A power system dispatching method and equipment

Abstract

This invention provides a power system dispatching method and apparatus. The method includes: constructing a historical dispatching knowledge graph dataset of the power system; performing fusion similarity matching on dispatching data and dispatching knowledge data; generating reference generation curves through a nearest neighbor propagation clustering algorithm; and constructing a dispatching recommendation knowledge graph. After fusion similarity matching of dispatching data and dispatching knowledge data, based on the fusion similarity ranking result, multiple historical day dispatching knowledge items with the highest similarity are selected as a subset of historical dispatching knowledge. Each generation curve data item in the historical dispatching knowledge subset is then aggregated to form the generation curve for the dispatching day. The power system is dispatched based on the generation curve. This method generates generation curves quickly and significantly improves the efficiency and quality of the decision-making process for dispatching personnel, greatly reducing repetitive and tedious manual work. Furthermore, the apparatus provided by this invention can execute the above method.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a power system dispatching method and equipment. Background Technology

[0002] In the context of big data in the power sector, the factors influencing power system dispatch have become more diverse and complex. Power system dispatchers need to comprehensively consider factors such as current weather conditions, relevant policies, system operating conditions, and maintenance plans to quickly formulate reasonable dispatch schemes. This requires power system dispatchers to make dispatch decisions rapidly by analyzing massive amounts of dispatch data, a process that is extremely time-consuming and labor-intensive.

[0003] Applying visualization technology to present the dispatch data that the power dispatch system needs to process can significantly improve work efficiency, promptly and accurately identify problems in the power dispatch system, and realize the image-based display of information data. Complex data will be presented as intuitive images, helping dispatch operators to quickly determine the cause of power failures and take targeted measures, thereby ensuring the stable operation of the power dispatch system.

[0004] In power dispatching, the application of deep neural network technology is beneficial to improving the efficiency of information extraction and processing during the operation process, thereby realizing parallel processing of dispatching data, online learning, and associative memory. In practical applications involving small-scale data, it can quickly diagnose faults and ensure the stable operation of the dispatching system. However, when analyzing massive dispatching data, neural networks have a slow convergence speed and consume a long time, which cannot meet the needs of rapid dispatching decision-making. Summary of the Invention

[0005] This invention provides a power system dispatching method and device to address the shortcomings of existing neural networks in analyzing massive dispatching data, which have slow convergence speed, long processing time, and inability to meet the needs of rapid dispatching decisions.

[0006] This invention provides a power system dispatching method, comprising the following steps:

[0007] Construct a knowledge graph dataset of historical dispatching in the power system, including historical dispatching data and historical dispatching knowledge data;

[0008] The scheduling data on the scheduling day is matched with the historical scheduling data to obtain the first similarity dataset of the scheduling data; wherein, the scheduling day refers to the date on which the power system needs to be scheduled, which can be the current day or a future day;

[0009] The scheduling text data of the scheduling day is matched with the historical scheduling knowledge data to obtain a second similarity dataset of scheduling text data.

[0010] The data from the two similarity datasets are normalized and then fused in units of a predetermined time length to obtain a fused similarity dataset; wherein each data point in the fused similarity dataset represents the fused similarity value corresponding to each predetermined time length;

[0011] The top-ranked data in the fusion similarity dataset are matched one by one with the historical scheduling knowledge data to obtain multiple data subsets, which are then aggregated into the power generation curve for the scheduling day.

[0012] The power system is regulated according to the aforementioned power generation curve. Constructing a historical dispatch knowledge graph dataset for the power system specifically includes the following steps:

[0013] Acquire historical scheduling data and historical scheduling knowledge data, and store them in a CSV file in days;

[0014] Using the historical scheduling data and historical scheduling knowledge data, entity node data and attribute data of the entity nodes are created. The relationship between the entity node data and the attribute data of the entity nodes is used to form the relationship edge data between entities, thus completing the construction of the power system historical scheduling knowledge graph.

[0015] A graph database platform is used to store the entity node data, the attribute data of the entity nodes, and the relationship edge data between the entity nodes in a graph structure.

[0016] Furthermore, in the step of matching the scheduling data of the scheduling day with the historical scheduling data to obtain the first similarity dataset of the scheduling data, the matching method used is cosine similarity matching.

[0017] Furthermore, in the step of matching the scheduling text data of the scheduling day with the historical scheduling knowledge data to obtain the second similarity dataset of scheduling knowledge data, the matching method used is to perform similarity matching using the BM25 algorithm based on a probability retrieval model.

[0018] The process of using the top-ranked data points from the fused similarity dataset and matching them one by one with the historical scheduling knowledge data to obtain multiple corresponding data subsets, and then aggregating them into the power generation curve for the scheduling day, specifically includes the following steps:

[0019] The data in the fusion similarity dataset are sorted from largest to smallest according to their fusion similarity values;

[0020] Based on the sorting, select multiple fusion similarity values ​​for a predetermined number of days;

[0021] Based on multiple fusion similarity values ​​for the selected predetermined number of days, the corresponding historical scheduling knowledge data for the predetermined number of days is selected from the historical scheduling knowledge data of the power system historical scheduling knowledge graph.

[0022] From the historical scheduling knowledge data of the selected corresponding predetermined number of days, power generation curve data is selected one by one for each historical scheduling knowledge data to form a power generation curve dataset;

[0023] The power of each power generation curve in the power generation curve dataset is selected from multiple time periods and used to form multiple daily power generation vectors.

[0024] The obtained multiple daily power generation vectors are used as samples to be clustered to construct a sample set to be clustered;

[0025] The daily power generation vector of each sample set to be clustered is standardized.

[0026] The nearest neighbor propagation clustering algorithm is used to cluster multiple samples in the sample set to be clustered into one class, and the cluster center of this class is calculated as the power generation curve.

[0027] The acquired historical power system dispatch data and historical dispatch knowledge are the historical power system dispatch data and historical dispatch knowledge data of the previous year.

[0028] The dispatch data includes at least one of the following: voltage, line power, load, generator output, air quality, temperature, relative humidity, wind direction, wind speed, and solar irradiance.

[0029] The dispatch knowledge data includes dispatch text data and power generation curves; the dispatch text data includes at least one of the following: power generation guidance policies, maintenance plans, dispatch logs, and operation methods for dispatching personnel.

[0030] Another aspect of the present invention provides an electronic device, including a processor and a memory, the memory storing program instructions, characterized in that: the processor executes the program instructions to perform the power system dispatching method in the optional implementation described above.

[0031] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a power system historical dispatch knowledge graph dataset from multi-source heterogeneous data involved in power system dispatch operations, such as large-scale historical dispatch data and historical dispatch knowledge; it simultaneously performs similarity matching from both historical dispatch data and historical dispatch text data to obtain multiple generation curves, and uses a nearest neighbor propagation clustering algorithm to obtain cluster centers as generation curves, which are then used for power system dispatch. This solution is time-efficient and can meet the needs of rapid dispatch decision-making; at the same time, it can comprehensively consider factors such as current weather conditions, relevant policies, system operating conditions, and maintenance plans to quickly formulate reasonable dispatch schemes, which helps to achieve automated and intelligent dispatch, greatly reduces the search space for optimal dispatch schemes, improves the decision-making and predictive capabilities of planning and operation personnel, reduces the workload of dispatch personnel, and improves the work efficiency of dispatch departments. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of the method of the present invention;

[0034] Figure 2 It is a cosine similarity space graph;

[0035] Figure 3 This is a flowchart of the nearest neighbor propagation clustering (AP) algorithm;

[0036] Figure 4 This is a graph showing the aggregated results of the dispatch reference power generation curve. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] This invention provides a power system dispatching method, comprising:

[0039] S1: Construct a historical dispatch knowledge graph dataset for the power system, including historical dispatch data and historical dispatch knowledge data;

[0040] S2: Match the scheduling data of the scheduling day with the historical scheduling data to obtain the first similarity dataset of the scheduling data;

[0041] S3: Match the scheduling text data of the scheduling day with the historical scheduling knowledge data to obtain the second similarity dataset of the scheduling text data;

[0042] S4: Normalize the data from the two similarity datasets and then fuse them in units of a predetermined time length to obtain a fused similarity dataset.

[0043] S5: Using the top-ranked data in the fusion similarity dataset, match each data set with the historical scheduling knowledge data to obtain multiple data subsets, and aggregate them into the power generation curve for the scheduling day;

[0044] S6: Regulate the power system according to the power generation curve.

[0045] The dispatching day refers to the date on which the power system needs to be dispatched; it can be the current day or a future day.

[0046] Dispatch data includes at least one of the following: voltage, line power, load, generator output, air quality, temperature, relative humidity, wind direction, wind speed, and solar irradiance.

[0047] The dispatch knowledge data includes dispatch text data and power generation curves; the dispatch text data includes at least one of the following: power generation guidance policies, maintenance plans, dispatch logs, and operation methods for dispatch personnel.

[0048] In step S1, the historical scheduling data is a collection of power system scheduling data for each day prior to the current date; preferably, the historical scheduling data is a collection of power system scheduling data for each day of the previous year; the historical scheduling knowledge data is a collection of power system scheduling knowledge data for each day prior to the current date; preferably, the historical scheduling knowledge data is a collection of power system scheduling knowledge data for each day of the previous year.

[0049] Specifically, step S1 includes the following detailed steps:

[0050] S101: Obtain historical scheduling data and historical scheduling knowledge data, and store them in a CSV file in days;

[0051] S102: Using the historical scheduling data and historical scheduling knowledge data, create entity node data and attribute data of the entity nodes, and use the relationship between the entity node data and the attribute data of the entity nodes to form entity relationship edge data, thereby completing the construction of the power system historical scheduling knowledge graph;

[0052] S103: A graph database platform is used to store the entity node data, the attribute data of the entity nodes, and the relationship edge data between the entity nodes in a graph structure.

[0053] Among them, the storage method of S101 realizes the data storage of time slices.

[0054] In S102, entity nodes and relationship edges between entities can be quickly created by calling the Cypher statement in the Neo4j graph database using a programming language.

[0055] Neo4j is a high-performance, NoSQL graph database that stores structured data on a network rather than in tables. It's an embedded, disk-based, fully transactional persistence engine that stores structured data on a network (mathematically called a graph) instead of tables. Neo4j can also be viewed as a high-performance graph engine with all the features of a mature database. The Neo4j database platform stores data in a graph structure, storing nodes, attributes, and edges. Attributes and nodes are stored separately, and the relationships between attributes and nodes form edges, which greatly improves the performance of complex relationship analysis in graph databases. Knowledge graphs built using the Neo4j graph database platform provide the ability to analyze complex problems from a relational perspective, efficiently providing decision support for intelligent scheduling and recommendation assistance, and helping scheduling staff quickly extract valuable information from massive and diverse structured and unstructured data.

[0056] Using the Python programming language as an example, the following is an example of creating entity nodes:

[0057] CREATE(g:G en{name:′G en1′})RETURN g

[0058] The CREATE statement creates nodes. `g` is the node name, and `Gen` is the node's label, indicating that the node type is a generator node. The curly braces enclose the node's attributes. This generator node has a `name` attribute with a value of `Gen1`, indicating that this is the first generator set.

[0059] An example of creating an edge to establish a relationship between entities is shown below:

[0060] MATCH(g:G en{name:′G en1′})

[0061] MATCH(c:gen_cost2{name:'gen1_cost2',gencost2:0.3})

[0062] MERGE(c)-[:gen_cost2]→(g)

[0063] The `MATCH` statement is a query operation that matches a certain pattern, which can be a simple node or association, or a complex relational path. In the example, it matches two nodes: generator node `g` and generator cost coefficient node `c`. The `MERGE` statement creates a relationship between nodes, creating an edge from node `c` to node `g` with a relationship type of `gen_cost2`. Establishing this relationship facilitates efficient searching and querying of the generator cost coefficient for the first generator set.

[0064] In step S2:

[0065] The matching method used is to perform matching using the cosine similarity matching algorithm;

[0066] Cosine similarity measures the similarity between two vectors by measuring the cosine of the angle between their inner product spaces. It has been widely used in information retrieval, relation analysis, and data mining. Compared to commonly used metrics such as Euclidean distance, it is more suitable for evaluating the similarity between scheduling data on a given day and historical scheduling data.

[0067] When calculating the cosine similarity of the scheduling data, the nodes of the scheduling data for the scheduling day—including voltage, line power, load, generator output, air quality, temperature, relative humidity, wind direction, wind speed, and / or solar irradiance—are used to construct a scheduling data vector d for that day. The scheduling data for each day in the historical scheduling data is then used to construct a historical day scheduling data vector D. The historical day with the highest similarity is then identified by calculating the cosine similarity between vector d and vector D.

[0068] The expression for calculating cosine similarity is as follows:

[0069]

[0070] In the formula: variable d = [d1, d2, ..., d n ];D=[D1,D2,…,D n ]; n is the number of scheduling data nodes for similarity matching.

[0071] Based on the cosine similarity calculation formula, if the number of sampling points is 3, the cosine similarity space is as follows: Figure 2 As shown in the spatial diagram, if two vectors have the same direction, the cosine similarity is 1; if the two vectors have perpendicular or opposite directions, the cosine similarity is 0 and -1, respectively.

[0072] Specifically, cosine similarity is used to determine the directional differences between two high-dimensional vectors. Therefore, analyzing the cosine similarity between vector d and vector D, the closer the cosine similarity is to 1, the higher the similarity between the two vectors, and the higher the correlation of the scheduling data.

[0073] The similarity(d,D) obtained from the above steps is the first similarity dataset.

[0074] In step S3:

[0075] The matching method used is similarity matching based on the BM25 algorithm, which is based on a probability retrieval model.

[0076] in,

[0077] The nodes of the dispatch text data on the dispatch day—such as power generation guidance policies, maintenance plans, dispatch logs, or operating methods of power system dispatchers—are subjected to feature extraction and decomposition to generate several feature words k. i For each day's scheduling knowledge data in the historical scheduling knowledge data, construct historical daily scheduling data K, and calculate each feature k. i The correlation score with K. Finally, k i The relevance scores relative to K are weighted and summed to obtain k. i The correlation score between K and K.

[0078] Specifically, the BM25 algorithm can generally be simplified as follows:

[0079]

[0080] In the formula, k represents the scheduling text data for the scheduling day, K represents the scheduling knowledge data for a certain historical day, n represents the number of feature words in the scheduling text data for the scheduling day, and k i For the i-th feature word in the daily scheduling text data, W i R(k) represents the weight of the feature word. i K) is the correlation score between the feature word and the historical daily scheduling knowledge data.

[0081]

[0082] In the formula, N represents the total number of historical days, and df i Indicates the presence of feature word k i The number of historical days.

[0083]

[0084] L = a * (1 - c + c * dl / (avg_dl))

[0085] In the formula, a, b, and c are adjustment factors. a adjusts the influence of feature word frequency in the historical scheduling knowledge data on the matching score; b adjusts the influence of feature word frequency in the scheduling text data of the scheduling day on the matching score; and c adjusts the influence of the number of historical scheduling knowledge data points on the matching score. The values ​​of a, b, and c range from (0, 1), and a + b + c = 1. i Indicates the feature word k i Frequency of scheduling knowledge data in historical days; kf i Indicates the feature word k i The frequency of scheduling text data on the scheduling day; dl is the number of scheduling knowledge data on a historical day; avg_dl is the average number of scheduling knowledge data across all historical days.

[0086] The matching score Score(k, K) of the scheduling text data k for the scheduling day and the scheduling knowledge data Ki (i = 1, 2... N) for each historical day is obtained using the BM25 algorithm, which is the second similarity dataset.

[0087] In step S4, the predetermined time length is one day;

[0088] The normalization process is described in detail below:

[0089] After normalizing the first similarity dataset and the second similarity dataset according to the following formula, the similarity is fused according to the following formula to obtain the fused similarity dataset.

[0090] sim(i) = β*similarity i (d,D)+γ*Score i (k, K)

[0091] In the formula, sim(i) represents the fusion similarity value between the scheduling data and scheduling text data of the scheduling day and the scheduling data and scheduling knowledge data of the i-th historical day. The set of sim(i) for (i = 1, 2...N) is the fusion similarity dataset. β and γ are fusion factors, and β + γ = 1.

[0092] Step S5 specifically includes the following steps:

[0093] S501: Sort the data in the fusion similarity dataset from largest to smallest according to the fusion similarity value;

[0094] S502: Select multiple fusion similarity values ​​for a predetermined number of days according to the sorting.

[0095] S503: Based on multiple fusion similarity values ​​of the selected predetermined number of days, select the corresponding historical scheduling knowledge data for the predetermined number of days from the historical scheduling knowledge data of the power system historical scheduling knowledge graph;

[0096] The reservation period is 10 days or more; 10 days is preferred.

[0097] S504: From the historical scheduling knowledge data of the selected corresponding number of days, select power generation curve data one by one to form a power generation curve dataset for each historical scheduling knowledge data.

[0098] S505: Select the power of multiple time periods for each power generation curve in the power generation curve dataset and construct multiple daily power generation vectors.

[0099] S506: Construct a cluster sample set using the obtained multiple daily power generation vectors as samples to be clustered;

[0100] S507: Standardize the data of each daily power generation vector in the sample set to be clustered;

[0101] S508: Use the nearest neighbor propagation clustering algorithm to cluster multiple samples in the sample set to be clustered into one class, and calculate the cluster center of the class as the power generation curve.

[0102] In step S505, for each power generation curve, a time interval of 15 minutes is used to form 96 time periods; the power of each power generation curve in the 96 time periods is used to form a daily power generation vector, and n daily power generation vectors are used as samples to be clustered to construct a sample set to be clustered.

[0103] In step S507, the daily power generation vectors in the sample set to be clustered are preprocessed using the following formula for data standardization:

[0104]

[0105]

[0106]

[0107] Where, p′ kt Let be the generator active power value for the t-th time period on day k. Let S(p′) be the average active power of the generator in the t-th time period. t Let p be the standard deviation of the generator active power in the t-th time period. kt Let n be the generator active power value in the t-th time period on the k-th day after standardization preprocessing, where n is the total number of power generation days in the sample set to be clustered, and t ranges from 1 to 96.

[0108] Furthermore, based on the nearest neighbor propagation clustering algorithm, n samples in the set to be clustered are clustered into one class, and the cluster center of this class is calculated:

[0109] Among them, the nearest neighbor propagation clustering algorithm (AP) is a graph theory-based clustering algorithm. Its basic idea is to treat all samples to be clustered as nodes in a network, and each as a potential cluster center. The samples are connected by similarity lines to form a network (similarity matrix S). Then, through the transmission of messages (attractiveness and availability) along the edges of the network, the cluster centers of the sample set are calculated. A flowchart of the nearest neighbor propagation clustering algorithm is attached. Figure 3 As shown.

[0110] The similarity matrix S(j, h) is calculated based on the standardized and preprocessed sample set to be clustered. S(j, h) represents the ability of data point h to serve as the cluster center of data point j, and negative Euclidean distance is generally used.

[0111] S(j, h) = -||x j -x h || 2

[0112] For all samples to be clustered in the network, using the concept of adjacency matrices in graph theory, we can calculate the attraction matrix R(j, h) and the membership matrix A(j, h). Here, R(j, h) represents the attractiveness of each candidate cluster center h relative to other candidate cluster centers h′ to the sample j to be clustered, and A(j, h) represents the membership degree of each sample j to the candidate cluster center h.

[0113]

[0114]

[0115] A t+1 (j, h)=∑ j′≠h max{0,R t+1 (j′,h)}

[0116] To avoid oscillations in R(j,h) and A(j,h) during iterative calculations, a decay factor λ is introduced:

[0117] R t+1 (j, h) = (1-λ)R t+1 (j, h) + λR t (j, h)

[0118] A t+1 (j, h)=(1-λ)A t+1(j, h) + λA t (j, h)

[0119] The attenuation factor λ ranges from (0, 1).

[0120] The AP algorithm achieves clustering by iteratively updating the values ​​of the attraction matrix R(j,h) and the membership matrix A(j,h). When R(j,h) and A(j,h) reach stability or the maximum number of iterations is reached, the algorithm terminates and selects the sample to be clustered with the largest R(j,h) + A(j,h) as the cluster center c, which is then used as the power generation curve.

[0121] The aggregation results of the power generation curves in the above steps are shown in the attached figure. Figure 4 As shown in the figure, the horizontal axis represents the time period, and the vertical axis represents the power generation value of the generator set.

[0122] For example:

[0123] The scheduling knowledge, such as power generation plans, with the highest similarity over the ten days is searched in the historical knowledge graph. For the power generation plans with the highest similarity matching over the ten days, the time granularity is set to 15 minutes, that is, the total power generation demand of the generator units is stored in a time period of 15 minutes, resulting in 96 time periods per day. The power of each power generation curve in the 96 time periods constitutes a daily power generation vector, and the ten daily power generation vectors are used as samples to construct a clustering sample set. The nearest neighbor propagation clustering algorithm (AP) is used to perform cluster analysis on these ten curves, and the obtained cluster centers are used as the power generation curves for scheduling days to realize power system scheduling. The scheduling scheme is as follows: the entire scheduling day is divided into 96 time periods, each time period is 15 minutes, and the scheduling knowledge data of each time period of the cluster centers and / or Figure 4 The power curve in the data is used as the scheduling knowledge data for the corresponding time period of the scheduling day for power dispatching.

[0124] Another aspect of the present invention provides an electronic device, including a processor and a memory, the memory storing program instructions, characterized in that: the processor executes the program instructions to perform the power system dispatching method in the optional implementation described above.

[0125] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the embodiments of the present invention.

[0126] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A power system dispatching method characterized by, The method comprises the following steps: constructing a power system historical scheduling knowledge graph dataset, including historical scheduling data and historical scheduling knowledge data; matching the scheduling data of the scheduling day with the historical scheduling data to obtain a first similarity dataset of the scheduling data; matching the scheduling text data of the scheduling day with the historical scheduling knowledge data to obtain a second similarity dataset of the scheduling text data; normalizing the data in the two similarity datasets and fusing them in units of a predetermined time length to obtain a fused similarity dataset; wherein each data in the fused similarity dataset represents a fused similarity value corresponding to each predetermined time length; adopting a plurality of data ranked at the front in the fused similarity dataset, matching them one by one with the historical scheduling knowledge data to obtain a plurality of corresponding data subsets, and aggregating them into a generation curve of the scheduling day; regulating and controlling the power system according to the generation curve; The scheduling data comprises at least one of the following: voltage, line power, load, generator output, air quality, temperature, relative humidity, wind direction, wind speed, and solar radiation intensity. The scheduling knowledge data comprises scheduling text data and a generation curve; the scheduling text data comprises at least one of the following: generation guidance policy, maintenance plan, scheduling log, and operation method of regulation and control personnel.

2. The power system dispatching method of claim 1, wherein The method of constructing a power system historical scheduling knowledge graph dataset specifically comprises the following steps: obtaining historical scheduling data and historical scheduling knowledge data, which are stored in CSV files in units of days; creating entity node data and attribute data of the entity node by using the historical scheduling data and the historical scheduling knowledge data, and constructing an entity relationship edge data by using the relationship between the entity node data and the attribute data of the entity node, to complete the construction of the power system historical scheduling knowledge graph; adopting a graph database platform to store the entity node data, the attribute data of the entity node, and the relationship edge data between the entity nodes in the structure of a graph.

3. The power system dispatching method of claim 1, wherein In the step of matching the scheduling data of the scheduling day with the historical scheduling data to obtain a first similarity dataset of the scheduling data, the matching method used is cosine similarity matching.

4. The power system dispatching method of claim 1, wherein In the step of matching the scheduling text data of the scheduling day with the historical scheduling knowledge data to obtain a second similarity dataset of the scheduling knowledge data, the matching method used is similarity matching by using a BM25 algorithm based on a probabilistic retrieval model.

5. The power system dispatching method of claim 1, wherein The predetermined time length is one day.

6. The power system dispatching method of claim 1, wherein The method of adopting a plurality of data ranked at the front in the fused similarity dataset, matching them one by one with the historical scheduling knowledge data to obtain a plurality of corresponding data subsets, and aggregating them into a generation curve of the scheduling day specifically comprises the following steps: sorting the data in the fused similarity dataset according to the fused similarity values from large to small; selecting a plurality of fused similarity values of a predetermined number of days according to the sorting; selecting corresponding historical scheduling knowledge data of a predetermined number of days from the historical scheduling knowledge data of the power system historical scheduling knowledge graph according to the selected plurality of fused similarity values of a predetermined number of days; selecting, from the selected corresponding historical scheduling knowledge data of the predetermined days, power generation curve data sets each composed of power generation curve data of each historical scheduling knowledge data; selecting, from each power generation curve in the power generation curve data set, power of a plurality of time periods, and constructing a plurality of daily power generation power vectors; constructing a to-be-clustered sample set by taking the obtained plurality of daily power generation power vectors as to-be-clustered samples; performing data standardization processing on each daily power generation power vector in the to-be-clustered sample set; using a nearest neighbor propagation clustering algorithm to cluster the plurality of to-be-clustered samples in the to-be-clustered sample set into a class, and calculating a clustering center of the class as a power generation curve.

7. The power system dispatching method of claim 2, wherein The obtained power system historical scheduling data and historical scheduling knowledge are power system historical scheduling data and historical scheduling knowledge data of the previous year.

8. A power system dispatching device comprising a processor and a memory, the memory storing program instructions, characterized in that: The processor runs program instructions to implement the power system scheduling method of any one of claims 1 to 7.

Images