CN116010831A - Combined clustering scene reduction method and system based on potential decision result - Google Patents

Abstract

The invention discloses a combined clustering scene reduction method and a system based on potential decision results, wherein uncertainty factors of all nodes in a power system are expressed as a multi-dimensional curve input domain in a time sequence curve form; constructing a power supply planning problem solving model considering operation check under different time scale scenes based on a multidimensional curve input domain, and converting the multidimensional curve input domain into a potential decision result domain for cluster analysis; carrying out SOM neural network clustering on the potential decision result to obtain a preliminary clustering result; and finally classifying the preliminary clustering result by adopting a SOM and k-medoids combined clustering method to obtain a typical scene adopted by final operation check. The method is used for clustering in the power supply planning typical scene operation check to obtain the most representative scene set, so that the complexity of solving the model is reduced to the greatest extent under the condition of guaranteeing the reliability of the check result, and the calculation efficiency is improved.

Description

Combined clustering scene reduction method and system based on potential decision result

Technical Field

The invention belongs to the technical field of power supply planning, and particularly relates to a combined clustering scene reduction method and system based on potential decision results.

Background

Conventional power planning problems meet the power requirements of a regional target year by deciding when and where to launch what capacity gensets. With the large-scale grid connection of renewable energy sources at the power supply side and the continuous improvement of the load level at the demand side, a decision maker needs to perform operation check on the proposed power investment decision result in order to cope with power fluctuation from the two sides of the source-load. The time scale is usually in the order of hours or even minutes no matter the machine set combination model or the economic dispatch model is used for checking, the related variables comprise the moment-by-moment output and the moment-by-moment fluctuation condition of load power of all the machine sets of all the types, and the existence of a plurality of nodes of the whole system, so that the solved problem becomes a large-scale Mixed Integer Linear Programming (MILP) problem. Meanwhile, as the planning period grows and the number of nodes in the system increases, the complexity of the solved problem will further increase.

Therefore, in order to cope with the complexity of solution caused by annual time-by-time operation check, the operation check of a typical scene is generally adopted to replace the operation check with high precision, which involves a scene cut technology. The traditional power planning scene reduction method generally adopts a heuristic method, namely, manually selecting a typical scene according to typical labels such as seasons, holidays and the like or heavy point indexes such as a load curve, a new energy output curve and the like. However, as the permeability of new energy increases gradually, the mutual coupling of multiple energy systems at the demand side causes uncertainty from both sides of the source and the load to promote the generation of more kinds of uncertainty scenes. The method is only used for selecting the complex power network which is not suitable for high-proportion new energy access according to a single standard through subjective judgment. Therefore, the current power supply planning decision maker adopts a clustering technology to reduce the number of typical scenes and improve the scene selection effect.

The current widely adopted clustering method comprises spectral clustering, k-means clustering and hierarchical clustering, and meanwhile, most papers can also draw conclusions, and no absolute optimal one exists in all different clustering algorithms. The scene reduction method performs cluster analysis by using the input fields of the planning model, such as a historical load curve and a historical new energy output curve, and has the advantages of simplicity in operation and capability of retaining the time correlation of the time sequence. However, for the power planning model, since the relationship between the output result and the input variable is highly nonlinear, two scenes with the same category in the input domain cluster may result in different decision results, so the clustering result based on the power planning input data often does not have good practical value, and may not have a representative effect on the final scene cut.

Therefore, the scene reduction method based on the investment decision cost can more directly judge the scenes with similar results from a plurality of scenes so as to make more accurate clustering, in addition, the efficiency of clustering can be greatly improved by reducing the dimension of a plurality of input data with higher dimension into potential investment results, and the difficulty of calculation and solution is reduced, so that the clustering method based on the investment decision cost is necessary to be summarized so as to improve the efficiency of typical scene analysis.

Disclosure of Invention

The technical problem to be solved by the invention is to provide a combined clustering scene reduction method and a system based on potential decision results for solving the technical problem of scene number reduction in power planning operation simulation aiming at the defects in the prior art.

The invention adopts the following technical scheme:

a combined clustering scene reduction method based on potential decision results comprises the following steps:

s1, representing uncertainty factors of all nodes in a power system as a multi-dimensional curve input domain in a time sequence curve mode;

s2, under the condition of constructing different time scales based on the multidimensional curve input domain obtained in the step S1, a power supply planning problem solving model for operation check is considered, and the multidimensional curve input domain is converted into a potential decision result domain for cluster analysis;

s3, carrying out SOM neural network clustering on the potential decision result obtained in the step S2 to obtain a preliminary clustering result;

s4, carrying out final classification on the preliminary clustering result obtained in the step S3 by adopting a SOM and k-medoids combined clustering method to obtain a typical scene adopted by final operation check.

Specifically, in step S1, the multidimensional curve input field Z is expressed as:

Where K/K is scene number/scene set,G ^RE the number of the power supplies of the new energy sources is the number,

for the multidimensional input curve vector in the scene k, N is the total number of nodes in the power system, T is the time of day in the scene k, and R is the real number domain.

Specifically, in step S2, the scene-based model is used as a substitute for the power planning solution model based on full scene operation check, the solved scene-based model is simplified to be a power planning model only containing a single scene for solving, the weight of a typical day is set to be 1, the power production sequence under the scene k is obtained by solving the model of the single scene, the production cost of various power production sequences is correspondingly multiplied, a potential investment decision result is obtained, and a potential decision result domain Γ formed by the potential decision result is constructed.

Further, the potential decision result domain is specifically Γ:

wherein ,

for potential investment decision results, K/K is scene number/scene set, G is total number of units, Y is time of day in scene K, and R is real number domain.

Further, solving the power planning model containing only a single scene is expressed as:

wherein ,OBJ_OD As objective function, Y is the number of planning years, G is the total number of units, a _y,g Is the unit capacity cost of the unit, x _y,g The integer variable for the decision making of annual y-sets g, T is the total number of moments in a year,

for cost coefficients associated with continuous variables, p _y,g,t Continuous variable for the operational decision of annual y-set g time t,/for the annual y-set g time t>

Is a cost coefficient related to 0-1 variable, u _y,g,t 0-1 variable for operational decision of annual y-set g moment t, y is planning annual number, A _y-1,g For constraint coefficient related to the number of production units in y-1 year, x _y-1,g and B_y,g For constraint coefficients related to the number of production units in y years, x is _y,g Integer variable d for construction decision of annual y-set g _y,g To put into production the upper limit constraint of the variable C _y,g,t For the proportionality coefficient related to the number of the production units at each moment, D _y,g As a scaling factor related to the continuous variable of the operational decision of the annual y-group at time g, p _y,g,t Continuous variable for operational decision of annual y-set g moment t, E _y,g For the scaling factor related to the 0-1 variable of the operational decision of the annual y-set at the moment t, u _y,g 0-1 variable, e for operational decision of annual y-set at g moment t _y,g,t Is the upper limit in the operational decision constraint.

Specifically, in step S3, the input elements of the SOM neural network are read, i.e. through steps S1 andin the step S2, the input of SOM neural network is composed of K M-dimensional vectors at decision result of single field Jing Qian

Defining each neuron in the neural network with its weight vector +.>

L is the number of neurons, randomly initialized +.>

Repeating the following steps for a given number of iterations:

selecting an input element

Calculating the +.>

The number of the corresponding best matching unit;

using

The information of the optimal matching unit and adjacent neurons is updated;

the learning process ends when the learning efficiency is less than a predefined threshold or an upper limit of the number of iterations is reached.

Further, a quantization error QE is introduced for characterization, the index represents an average distance between each input element and the best matching unit, and a calculation formula is as follows:

wherein ,

for input element +.>

Is used for matching the unit weight vector.

Specifically, step S4 specifically includes:

s401, introducing a dissimilarity matrix to record the distance between an input element and each neuron, and representing the similarity between the input element and each neuron by the distance;

s402, introducing a contour coefficient to judge the optimal clustering quantity, and selecting the quantity of clusters with the highest contour coefficient SC as the final clustering quantity;

s403, initializing a clustering center by adopting a k-means++ algorithm, and conforming to the principle of increasing the diffusivity of an initial centroid set;

S404, classifying elements except the clustering centers, calculating Euclidean distances from the elements to the clustering centers, and classifying the elements into the class of the closest clustering center;

s405, searching the elements with the smallest sum of the distances between the elements in each class and other elements in the class except the clustering center, and taking the elements as a new clustering center;

s406, repeating the steps S402 to S405, further clustering the preliminary clustering result based on SOM to obtain a neuron classification result after merging and reducing, and performing k-means clustering again on the input elements in each class to obtain the number of the mass centers in each class element as a typical scene reduction result.

Further, in step S402, the contour coefficient sc (i) is:

the total profile coefficient SC of the clusters is:

wherein a (i) is the average distance from the ith element to other elements in the same cluster, b (i) is the minimum value of the average distances from the ith element to all elements in other clusters, and N is the total number of nodes in the power system.

In a second aspect, an embodiment of the present invention provides a combined clustering scene reduction system based on potential decision results, including:

the representation module is used for representing uncertainty factors of all nodes in the power system into a multi-dimensional curve input domain in a time sequence curve mode;

The conversion module is used for constructing a power supply planning problem solving model considering operation check under different time scale scenes based on the multidimensional curve input domain obtained by the representation module, and converting the multidimensional curve input domain into a potential decision result domain for cluster analysis;

the clustering module is used for carrying out SOM neural network clustering on the potential decision result obtained by the conversion module to obtain a preliminary clustering result;

and the output module is used for carrying out final classification on the preliminary clustering result obtained by the clustering module by adopting a SOM and k-means combined clustering method to obtain a typical scene adopted by final operation check.

Compared with the prior art, the invention has at least the following beneficial effects:

the combined clustering scene reduction method based on the potential decision result is used for reducing the number of check scenes in power planning operation check, improving the scene selection efficiency and reducing the scale and difficulty of operation check operation on the premise of ensuring the operation check reliability.

Furthermore, reading relevant node load data and renewable energy output data from 365 scenes all the year round is a core for constructing a scene input domain, is also a basis for checking a power supply planning scheme, and ensures the integrity and reliability of the input data, thereby being a necessary condition for improving the operation checking reliability.

Further, modeling is carried out on power investment decisions and operation check under different scene dimensions, and the simulation check from the most precise and accurate full scene model to the typical day operation simulation to the single typical day is carried out, so that the difficulty of solving the problem of the decision model is reduced step by step, and the efficiency of obtaining a single potential decision result is improved. Meanwhile, the operation simulation based on the full scene and the operation simulation based on the typical scene check can play a role in checking the scene reduction effect from the other aspect.

Furthermore, by solving the power supply planning investment decisions of all scenes, the potential result domain Γ of the investment decisions is obtained, and further, the conversion of the multidimensional input domain into the input domain with lower dimension is realized, and convenience is provided for further cluster analysis. Meanwhile, the potential result field explicitly represents the construction cost problem concerned by the power investment decision, thereby providing key characteristics in cluster analysis.

Furthermore, the power planning model of the whole scene is simplified into the power planning model only comprising a single scene for solving, the operation constraint of each scene at each moment is disassembled, the operation constraint of each scene at each moment is simplified into the operation constraint of each scene at each moment, the complexity of solving the problem can be greatly reduced, and the solving efficiency is improved. And solving through a power supply planning model under a single scene to obtain potential decision results for cluster analysis.

Furthermore, the reliability of the clustering algorithm is greatly improved by performing dimension reduction processing on the input domain, and because the similarity measurement result is unreliable in a high-dimension space, mapping the high-dimension data to the low dimension is a necessary condition before clustering. Through principal component analysis, the multidimensional data in the input domain can be better subjected to feature analysis, the vector with larger influence on the whole input data is summarized, and the dimension of the input vector can be obviously reduced without losing the original feature attribute of the data to influence the result by retaining the feature vector corresponding to the principal component.

Furthermore, the average distance between the input element and the neuron in the SOM neural network is described by introducing the quantization error QE, so that the optimal matching unit of each input element can be obtained, the preliminary clustering of the SOM neural network is further realized, and the category of each input element is obtained.

Furthermore, the SOM neural network method is adopted to cluster potential decision results, and the decision results are initially clustered through the interaction mode and the competition mechanism of the self-organizing neural network, so that the clustering effect is improved through the artificial intelligence method. Meanwhile, by means of a competition mechanism of the SOM neural network, the result of the clustering method can have self-organizing characteristics, the unsupervised clustering algorithm k-means is adopted to recluster the primary result, firstly, a dissimilarity matrix is introduced to record the distance between the input vector and each neuron vector, and then the primary clustering result is fully captured and then further clustering is carried out. The adoption of the method which is more suitable for solving the median of the potential decision results ensures the applicability of the clustering method. Meanwhile, a k-means++ algorithm is adopted to initialize the clustering center, so that the high efficiency of the clustering algorithm is ensured.

Further, through the combined clustering of the SOM neural network and the k-means, the optimal k-modoids clustering quantity is judged by utilizing the contour coefficients, further clustering is carried out on the preliminary results, the numbers of the centroids in each category are solved, and the final scene reduction result is obtained.

It will be appreciated that the advantages of the second aspect may be found in the relevant description of the first aspect, and will not be described in detail herein.

In summary, the method is used for clustering in the power supply planning typical scene operation check to obtain the most representative scene set, so that the complexity of solving the model is reduced to the greatest extent under the condition of ensuring the reliability of the check result, and the calculation efficiency is improved.

The technical scheme of the invention is further described in detail through the drawings and the embodiments.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a graph of normalized potential decision results;

FIG. 3 is a graph of SOM preliminary clustering results, wherein (a) is a uniform distance matrix and (b) is an input element distribution map;

FIG. 4 is a graph of profile coefficients for a dissimilarity matrix cluster;

FIG. 5 is a diagram of potential decision results for a reduced scenario;

FIG. 6 is a multi-dimensional input graph of a reduced scene;

Fig. 7 is a comparison chart of the power supply cumulative installation result.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. 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.

In the description of the present invention, it will be understood that the terms "comprises" and "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

It is also to be understood that the terminology used in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

It should be further understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations, e.g., a and/or B, may represent: a exists alone, A and B exist together, and B exists alone. In addition, the character "/" herein generally indicates that the front and rear associated objects are an "or" relationship.

It should be understood that although the terms first, second, third, etc. may be used to describe the preset ranges, etc. in the embodiments of the present invention, these preset ranges should not be limited to these terms. These terms are only used to distinguish one preset range from another. For example, a first preset range may also be referred to as a second preset range, and similarly, a second preset range may also be referred to as a first preset range without departing from the scope of embodiments of the present invention.

Depending on the context, the word "if" as used herein may be interpreted as "at … …" or "at … …" or "in response to a determination" or "in response to detection". Similarly, the phrase "if determined" or "if detected (stated condition or event)" may be interpreted as "when determined" or "in response to determination" or "when detected (stated condition or event)" or "in response to detection (stated condition or event), depending on the context.

Various structural schematic diagrams according to the disclosed embodiments of the present invention are shown in the accompanying drawings. The figures are not drawn to scale, wherein certain details are exaggerated for clarity of presentation and may have been omitted. The shapes of the various regions, layers and their relative sizes, positional relationships shown in the drawings are merely exemplary, may in practice deviate due to manufacturing tolerances or technical limitations, and one skilled in the art may additionally design regions/layers having different shapes, sizes, relative positions as actually required.

The invention provides a combined clustering scene reduction method based on potential decision results, which is used for carrying out combined clustering on the potential decision results through a combined clustering method of SOM and k-medoids to obtain a scene set with the most representative decision results. Performing dimension reduction on the operation check model facing the full scene to obtain a typical scene operation check model endowed with weight, performing dimension reduction on input domains of each typical scene by a principal component analysis method, and reducing calculation pressure brought by high-dimension input domains to clustering; potential decision results of all single scenes are obtained through solving the single scene simple model, the results are used as input data of SOM cluster analysis, the simple model is solved one by one to replace the complex model for solving the whole scene, and the calculation difficulty is greatly reduced; and the dissimilarity matrix and the contour coefficient are introduced to judge the optimal k-medoids clustering quantity, then the k-medoids method is adopted to cluster potential decision results, and the degree of proximity to panoramic simulation results is improved under the clustering of the combination method, so that the result deviation generated by typical scene simulation is reduced.

Referring to fig. 1, the method for reducing combined clustering scenes based on potential decision results of the invention comprises the following steps:

s1, taking uncertainty factors of all nodes in a system as input data of a model in a time sequence curve mode;

firstly, acquiring required data from related departments, wherein the required data comprises annual load data of each node in the system and annual output conditions of callable resources at a load side; the annual output condition of the renewable energy unit in the system.

The load power data of each node in each scene k in the whole year is defined as:

wherein T is the time of day in scene k; n is the total number of nodes within the system under study.

Similarly, a wind power resource curve and a photovoltaic resource curve at each moment in each scene k in the whole year are defined as:

wherein ,G^WT The number of wind farms is the number; g ^S The number of the photovoltaic power stations is the number of the photovoltaic power stations.

As load side callable resources are increasingly increasing, the role played by flexible resources represented by demand response in power systems is increasingly important, and the uncertainty of the flexible resources should be considered in various scenarios, so that the upper limit of the response capacity of the flexible resources on the demand side is defined as:

all data information needed to be contained in the input domain is obtained, and four types of data vectors are spliced according to a modeling sequence on the premise of guaranteeing the time sequence: sequentially a load curve, a wind power resource output curve, a photovoltaic resource output curve and a demand side flexible resource demand response upper limit curve in a scene k; the vector obtained by splicing is a multidimensional input curve vector in a scene k:

The load, new energy and demand response resource multidimensional curve input domain suitable for the power supply planning operation check model is formed by the following steps:

wherein K is the number of scenes; g ^RE The number of the new energy power supplies.

S2, constructing a power supply planning problem solving model considering operation check under the scene of different time scales based on the defined input domain, and converting the multidimensional curve input domain into a potential decision result domain for cluster analysis;

firstly, defining a power supply planning solution model (FD) based on Full scene operation check, and performing operation check by adopting an input resource curve and a load curve of a Full period in a planning period:

wherein y is the planning year number; y is the number of planned years; G/G is the number of units/total number of units;

for time of day number/total number of times of year, wherein +.>

x _y,g Integer variables for the construction decision of the annual y-set g; p is p _y,g,t /u _y,g,t The continuous variable and 0-1 variable of the operational decision at the time t of the annual y-set g are adopted.

Equation (7) is an objective function of the power supply planning problem, targeting a minimized total projected cost and operating cost, where the operating cost includes both the continuous variable p _y,g,t Related operation costs, such as variable operation costs including fuel cost, carbon emission cost and the like of the traditional thermal power generating unit, and electric discarding punishment cost and the like of the new energy generating unit; also comprises the running variable u of the unit 0-1 _y,g,t Related costs, such as start-stop costs of thermal power generating units, and the like. And (3) planning investment decision-related constraints for the power supply, wherein the constraints comprise a production variable constraint, a new energy installed capacity duty ratio constraint, an investment budget upper limit constraint, an installed adequacy constraint and the like. The formula (9) is a constraint related to operation decisions at each moment in a planning period, and comprises upper and lower limit constraints of output force of various types of units, constraint of demand response calling moment, power balance constraint, section constraint and the like of the system at each moment in different scenes.

The decision variables related to the FD problem not only comprise power investment decision variables of units year by year, but also comprise output, load size and demand side response capacity of different types of units in each scene year by year, and the FD problem has the characteristics of large variable quantity, complex time sequence coupling constraint and high solving difficulty. In order to reduce the solving time, the equivalent and simplification can be carried out by adopting technologies such as combination and simplification of a thermal power unit.

Therefore, to reduce the complexity of the overall model solution, a scene-based model (Model based on scenario, abbreviated SD) is employed as a surrogate for the FD problem.

Wherein t is the time number in the scene; t is a set of moments within the scene t=24; K/K is scene number/scene set; omega _k For the weight of scene k, the number and weight of scenes can be obtained by a scene reduction method.

It should be noted that, the method of using the typical scene simulation cannot guarantee the scene continuity required by the unit combination climbing constraint and the stored energy time sequence balance constraint in the selected typical scene, so this type of constraint needs to be decoupled among the scenes.

Aiming at the combined clustering method based on potential decision results adopted by the method, the method for solving the models of a plurality of typical scenes is not an optimal method, and further simplifying the models, simplifying the solved SD problem into a power planning model (OD) only containing a single scene for solving, wherein the expression of the solving model is as follows:

because only a single scene is adopted to perform the simulation of operation check, the weight of a typical day is set to be 1, and the power supply production sequence under the scene k is obtained by solving the model of the single scene:

for planning problems, the cost of investment decision is often concerned, so that the investment decision result is obtained by correspondingly multiplying the investment cost of each type of power supply production sequence:

constructing a potential decision result domain Γ consisting of potential decision results:

The multidimensional input domain X is converted into a potential decision result domain Γ for cluster analysis, the potential decision result domain Γ is obviously reduced in the dimension of the input vector, and the contained information more obviously reflects the decision effect.

S3, performing SOM neural network clustering on the obtained potential decision result;

firstly, reading input elements of the SOM neural network, namely, the single field Jing Qian processed by the step S1 and the step S2 is used for determining the result, and the input of the SOM neural network comprises K M-dimensional vectors

Is a set of (3). At the same time, each neuron in the neural network is defined with its weight vector +.>

Where L is the number of neurons. Random initialization +.>

Repeating the following steps for a given number of iterations:

selecting an input element

Calculating the +.>

Number of the corresponding best matching unit:

using

The best matching unit and the weight vector of the adjacent neurons:

wherein alpha is learning efficiency, and decreases with increasing iteration times; h (s, l, r) is a neighborhood function, typically using a gaussian mexico cap function, and r is a neighborhood function radius.

The learning process ends when the learning efficiency is less than a predefined threshold or an upper limit of the number of iterations is reached. In order to screen out the SOM neural network with the best performance, a quantization error QE is introduced for depiction, the index represents the average distance between each input element and the best matching unit, and the calculation formula is as follows:

wherein ,

for input element +.>

Is used for matching the unit weight vector.

S4, carrying out final classification on the preliminary clustering result by adopting a k-medoids clustering algorithm.

S401, in order to fully capture the primary clustering result information, introducing a dissimilarity matrix to record the distance between the input element and each neuron, and representing the similarity between the input element and each neuron through the distance. The smaller the distance between the neuron and the input element indicates the better the neuron's representativeness of the data.

The specific expression of the similarity matrix is as follows:

s402, selecting the clustering quantity

In order to determine the clustering quantity of k-medoids, introducing a contour coefficient to judge the optimal clustering quantity, wherein the contour coefficient of a single element can measure the similarity degree between the element and the element in the self class and the similarity degree between the element and the element in other classes, and a specific calculation formula is as follows:

wherein a (i) is the average distance from the ith element to other elements in the same cluster, and represents the dissimilarity in the cluster; b (i) is the minimum value of the average distance from the ith element to all elements in other clusters, and represents dissimilarity among clusters.

The total profile coefficient SC of the clusters is:

/>

the range of values of the profile coefficients is [ -1,1], and the closer the value is to 1, the better the clustering performance is, and the closer the value is to-1, the worse the clustering performance is. The number of clusters with the highest profile factor is thus selected as the final number of clusters.

S403, initializing a clustering center

After the number of clusters is determined, a k-means++ algorithm is adopted to initialize the cluster centers, and the principle of increasing the diffusion degree of the initial centroid set is followed.

S404, classifying elements except the clustering centers, calculating Euclidean distances between the elements and each clustering center, and classifying the elements into the class of the closest clustering center.

S405, searching the elements with the smallest sum of the distances from other elements in the class among the other elements except the clustering center in each class, and taking the elements as a new clustering center.

And S406, repeating the steps S402 to S405, and further clustering the preliminary clustering result based on the SOM to obtain the neuron classification result after combination and reduction. And then k-means clustering is carried out again on the input elements in each class, and the number of the mass centers in each class of elements is calculated and used as a typical scene reduction result.

So far, the whole process of carrying out combined clustering on the potential decision results to achieve the purpose of scene reduction is completed.

In still another embodiment of the present invention, a system for reducing a combined clustering scene based on a potential decision result is provided, where the system can be used to implement the method for reducing a combined clustering scene based on a potential decision result, and specifically, the system for reducing a combined clustering scene based on a potential decision result includes a representation module, a transformation module, a clustering module, and an output module.

The system comprises a representation module, a control module and a control module, wherein the representation module represents uncertainty factors of all nodes in the power system as a multi-dimensional curve input domain in a time sequence curve mode;

the conversion module is used for constructing a power supply planning problem solving model considering operation check under different time scale scenes based on the multidimensional curve input domain obtained by the representation module, and converting the multidimensional curve input domain into a potential decision result domain for cluster analysis;

the clustering module is used for carrying out SOM neural network clustering on the potential decision result obtained by the conversion module to obtain a preliminary clustering result;

and the output module is used for carrying out final classification on the preliminary clustering result obtained by the clustering module by adopting a SOM and k-means combined clustering method to obtain a typical scene adopted by final operation check.

In yet another embodiment of the present invention, a terminal device is provided, the terminal device including a processor and a memory, the memory for storing a computer program, the computer program including program instructions, the processor for executing the program instructions stored by the computer storage medium. The processor may be a central processing unit (Central Processing Unit, CPU), but may also be other general purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc., which are the computational core and control core of the terminal adapted to implement one or more instructions, in particular to load and execute one or more instructions to implement the corresponding method flow or corresponding functions; the processor of the embodiment of the invention can be used for the operation of a combined clustering scene reduction method based on potential decision results, and comprises the following steps:

The uncertainty factors of all nodes in the power system are expressed as a multi-dimensional curve input domain in the form of a time sequence curve; constructing a power supply planning problem solving model considering operation check under different time scale scenes based on a multidimensional curve input domain, and converting the multidimensional curve input domain into a potential decision result domain for cluster analysis; carrying out SOM neural network clustering on the potential decision result to obtain a preliminary clustering result; and finally classifying the preliminary clustering result by adopting a SOM and k-medoids combined clustering method to obtain a typical scene adopted by final operation check.

In a further embodiment of the present invention, the present invention also provides a storage medium, in particular, a computer readable storage medium (Memory), which is a Memory device in a terminal device, for storing programs and data. It will be appreciated that the computer readable storage medium herein may include both a built-in storage medium in the terminal device and an extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also stored in the memory space are one or more instructions, which may be one or more computer programs (including program code), adapted to be loaded and executed by the processor. The computer readable storage medium may be a high-speed RAM Memory or a Non-Volatile Memory (Non-Volatile Memory), such as at least one magnetic disk Memory.

One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement the respective steps of the above-described embodiments with respect to a combined clustered scene reduction method based on potential decision results; one or more instructions in a computer-readable storage medium are loaded by a processor and perform the steps of:

the uncertainty factors of all nodes in the power system are expressed as a multi-dimensional curve input domain in the form of a time sequence curve; constructing a power supply planning problem solving model considering operation check under different time scale scenes based on a multidimensional curve input domain, and converting the multidimensional curve input domain into a potential decision result domain for cluster analysis; carrying out SOM neural network clustering on the potential decision result to obtain a preliminary clustering result; and finally classifying the preliminary clustering result by adopting a SOM and k-medoids combined clustering method to obtain a typical scene adopted by final operation check.

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. 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 analysis

In order to verify the effectiveness of the method provided by the invention, an XJTU-ROTS system is selected for calculation and analysis. In order to fully embody the intermittence and fluctuation of the new energy, the new energy curve of the test system is replaced by the new energy curve of a certain practical system in northwest, and the historical data set contains 365 days. The power sources and wind and light resources in the system are shown in table 1. Planning is carried out by adopting a single-node model, but the differential resource characteristics of new energy sources in each region are reserved.

Table 1 system has power supply and wind and light resource conditions

In order to control solving time, the types of the thermal power generating units with the same technical parameters are combined, so that the number of variables ranging from 0 to 1 is reduced. The parameters of the power supply to be selected and the stored energy are shown in tables 2 and 3.

Table 2 system selected power parameters

Table 3 system selected energy storage parameters

According to the method provided by the invention, potential decision results of a single scene model solving power supply planning problem under 365 scenes are solved first, and the results are shown in fig. 2.

And then, clustering by adopting an SOM neural network by taking potential decision results of each scene as input vectors, and performing parameter selection through Monte Carlo simulation, wherein the clustering result and the input element distribution of the SOM neural network with the best performance are shown in a figure 3.

Based on the SOM preliminary clustering result, the dissimilarity matrix between the neurons and the input elements is used as the input data of k-means clustering, and the contour coefficients under different clustering numbers are calculated, wherein the result is shown in figure 4.

And obtaining a classification result of the input elements according to the optimal matching relation in the SOM, taking the mass centers of the input elements of each type as a clustering center by using a k-means method again, wherein the obtained 16 potential decision results are shown in figure 5, and the multidimensional input domain curve is shown in figure 6.

To verify the effectiveness of the proposed method, a comparative example was designed:

benchmark: planning and solving by using an FD model based on the whole input domain;

case_1: performing scene reduction on the input domain clusters by using a k-medoids method, and performing planning solution by using an SD model;

case_2: performing scene reduction on the potential decision result domain clusters by using a k-medoids method, and performing planning solution by using an SD model;

case_3: and performing scene reduction on the potential decision result domain clusters by using an SOM+k-means combined clustering method, and performing planning solution by using an SD model.

The Benchmark algorithm is introduced as a reference for other algorithms, and the power supply cumulative installed capacity of the four algorithms is shown in fig. 7 under the optimum scene reduction number.

Further, the deviation of the installed capacity of the case_1 to case_3 power supply from the reference example result is measured by introducing a Normalized Root Mean Square Error (NRMSE), and the calculation formula of the NRMSE is as follows:

wherein G is the number of power supply types; s is S _g A calculated value for the source g-machine result;

is a reference value for the result of the power supply g.

The calculated number of scene cuts and NRMSE for each example are shown in table 4:

table 4 NRMSE for comparative example

/>

From the results, from case_1 to case_3, the NRMSE value of the power supply project result decreases, case_2 decreases 0.0473 compared with case_1, indicating that scene reduction based on the potential decision result is closer to the actual result than scene reduction based on the input domain data. Compared with Case_2, case_3 has the NRMSE value reduced by 0.0294, which indicates that the SOM and k-medoids combined clustering scene reduction method has better performance effect compared with the k-medoids clustering scene reduction method.

In summary, the combined clustering scene reduction method and system based on the potential decision result can improve the calculation efficiency of power supply planning operation check. Compared with the traditional panoramic simulation and typical day simulation, the invention adopts the SOM self-organizing neural network and the k-medoids clustering algorithm to cluster the potential decision results of multiple scenes all the year round, avoids the high dimensionality and complex calculation of the full scene simulation and also avoids the subjectivity and randomness of manually selecting typical days. By constructing power supply planning models with different scene scales, a model system for quick solving and accuracy checking is summarized. The method provided by the invention can rapidly summarize the characteristics of potential decision results under different scenes, efficiently cluster the characteristics by utilizing the advantages of an artificial intelligent algorithm, and finally perform deviation assessment on the clustered results and the standard simulation results by introducing an appropriate evaluation mechanism so as to complete the reduction of typical scenes of annual operation check in a relatively complete system. When the uncertainty of the new energy output prediction is difficult to solve and a plurality of flexible resources on the load side are continuously aggregated, the combined clustering algorithm based on the potential decision result can reduce the calculation problem caused by high-latitude and multi-scene operation simulation, improve the efficiency of operation simulation in power planning, ensure that the difference between the operation checking result and the panoramic operation checking result is kept within an acceptable range, and is very suitable for improving the calculation efficiency of the overall model solution of the power planning.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, specific names of the functional units and modules are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present application. The specific working process of the units and modules in the above system may refer to the corresponding process in the foregoing method embodiment, which is not described herein again.

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

In the embodiments provided in the present invention, it should be understood that the disclosed apparatus/terminal and method may be implemented in other manners. For example, the apparatus/terminal embodiments described above are merely illustrative, e.g., the division of the modules or units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection via interfaces, devices or units, which may be in electrical, mechanical or other forms.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated modules/units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the present invention may implement all or part of the flow of the method of the above embodiment, or may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and when the computer program is executed by a processor, the computer program may implement the steps of each of the method embodiments described above. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RandomAccess Memory, RAM), an electrical carrier wave signal, a telecommunications signal, a software distribution medium, etc., it should be noted that the computer readable medium may contain content that is appropriately increased or decreased according to the requirements of jurisdictions and patent practices, such as in certain jurisdictions, according to the jurisdictions and patent practices, the computer readable medium does not contain electrical carrier wave signals and telecommunications signals.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The above is only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited by this, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the claims of the present invention.

Claims (10)

1. The combined clustering scene reduction method based on the potential decision result is characterized by comprising the following steps of:

s1, representing uncertainty factors of all nodes in a power system as a multi-dimensional curve input domain in a time sequence curve mode;

s2, under the condition of constructing different time scales based on the multidimensional curve input domain obtained in the step S1, a power supply planning problem solving model for operation check is considered, and the multidimensional curve input domain is converted into a potential decision result domain for cluster analysis;

S3, carrying out SOM neural network clustering on the potential decision result obtained in the step S2 to obtain a preliminary clustering result;

s4, carrying out final classification on the preliminary clustering result obtained in the step S3 by adopting a SOM and k-medoids combined clustering method to obtain a typical scene adopted by final operation check.

2. The method for combined clustering scene cut-down based on potential decision results according to claim 1, wherein in step S1, the multidimensional curve input field Z is expressed as:

wherein K/K is scene number/scene set, G ^RE The number of the power supplies of the new energy sources is the number,

for the multidimensional input curve vector in the scene k, N is the total number of nodes in the power system, T is the time of day in the scene k, and R is the real number domain.

3. The method for reducing combined clustering scenes based on potential decision results according to claim 1, wherein in step S2, a scene-based model is used as a substitute for a power planning solution model based on full scene operation check, the solved scene-based model is simplified into a power planning model only containing a single scene to be solved, the weight of a typical day is set to be 1, a power supply production sequence under a scene k is obtained by solving the model of the single scene, the production cost of various types of power supply production sequences is correspondingly multiplied, potential investment decision results are obtained, and a potential decision result domain Γ formed by the potential decision results is constructed.

4. The method for reducing combined clustering scene based on potential decision result according to claim 3, wherein the potential decision result field is specifically Γ:

wherein ,

for potential investment decision results, K/K is scene number/scene set, G is total number of units, Y is time of day in scene K, and R is real number domain.

5. The method for reducing combined clustering scenes based on potential decision results according to claim 3, wherein solving the expression of a power planning model containing only a single scene is:

wherein ,OBJ_OD As objective function, Y is the number of planning years, G is the total number of units, a _y,g Is the unit capacity cost of the unit, x _y,g The integer variable for the decision making of annual y-sets g, T is the total number of moments in a year,

for cost coefficients associated with continuous variables, p _y,g,t Continuous variable for the operational decision of annual y-set g time t,/for the annual y-set g time t>

Is a cost coefficient related to 0-1 variable, u _y,g,t 0-1 variable for operational decision of annual y-set g moment t, y is planning annual number, A _y-1,g For constraint coefficient related to the number of production units in y-1 year, x _y-1,g and B_y,g For constraint coefficients related to the number of production units in y years, x is _y,g Integer variable d for construction decision of annual y-set g _y,g To put into production the upper limit constraint of the variable C _y,g,t For the proportionality coefficient related to the number of the production units at each moment, D _y,g As a scaling factor related to the continuous variable of the operational decision of the annual y-group at time g, p _y,g,t Continuous variable for operational decision of annual y-set g moment t, E _y,g For the scaling factor related to the 0-1 variable of the operational decision of the annual y-set at the moment t, u _y,g 0-1 variable, e for operational decision of annual y-set at g moment t _y,g,t Is the upper limit in the operational decision constraint.

6. The method for reducing combined clustering scene based on potential decision result as set forth in claim 1, wherein in step S3, input elements of the SOM neural network are read, that is, the single field Jing Qian processed in step S1 and step S2 is used for decision result, and the input of the SOM neural network is that the input element comprises K M-dimensional vectors

Defining each neuron in the neural network with its weight vector +.>

L is the number of neurons, randomly initialized +.>

Repeating the following steps for a given number of iterations:

selecting an input element

Calculating the +.>

The number of the corresponding best matching unit;

using

The information of the optimal matching unit and adjacent neurons is updated;

the learning process ends when the learning efficiency is less than a predefined threshold or an upper limit of the number of iterations is reached.

7. The method for reducing combined clustering scenes based on potential decision results according to claim 6, wherein quantization error QE is introduced for characterization, the index represents an average distance between each input element and its best matching unit, and the calculation formula is:

wherein ,

for input element +.>

Is used for matching the unit weight vector.

8. The method for reducing combined clustering scenes based on potential decision results according to claim 1, wherein step S4 is specifically:

s401, introducing a dissimilarity matrix to record the distance between an input element and each neuron, and representing the similarity between the input element and each neuron by the distance;

s402, introducing a contour coefficient to judge the optimal clustering quantity, and selecting the quantity of clusters with the highest contour coefficient SC as the final clustering quantity;

s403, initializing a clustering center by adopting a k-means++ algorithm, and conforming to the principle of increasing the diffusivity of an initial centroid set;

s404, classifying elements except the clustering centers, calculating Euclidean distances from the elements to the clustering centers, and classifying the elements into the class of the closest clustering center;

s405, searching the elements with the smallest sum of the distances between the elements in each class and other elements in the class except the clustering center, and taking the elements as a new clustering center;

S406, repeating the steps S402 to S405, further clustering the preliminary clustering result based on SOM to obtain a neuron classification result after merging and reducing, and performing k-means clustering again on the input elements in each class to obtain the number of the mass centers in each class element as a typical scene reduction result.

9. The method for combined clustering scene reduction based on potential decision results according to claim 8, wherein in step S402, the contour coefficient sc (i) is:

the total profile coefficient SC of the clusters is:

wherein a (i) is the average distance from the ith element to other elements in the same cluster, b (i) is the minimum value of the average distances from the ith element to all elements in other clusters, and N is the total number of nodes in the power system.

10. A combined clustering scene cut-down system based on potential decision results, comprising:

the representation module is used for representing uncertainty factors of all nodes in the power system into a multi-dimensional curve input domain in a time sequence curve mode;

the conversion module is used for constructing a power supply planning problem solving model considering operation check under different time scale scenes based on the multidimensional curve input domain obtained by the representation module, and converting the multidimensional curve input domain into a potential decision result domain for cluster analysis;

The clustering module is used for carrying out SOM neural network clustering on the potential decision result obtained by the conversion module to obtain a preliminary clustering result;

and the output module is used for carrying out final classification on the preliminary clustering result obtained by the clustering module by adopting a SOM and k-means combined clustering method to obtain a typical scene adopted by final operation check.

Images