolution neural network model is designed to be a CNN convolution neural network, convolution kernels with different preset numbers and different sizes are selected to respectively carry out convolution operation on the output of the cyclic layer, and each convolution output matrix is obtained, wherein the number of columns of the convolution kernels is the same as the number of columns of the output of the cyclic layer, and each step of convolution operation comprises the following formula:
h c (i)＝f(W c ·h 2 (i:i+l-1)+b c )
hr c (i)＝relu(h c (i))
wherein hc (i) is the convolution output matrix, hr c (i) I is the number of rows of the convolution kernel, i=1, 2,.. 2 (i: i+l-1) a matrix formed for the i-th to i+l-1 th row vectors of the cyclic layer output, b c As a convolution operation bias term, relu is an activation function of the CNN convolution neural network;
designing a pooling layer of the cyclic convolutional neural network model as a maximum element pooling layer, selecting the maximum element in the cyclic layer output and the convolutional vector as a characteristic value, and cascading all the characteristic values to form a characteristic vector.
6. The energy storage planning method of the power system according to claim 1, wherein the clustering analysis is performed on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set, specifically:
Taking the complex fault scene set and the source load fluctuation scene set as training sets of a scene aggregation algorithm, selecting a plurality of typical fault scenes as scene clustering centers, solving the spatial distances between all the complex fault scenes and an initial scene, classifying the complex fault scenes and the initial scene into the nearest initial scene, and obtaining scene sets of all categories; the initial scene comprises an extreme event scene set and a source load fluctuation scene set;
and updating each scene clustering center according to each category of scene set, and performing loop iteration until each scene clustering center meets the clustering requirement, thereby obtaining the typical key scene set.
7. The energy storage planning method of the power system according to claim 1, wherein the generating the extreme event scene set and the source load fluctuation scene set according to the scene data of the power system is specifically as follows:
obtaining a source load fluctuation scene set according to new energy fluctuation data and load fluctuation data in a conventional scene in the historical data of the power system;
and carrying out extreme event key parameter simulation on the extreme event data of the power system to generate the extreme event scene set.
8. The energy storage planning method of the power system according to claim 1, wherein the energy storage planning double-layer model is built according to energy storage configuration investment and energy storage scheduling operation, specifically:
Taking energy storage investment, load reduction and weak link reinforcement cost as a first objective function, taking allowable installation maximum power and capacity as first constraint conditions, taking energy storage power energy configuration and key equipment reinforcement measures as first decision variables, and building an upper model of the energy storage configuration investment;
taking load reduction cost as a second objective function, taking energy storage scheduling operation constraint as a second constraint condition, taking an energy storage scheduling strategy, a line and unit recovery strategy as a second decision variable, and building a lower model of the energy storage scheduling operation; the energy storage scheduling operation constraint comprises an energy storage scheduling constraint, a system power balance constraint, a voltage constraint, an energy storage electric quantity balance and an energy storage charge state;
the upper model provides a planning scheme for the lower model, the lower model provides an energy storage operation simulation strategy for the upper model, and the energy storage planning double-layer model is built.
9. An energy storage planning device for an electrical power system, comprising: the system comprises a scene data module, a fault probability module, a cluster analysis module and an energy storage planning module;
the scene data module is used for generating an extreme event scene set and a source load fluctuation scene set according to scene data of the power system; the scene data comprise extreme event data, new energy fluctuation data and load fluctuation data in a conventional scene;
The fault probability module is used for driving physical elements of the power system by deep neural network data according to different key types of operation scenes to obtain an element fault probability model, and simulating Monte Carlo scenes of the element fault probability model and the extreme event scene set to generate a complex fault scene set;
the cluster analysis module is used for carrying out cluster analysis on the complex fault scene set and the source load fluctuation scene set to generate a typical key scene set;
the energy storage planning module is used for constructing an energy storage planning double-layer model according to energy storage configuration investment and energy storage scheduling operation, solving the energy storage planning double-layer model according to the typical key scene set and generating an energy storage planning scheme.
10. A method of regulating an electrical power system, comprising: according to an energy storage planning scheme, adjusting energy storage configuration of a power system, and regulating and controlling the operation of the power system; wherein the energy storage planning scheme is obtained according to the energy storage planning method of the power system according to any one of claims 1 to 8.
CN202310420259.0A 2023-04-19 2023-04-19 Energy storage planning method and equipment for power system Pending CN116432968A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310420259.0A CN116432968A (en) 2023-04-19 2023-04-19 Energy storage planning method and equipment for power system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310420259.0A CN116432968A (en) 2023-04-19 2023-04-19 Energy storage planning method and equipment for power system

Publications (1)

Publication Number Publication Date
CN116432968A true CN116432968A (en) 2023-07-14

Family

ID=87094126

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310420259.0A Pending CN116432968A (en) 2023-04-19 2023-04-19 Energy storage planning method and equipment for power system

Country Status (1)

Country Link
CN (1) CN116432968A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117314043A (en) * 2023-08-23 2023-12-29 华北电力大学 Scene-driven comprehensive energy complementary capacity planning method and system
CN117556970A (en) * 2024-01-12 2024-02-13 杭州鸿晟电力设计咨询有限公司 Power distribution network planning method and system based on data driving

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117314043A (en) * 2023-08-23 2023-12-29 华北电力大学 Scene-driven comprehensive energy complementary capacity planning method and system
CN117556970A (en) * 2024-01-12 2024-02-13 杭州鸿晟电力设计咨询有限公司 Power distribution network planning method and system based on data driving
CN117556970B (en) * 2024-01-12 2024-04-09 杭州鸿晟电力设计咨询有限公司 Power distribution network planning method and system based on data driving

Similar Documents

Publication Publication Date Title
CN112733462B (en) Ultra-short-term wind power plant power prediction method combining meteorological factors
Ghaderi et al. Deep forecast: Deep learning-based spatio-temporal forecasting
CN116432968A (en) Energy storage planning method and equipment for power system
CN108365608B (en) Uncertain optimization scheduling method and system for regional energy Internet
ES2644528T3 (en) Method for the computer-aided determination of the use of electric power produced by a power generation plant, particularly a renewable power generation plant
CN103489038A (en) Photovoltaic ultra-short-term power prediction method based on LM-BP neural network
CN103729695A (en) Short-term power load forecasting method based on particle swarm and BP neural network
Zhou et al. Deep learning-based rolling horizon unit commitment under hybrid uncertainties
CN113313306B (en) Elastic neural network load prediction method based on improved wolf optimization algorithm
CN110942205A (en) Short-term photovoltaic power generation power prediction method based on HIMVO-SVM
CN112785027A (en) Wind-solar-storage combined power generation system confidence capacity evaluation method and system
CN110991722A (en) Power load prediction method
CN109583655A (en) A kind of hair transmission of electricity multistage joint Expansion Planning method and system
Eleni et al. New concerns on fuzzy cognitive maps equation and sigmoid function
CN111027760B (en) Power load prediction method based on least square vector machine
CN116300755A (en) Double-layer optimal scheduling method and device for heat storage-containing heating system based on MPC
CN112231845B (en) Stratospheric airship height control method and system
CN111401659A (en) Ultra-short-term or short-term photovoltaic power generation power prediction method based on case reasoning
CN117932417A (en) Hierarchical consistency wind speed prediction method based on multi-time scale features
CN111105025A (en) Urban high-voltage distribution network blocking management method based on data-driven heuristic optimization
CN111027855A (en) Power system risk control method considering power transmission line meteorological disaster fault probability
CN105184383A (en) Urban mobile emergency power supply optimal scheduling method based on intelligent optimization method
CN115271254A (en) Short-term wind power prediction method for optimizing extreme learning machine based on gull algorithm
CN114897220A (en) Micro-grid short-term load prediction method based on GRU neural network
Zhai et al. Pre-event Resilience Enhancement Strategy for Distribution Systems Based on Dueling DDQN

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination