A large-scale evaluation method and system for distribution network projects based on big data

Abstract

The present invention belongs to the technical field of distribution network, and relates to a method and system for large-scale review of distribution network projects based on big data, the method comprising the following steps: collecting basic data and operation data of distribution network from business systems and data middle platforms through a data connection module; constructing a problem library in a problem library module, and collecting problems according to two types of equipment: distribution lines and distribution transformers; generating a demand library through a demand library module; data preprocessing; establishing a big data review model based on historical data, inputting the demand list of the project to be reviewed into the big data review model, and automatically calculating the review results through the big data review model; appeal process; expert manual review. The present invention establishes a complete set of processes of problem library-demand library-big data review-expert review, recommends expert review based on big data review results, can realize large-scale review of distribution network projects, greatly improves review efficiency and coverage, and effectively assists in accurate project establishment of distribution networks.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network, and more specifically, to a distribution network project review method and system based on big data. Background Art

[0002] The distribution network, as the "last mile" of the power system, extends to thousands of households. Its line length, number of equipment, and distribution locations are much higher than the main grid. Therefore, the number of new construction and renovation projects in the distribution network each year is far greater than that in other project areas. In the early stages of project establishment and evaluation, the distribution network faces more difficulties.

[0003] The distribution network project approval process is highly dependent on offline data. Usually, the line and equipment data in the planning area are collected and sorted manually offline, and the project approval needs of each power supply station are investigated. According to the urgency of the demand report, the site survey is carried out. Due to the limited number of manpower, it is often impossible to conduct on-site surveys for all the reported projects. Therefore, the supporting materials for the project approval are extremely dependent on the demand content reported by the power supply station.

[0004] Since the requirements submitted manually have not been uniformly verified, the integrity of the information and the quality of the data are often not guaranteed. In the review of power grid projects with higher voltage levels or other large projects, it is usually necessary to adopt a multi-expert, multi-review approach for a project to ensure the authenticity of the project establishment data and the rationality and necessity of the project. However, in the provincial-level review of medium and low voltage distribution network projects, due to the large number of projects, it is impossible to review them one by one, and only spot checks can be taken in different areas, and the proportion of spot checks is often less than 10%. The low review coverage rate leads to an increase in the proportion of problems in the later construction process, affecting the effectiveness of project implementation and causing waste of investment.

[0005] Improving the quality of distribution network project approval data and the coverage of preliminary review has a significant positive effect on improving its construction results. There are two main difficulties. First, the approval data is submitted manually offline, and it takes a lot of human resources to check them one by one. Second, at this stage, the approval of distribution network projects depends entirely on manual review. Manpower alone is not enough to analyze a large amount of project information. There is an urgent need for a data verification and review method that can cover all projects. By automatically processing the data information in the project approval process, batch screening of non-compliant issues can be carried out to assist manual large-scale project review. Summary of the invention

[0006] The purpose of the present invention is to provide a large-scale evaluation method and system for distribution network projects based on big data. From the perspective of distribution network project establishment, it includes a method for constructing a problem library and a demand library, a distribution network project evaluation system based on big data, and a complete set of push processes, and can complete batch project reserves through computer executable programs. Before the expert review, the full amount of project materials to be approved are automatically reviewed, and the project situation is evaluated from the dimensions of data authenticity, project establishment necessity, and project establishment rationality. Projects that do not meet the conditions for project establishment are screened to avoid unnecessary problems in the construction process that affect the effectiveness of project implementation, improve the accuracy of project establishment, and greatly reduce the workload of expert review and improve review efficiency.

[0007] The present invention is achieved through the following technical solutions.

[0008] A large-scale evaluation method for distribution network projects based on big data includes the following steps:

[0009] S1. Collect basic data and operation data of distribution network from business system and data center through data connection module;

[0010] S2. Build a problem library in the problem library module, collect problems according to the two types of equipment: distribution lines and distribution transformers. Before submitting each batch of projects, planning and design personnel will fill in the corresponding solutions in the problem library based on the on-site survey and current situation, that is, the problem is solved through operation and maintenance or project establishment; push the problems solved by operation and maintenance to the early warning work order and emergency repair work order of the intelligent power supply service command system, forming a closed loop in the early warning work order and emergency repair work order; push the problems solved by project establishment to the demand library; when all problems on a device are pushed to the next link, the problem cycle of the device is completed, and if new problems occur, recalculate;

[0011] S3. Generate a demand library through the demand library module. For the equipment and corresponding issues that need to be approved, the planning and design personnel fill in the project demand content. The status of the completed project demand is "pending approval". If the demand library accumulates multiple demands for the same equipment, the demand content is automatically bundled according to the equipment after completion, forming a "pending approval" project demand; the superior administrator account has approval authority, and the project demand approved by the superior administrator account is automatically pushed to the project review stage, and input into the big data review in the form of a project demand list; when all the demands on a device are pushed to the next stage, a demand cycle of the device is completed, and if new demands occur, they are recalculated;

[0012] S4, data preprocessing;

[0013] S5. Establish a big data review model based on historical data, input the list of project requirements to be reviewed into the big data review model, and automatically calculate the review results through the big data review model;

[0014] S6. Appeal process: Projects that fail the big data review in steps S4 to S5 can be divided into "incomplete information", "no necessity for project establishment" and "unreasonable project establishment" according to the review steps. Among them, projects with "no necessity for project establishment" do not provide an appeal channel and are not recommended for project establishment; projects with "incomplete information" and "unreasonable project establishment" are provided with an appeal channel. After rejection, the project can re-initiate the process from the problem library;

[0015] S7. Expert manual review: Projects that have passed the big data review will enter the expert manual review process. A spot check mechanism will be set up for the manual review. From the projects whose big data review results are "recommended for manual review", a number of projects that need to be spot checked will be selected, accounting for g%; from the projects whose big data review results are "passed", a number of projects that need to be spot checked will be selected, accounting for (100-g)%; the value g is determined by the review experts; after the corresponding number is determined, a list of spot-checked projects is automatically generated. The spot-checked projects shall be based on the manual review results, and the projects that are not spot-checked shall use the big data review results.

[0016] Further preferably, in the present invention, step S5 includes the following process:

[0017] S5-1, sample data set selection;

[0018] S5-2. Text processing: The sample data set stores unstructured data. To facilitate subsequent analysis, the fields used to describe the items are standardized.

[0019] S5-3, Project feature extraction: Step a. Taking 10kV line equipment and distribution transformer equipment as the main body, establish a feature equipment library, which is a project keyword set based on the standard equipment name; Step b. Use machine learning methods to extract and supplement keywords to obtain a keyword set based on the actual project library; Step c. Keyword classification and aggregation: Match the keyword set based on the standard equipment name in step a with the keyword set based on the actual project library in step b, and classify the keywords that can be used to describe the necessity of project establishment as standard equipment names; Step d. Feature structuring: Use the project name to associate with the main equipment to which it belongs, and count all the problems that occurred in the equipment during the evaluation time, including type, duration and number of times. The same type of problems can be divided into repeated, long-term and temporary problems. According to the project establishment orientation of each type of problem, feature items are counted; Combined with the project type classification in the project data, the equipment and problems are combined into second-order features;

[0020] S5-4, evaluation of the necessity of project establishment;

[0021] S5-5, evaluation of project rationality;

[0022] S5-6. Input the project to be reviewed into the big data review model, and obtain the big data review result through steps S5-2 to S5-5.

[0023] Further optimization, the process of evaluating the necessity of project establishment is as follows:

[0024] S5-41. Construct a decision tree model for evaluating the necessity of project establishment: Count the project establishment results under different combinations of equipment and problems, and construct a decision tree diagram based on the project type-equipment type-problem combination rules;

[0025] S5-42. Prune the decision tree;

[0026] S5-43. When the output results of the big data review model for each device are different, the combined output results are as follows:

[0027] ① When the output results of all devices included in the project are "passed", the combined output result is "passed review";

[0028] ② When the output results of each device included in the project are both "passed" and "failed", the combined output result is "manual review recommended";

[0029] ③ When the output results of all devices included in the project are "failed", the combined output result is "failed review".

[0030] S5-44. Extract projects from the historical project library and input them into the big data review model. Verify and correct the big data review model based on professional experience.

[0031] Further preferably, the decision tree pruning is:

[0032] When the result of a branch is underfitting, the branch should be pruned; the project approval rate for all equipment and problem combinations is calculated based on the project sample. When the project approval rate is lower than u1, it means that the branch has significant features that cannot pass the review, and the result is marked as failed; when the project approval rate is in the interval (u1, u2), it means that the branch does not improve the generalization performance of the model, that is, it does not have the ability to significantly distinguish the results. Applying the results of this branch will cause the model to underfit, and the branch should be pruned; when the project approval rate exceeds u2, it means that the branch has significant features that can pass the review, and the result is marked as passed. u1 is the lower limit of the project approval rate, and u2 is the upper limit of the project approval rate. The initialization parameters of u1 and u2 can be set to 0.25 and 0.75 respectively;

[0033] When the sample size is too small, if the number of passes and fails in a branch is less than the preset percentage threshold of the total number of projects, the branch is considered to have no significance in affecting the review results and is pruned.

[0034] Further preferably, the project rationality evaluation includes duplicate project evaluation and unreasonable expected results evaluation. The four additional items of project cost rationality, implementation carrying capacity, investment orientation fit and overall investment controllability are used as optional parameters of the big data review model, and are not used as the direct basis for whether the review is passed. Weights are configured and points are accumulated according to the overall planning direction of the project to be reviewed. When the points of the project additional items exceed the threshold, the output result "manual review is recommended" is

[0035] Further optimization, repeated project evaluation includes the following situations:

[0036] Complete duplication of project names: Compare the projects to be reviewed with those before the review by the same unit k For the historical project database that was officially approved in 2017, if there is a result in which the project name is completely repeated, the review result of the project will be "failed";

[0037] Partial duplication of project names: Compare the projects to be reviewed with those before the review by the same unit k Historical project database officially approved in the year, calculate the projects to be reviewed r and historical projects l The edit distance of the project to be reviewed is calculated r Percentage of string length or %, string similarity is 100- or When the similarity is greater than 95, the match is successful, that is, there is a suspected duplicate project. The equipment features of the project to be reviewed and the matching historical project are calculated through steps S5-2 and S5-3. If the feature overlap is greater than 50%, it is judged that there is a duplicate project, and the project review result is "failed".

[0038] Furthermore, the business system includes an intelligent power supply service command system, a power consumption information collection system, a distribution automation master station system, and a PMS2.0 system.

[0039] Furthermore, the problem categories in the problem library include distribution line tripping events, distribution line branch line failure and shutdown events, distribution line overload events, distribution lines not meeting communication requirements, unreasonable distribution line segmentation, distribution transformer overload events, substation outlet low voltage events, distribution transformer user low voltage events, and old high-loss transformers; and manually created problem categories are set.

[0040] Furthermore, the data preprocessing described in step S4 is: the requirement list of projects to be reviewed pushed to the project review stage through the problem library and the requirement library should contain fields for project name, project type, project content, equipment name, and problem details; for renovation projects, it is stipulated that each field in the requirement list of projects to be reviewed cannot be empty. If one field is empty, the big data review result of the project is "failed the review"; for new projects, it is stipulated that fields other than the equipment name cannot be empty. If one field is empty, the big data review result of the project is "failed the review"; the reason for failure to pass the review in this step is "incomplete information", and the information can be resubmitted in the problem library through the appeal process.

[0041] The present invention also provides a large-scale review system for distribution network projects based on big data, comprising a data connection module, a problem library module, a demand library module, a big data review module, an appeal module and a manual review module; the data connection module is used to connect the business system and the data middle platform, obtain the required data information, and push it to the problem library module and the big data review module; the problem library module calculates and generates various problem events based on basic data and operation data, and pushes them to the demand library module after manual filling; the demand library module is filled with demand content by planners, and a project demand list is formed after review by the superior administrator, and pushed to the big data review module; the big data review module trains the review model based on historical data and review rules, and forms a big data review result for the input project demand list; the appeal module provides an appeal channel for some projects that fail the review, and the materials can be returned to the problem library module through the appeal channel to resubmit the materials; the manual review module extracts projects according to the sampling rate customized by experts, and automatically outputs review opinions according to the big data review results and expert review results.

[0042] The present invention has the following characteristics:

[0043] First, establish an online submission process for project information, including the construction of a problem database and the generation of a demand database. The problem database is based on the business system or data center, and automatically collects problems based on equipment according to the established problem generation rules. The planning and design personnel of the reporting unit fill in the corresponding project establishment requirements based on the problem situation, reducing the workload of the fund collection process and ensuring the authenticity of the data. For some grassroots units that have project establishment requirements and cannot find problems in the system, the equipment can be manually entered into the problem database, and the requirements filled in the problem database will be automatically pushed to the demand database. All projects to be established will form a project requirement list through the demand database.

[0044] The second is to establish a big data review system for distribution network projects, establish a big data review model based on historical data, and calculate the review results after inputting the projects to be reviewed into the model. The sample data set comes from the historical project library, and the projects to be reviewed come from the project requirements list. Natural language processing technology is used to extract key information from the project content, and the key features of the project are obtained through feature project processing. The sample data is trained to establish a decision tree model, and some historical projects are extracted as test sets. After the model is corrected, the evaluation results of the data to be reviewed are calculated. After the big data review, a detailed list of projects that have passed, failed, and recommended for manual review is generated.

[0045] The third is to establish an appeal and manual review process. For projects that fail the big data review, they are classified into appealable and non-appealable categories according to the reasons for failure. Projects in the appealable category are given one opportunity to resubmit information from the question library. After submission, they will enter the big data review stage again. If the review passes, they will enter the manual review process. If the review fails, they will be included in the non-appealable category. Non-appealable projects are not recommended for project establishment in this batch of reviews. For projects that pass the big data review, a certain proportion of them will be randomly selected for manual review. The results of the manual review will be used for the projects that are randomly selected, and the results of the big data review will be used for the projects that are not randomly selected.

[0046] The present invention has the following advantages: first, a complete set of practical processes of problem library-demand library-big data review-expert review is established, and the data authenticity, information integrity and review coverage in the project establishment process are greatly improved; second, a big data review method for distribution network projects is designed, and all data are available based on existing systems. The algorithm has good practicality, strong professionalism and high review efficiency; third, expert review is recommended through big data review results, and there are fewer review links for projects that really need to be established and have high design quality, and the spot check rate of controversial projects is higher, which improves the overall spot check efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the large-scale evaluation system of distribution network projects based on big data of the present invention.

[0048] Figure 2 It is a diagram of a decision tree.

[0049] In the figure: 10-data connection module, 20-question library module, 30-demand library module, 40-big data review module, 50-appeal module, 60-manual review module. DETAILED DESCRIPTION

[0050] The present invention is further explained in detail below with reference to the accompanying drawings and embodiments.

[0051] Reference Figure 1 , a large-scale evaluation method for distribution network projects based on big data, comprising the following steps:

[0052] S1. Collect basic data and operation data of the distribution network from the business system and the data center through the data connection module 10; the business system includes the intelligent power supply service command system, the power consumption information collection system, the distribution automation master station system, and the PMS2.0 system.

[0053] S2. Construct a problem library in the problem library module 20. The problem library collects problems according to the two types of equipment: distribution lines and distribution transformers. All problems come from the business system or the data center. The problem categories in the problem library include distribution line tripping events, distribution line branch line failure shutdown events, distribution line overload events, distribution lines that do not meet the communication requirements, unreasonable distribution line segmentation, distribution transformer overload events, low voltage events at the substation outlet, low voltage events for distribution transformer users, and old high-loss transformers; and set up artificial new problem categories; before each batch of projects is submitted, the planning and design personnel will fill in the corresponding solutions in the problem library based on the on-site survey and the current situation, that is, the problem is solved through operation and maintenance or project establishment; the problems solved through operation and maintenance will be pushed to the early warning work order and emergency repair work order of the intelligent power supply service command system, forming a closed loop in the early warning work order and emergency repair work order; the problems solved through project establishment will be pushed to the demand library; when all problems on a device are pushed to the next link, the problem cycle of the device is completed, and if new problems occur, they will be recalculated.

[0054] S3. Generate a demand library through the demand library module 30. For the equipment and corresponding issues that need to be established, the planning and design personnel fill in the project demand content. The status of the completed project demand is "pending approval". If the demand library accumulates multiple demands for the same equipment, the demand content is automatically bundled according to the equipment after completion, forming an "pending approval" project demand; the superior administrator account has approval authority, and the project demands approved by the superior administrator account are automatically pushed to the project review link and input into the big data review in the form of a project demand list; when all the demands on a device are pushed to the next link, a demand cycle for the device is completed, and if new demands occur, they are recalculated.

[0055] S4. Data preprocessing: The list of project requirements to be reviewed that is pushed to the project review stage through the problem library and the requirements library should contain fields for project name (project quantity), project type, project content, equipment name, and problem details (problem type and quantity); for renovation projects, it is stipulated that the fields in the list of project requirements to be reviewed cannot be empty. If one field is empty, the big data review result of the project will be "failed the review"; for new projects, it is stipulated that the fields except the equipment name cannot be empty. If one field is empty, the big data review result of the project will be "failed the review"; the reason for failure to pass the review in this step is "incomplete information", and the information can be resubmitted in the problem library through the appeal process.

[0056] S5. Establish a big data review model based on historical data, input the list of project requirements to be reviewed into the big data review model, and automatically calculate the review results through the big data review model.

[0057] S6. Appeal process: Projects that fail the big data review in steps S4 to S5 can be divided into "incomplete information", "no necessity for project establishment" and "unreasonable project establishment" according to the review steps. Among them, projects with "no necessity for project establishment" do not provide an appeal channel and are not recommended for project establishment; projects with "incomplete information" and "unreasonable project establishment" are provided with an appeal channel, and after rejection, the project can re-initiate the process from the problem library.

[0058] S7. Expert manual review: Projects that pass the big data review will enter the expert manual review process. A spot check mechanism will be set up for manual review. From the projects with the big data review result of "recommended for manual review", a number of projects to be reviewed accounting for g% will be selected. From the projects with the big data review result of "passed", a number of projects to be reviewed accounting for (100-g)% will be selected. The value g is determined by the review experts. The projects to be randomly checked will be based on the manual review results, and the projects not to be randomly checked will use the big data review results.

[0059] In this embodiment, the question library includes the following question categories:

[0060] q1. Distribution line tripping event: Determine whether a tripping occurs based on the substation switch position information. If a distribution line trips once within a cycle, it will be included in the problem database and summarized and displayed according to the number of distribution line trips within the cycle.

[0061] q2. Distribution line branch fault shutdown event: determine whether there is a branch fault shutdown based on the distribution automation branch switch SOE information or the number of shutdown distribution transformers under the line. If the number of branch line faults connected to a distribution line exceeds the threshold within one cycle, it will be included in the problem library, and the number of branch line faults connected to each distribution line within the cycle will be summarized and displayed.

[0062] q3. Distribution line overload events: The occurrence of line overload events is determined based on whether the load current of the distribution line exceeds the safe operating current of the conductor. If a distribution line is overloaded once within one cycle, it will be included in the problem library and summarized and displayed according to the number of overloads of each distribution line within the cycle.

[0063] q4. The distribution line does not meet the interconnection requirements. According to the requirement that the overhead line should meet the single interconnection requirement, determine whether there is a single radial overhead line. Dynamically calculate within 1 cycle and display according to the most recent calculation result.

[0064] Q5. The distribution line segmentation is unreasonable. It is judged based on whether the distribution line is segmented and whether there are large branch lines (large branches are those with more than 30 distribution transformers or branch lines with a length of more than 10 km). For feeders without segmentation or with large branch lines, it is judged that the segmentation is unreasonable. It is dynamically calculated within 1 cycle and displayed according to the most recent calculation result.

[0065] q6. Distribution transformer overload event: The occurrence of distribution transformer overload event is determined based on whether the distribution transformer load power exceeds the capacity. If the distribution transformer is overloaded once within one cycle, it will be included in the problem library and summarized and displayed according to the number of overloads of each distribution transformer within the cycle.

[0066] q7. For low voltage events at the substation outlet, the occurrence of the event is determined based on whether the single-phase outlet voltage of the distribution transformer is less than 198V. If a low voltage occurs at the distribution substation outlet (low-voltage side of the distribution transformer) once within one cycle, it will be included in the problem database, and the number of low voltage events at the substation outlet of each distribution transformer within the cycle will be summarized and displayed.

[0067] q8. Low voltage event for distribution transformer users: The occurrence of the event is determined by whether the voltage at the user's metering point is less than 198V. If the sum of the number of low voltage events for all users connected to a distribution transformer within a cycle exceeds the threshold, it will be included in the problem library and the number of low voltage events for all users connected to each distribution transformer within the cycle will be summarized and displayed.

[0068] q9. Old high-loss transformers are determined based on the distribution transformer model and operating years. Distribution transformers with models of S9 and below, or with an operating life of more than 30 years, are judged to be old high-loss transformers. Dynamic calculations are performed within one cycle and the most recent calculation results are displayed.

[0069] Q10: For other issues that require project establishment but cannot be automatically collected online, set up a manually created issue category.

[0070] In this embodiment, step S5 includes the following process:

[0071] S5-1. Sample data set selection: After the demand library is established, historical project samples can be directly obtained from the demand library. When there is no demand library, all projects are directly entered into the project management system, and the projects that have been officially approved are pushed to the PMS2.0 system. Therefore, the project samples of the PMS2.0 system and the project management system are used as the historical project library. For the project samples that "passed the review", they can be obtained in the PMS2.0 system; for the project samples that "failed the review", the unapproved projects in the project management system can be selected, or based on the comparison between the project management system and the PMS2.0 system, the projects that are not in the PMS2.0 system can be selected.

[0072] S5-2. Text processing: The sample data set and project requirements store unstructured data. To facilitate subsequent analysis, the fields used to describe the project (project content or engineering quantity) must first be standardized.

[0073] 1) Text standardization: The same equipment may have multiple descriptions when filled in by different people. For example, "10kV", "10kv" and "10 kilovolts" describe the same meaning. When filled in incorrectly, "1kV", "0kV" and other forms may appear. During text preprocessing, common and ambiguous descriptions need to be standardized and replaced with standard descriptions using regularization methods.

[0074] 2) Word segmentation: Choose a suitable Chinese word segmentation tool to segment the project content and mark the parts of speech. You can choose methods such as jieba, SnowNLP, THULAC, NLPIR, etc., add the common writing of the equipment to the vocabulary, and correct the results after word segmentation. For example, words describing the main equipment such as "10kV overhead line", "distribution transformer", "pole-mounted circuit breaker", and "low-voltage line" should be ensured not to be separated; quote the public Chinese stop word list and remove stop words from the word segmentation results.

[0075] S5-3. Project feature extraction

[0076] Step a. With 10kV line equipment and distribution transformer equipment as the main body, establish a feature equipment library. The feature equipment library is a project keyword set based on the standard equipment name. The establishment of the feature equipment library can refer to the equipment list in the typical design, which should include the main equipment involved in the distribution network project and the equipment that affects the problem-solving attributes of the project, such as "10kV overhead line", "pole switch", "FTU", "lightning arrester", "distribution transformer", "JP cabinet", etc.

[0077] Step b. Use machine learning methods to extract and supplement keywords, and obtain a keyword set based on the actual project library: separate the words with numbers as the main body and the words with Chinese as the main body, for example, "100kVA" and "distribution transformer" are counted separately, and the digital part is recorded separately. The TF-IDF algorithm is used to analyze the frequency and weight of the Chinese vocabulary, as follows:

[0078] Calculate the frequency of words in the sample items TF :

[0079]

[0080] In the formula, TF i For vocabulary t i Frequency of words in the sample of items ,n i For vocabulary ti The number of times it appears in the sample of items, ∑ i n i is the total number of words in the Chinese vocabulary part of the project sample.

[0081] Calculate the inverse text frequency of a word IDF i :

[0082]

[0083] In the formula, IDF i For vocabulary t i The inverse text frequency, D is the total number of texts in the corpus, and the Jieba official corpus can be used for calculation, D i To include vocabulary t i The number of texts, calculated results idf i Used to indicate whether a word has the ability to distinguish.

[0084] pass TF - IDF = TF * IDF The operation filters out common words in the project sample and retains important words. After filtering, the top ranking words are obtained. x Item noun (equipment) is the keyword.

[0085] Step c. Keyword classification and collection: Since the project content is manually filled in, the equipment names in the project content are not standard equipment names. For example, the common writing of distribution transformers are "distribution transformer", "distribution station area", "station area", "transformer", "distribution transformer", "box transformer", "pole transformer", etc., but they actually describe the same equipment. Therefore, it is necessary to match the keyword set established based on the standard equipment name in step a with the keyword set established based on the actual project library in step b, and classify the keywords that can be used to describe the necessity of project establishment as standard equipment names. Standard equipment belonging to the line may include 10kV lines (including overhead and cable lines), switches (including primary equipment such as pole switches, ring network cabinets, disconnectors, and secondary equipment such as DTUs and FTUs), line lightning arresters, and insulators. Standard equipment belonging to distribution transformers may include distribution transformers, 10kV lines (supporting lines for distribution transformer construction), JP cabinets, and low-voltage lines.

[0086] Step d. Feature structuring: The project review result depends on the rationality and necessity of the project. The rationality of the project is a newly added indicator. This indicator is not considered for historical projects. The necessity of the project depends on the problems that the project needs to solve. Since the problem library has not been built when analyzing historical projects, it is necessary to obtain equipment records and historical problems from the data center or business system to evaluate the necessity of the project. The categories of historical problems that need to be obtained can refer to the content of the problem library construction. The time dimension involved is the year the project was issued and the previous year. Use the project name to associate with the main equipment, that is, the 10kV line or distribution transformer to which it belongs, and count all the problems that occurred in the equipment during the evaluation time, including type, duration and number. The same type of problems can be divided into categories such as repeated occurrence, long-term occurrence, and temporary problems. According to the project orientation of each type of problem, the feature items are counted.

[0087] In S5-2, the text processing of the project content is completed. Based on the keyword classification and aggregation in step c, the text processing results can be mapped to standard equipment names to form a key equipment set for each project. Combined with the project type classification in the project data (new construction of medium-voltage lines, transformation of medium-voltage lines, new construction of distribution transformers, and transformation of distribution transformers), the equipment and questions can be combined into a second-order m×n feature. Line equipment can be classified into 4 categories through dimensionality reduction: 10kV lines, switches, line arresters, and insulators. Distribution transformer equipment can be classified into 4 categories through dimensionality reduction: distribution transformers, 10kV lines, JP cabinets, and low-voltage lines. Therefore, m=4 in this embodiment. Similarly, in combination with the problem classification of the problem library, the problem characteristics can be reduced in dimension. Line problems can be further divided into line faults (distribution line tripping and distribution line branch line fault shutdown), distribution line overload, and grid problems (distribution lines do not meet the connection requirements, and distribution line segmentation is unreasonable). Distribution transformer problems can be further divided into distribution transformer body problems (distribution transformer overload, old high-loss transformer), low voltage problems at the substation outlet, and low voltage problems for users. That is, after the dimensionality reduction of line and distribution transformer problem characteristics, each contains 3 types of problems. Different equipment-problem combinations can be formed according to the presence or absence of a certain type of problem in each device.

[0088] S5-4. Evaluation of the necessity of project establishment

[0089] 1) Construct a decision tree model for evaluating the necessity of project establishment:

[0090] Considering that historical projects usually use the necessity of project establishment as the decisive factor for whether to pass the review, the project establishment results of historical project samples can be directly cited when evaluating the necessity of project establishment. The project establishment results under different combinations of equipment and problems are counted, and a decision tree diagram is constructed based on the project type-equipment type-problem combination rules. The decision tree construction method is as follows: Figure 2 shown.

[0091] 2) Pruning decision trees:

[0092] When the result of a branch is underfitting, the branch should be pruned. Based on the project samples, the project approval rate for all equipment and problem combinations is calculated. When the project approval rate is lower than u1, it means that the branch has significant features that cannot pass the review, and the result is marked as failed; when the project approval rate is in the interval (u1, u2), it means that the branch does not improve the generalization performance of the model, that is, it does not have the ability to significantly distinguish the results. Applying the results of this branch will cause the model to be underfitting, and the branch should be pruned; when the project approval rate reaches u2, it means that the branch has significant features that can pass the review, and the result is marked as passed. u1 is the lower limit of the project approval rate, and u2 is the upper limit of the project approval rate. The initialization parameters of u1 and u2 can be set to 0.25 and 0.75 respectively.

[0093] When the sample size is too small, if the number of passed and failed items in a branch is less than the preset percentage threshold of the total number of items, the branch is considered to have no significance in affecting the review results and is pruned. The initialization parameter of the preset percentage threshold can be set to 5%.

[0094] 3) Since the equipment involved in the project is not always a single device, it usually includes two or more types of equipment at the same time, that is, [equipment 1 equipment 2 equipment 3 equipment 4] = [1 1 0 1] or [1 0 0 1] and other 16 equipment combinations, so when the output results of the big data review model for each device are different, the combined output results are as follows:

[0095] ① When the output results of all devices included in the project are "passed", the combined output result is "passed review";

[0096] ② When the output results of each device included in the project are both "passed" and "failed", the combined output result is "manual review recommended";

[0097] ③ When the output results of all devices included in the project are "failed", the combined output result is "failed review".

[0098] 4) Extract projects from the historical project library and input them into the big data review model, and verify and correct the big data review model based on professional experience.

[0099] S5-5. Evaluation of project rationality

[0100] In the actual project construction and commissioning process, more and more problems are exposed due to unreasonable project establishment. Therefore, the evaluation of project rationality is added to the big data review model.

[0101] 1) Repeated project evaluation

[0102] In principle, the same device k ( k≥1) Duplicate project establishment is not allowed within the year.

[0103] ① The project names are completely repeated. Compare the project to be reviewed with the previous review of the same county company. k In the historical project database that was officially approved in the year, if there is a result in which the project name is completely repeated, the review result of the project will be "failed".

[0104] ② Part of the project names are repeated. Compare the projects to be reviewed with the projects before the review by the same county company. k Historical project database officially approved in the year, calculate the projects to be reviewed r and historical projects l The edit distance ( Levenshtein distance ), that is, the minimum number of editing operations required to transform one string into another. The calculation method is as follows:

[0105]

[0106] In the formula lev r,l ( i , j ) is the minimum edit distance, representing r Before i Characters change to l Before j The distance of characters relative to r Change to l For example, lev r,l ( i -1, j ) +1 means r Remove 1 character to match l , lev r,l ( i , j -1) +1 means r Add 1 character to match l , lev r,l ( i -1, j -1) +1 (ri≠lj) Indicates a match or no match, depending on r i ≠ l j Is it true?

[0107] calculate lev r,l ( i , j ) Projects awaiting review r Percentage of string length or%, string similarity is 100- or , the similarity is greater than p When the match is successful, there is a suspected duplicate project, it is recommended p =95, calculate the equipment features of the project to be reviewed and the matching historical project through steps S5-2 and S5-3. If the feature overlap is greater than 50%, it is judged that there is duplicate project establishment, and the project review result is "failed".

[0108] 2) Unreasonable evaluation of expected results

[0109] The effectiveness study of the commissioned projects shows that when the distribution transformer itself is not overloaded, the effect of simply replacing the distribution transformer (without supporting projects) in the project is poor. The "unreasonable evaluation of expected results" is configured as an optional parameter of the big data review model, that is, "unreasonable evaluation of expected results" = 0 (not calculated) or 1 (calculated), and the initial value defaults to 1.

[0110] 3) Other comments

[0111] In project review, there are additional concerns such as the rationality of project cost, implementation carrying capacity, fit with investment orientation, and overall investment controllability. Additional items are optional parameters of the big data review model and are not used as a direct basis for whether the review is passed. Weights and points can be configured according to the overall planning direction of the project to be reviewed. When the score of the project additional items exceeds the threshold, the output result "manual review recommended" is

[0112] S5-6. Input the project to be reviewed into the big data review model, and obtain the big data review result through steps S5-2 to S5-5.

[0113] In order to implement the above method, the present invention provides a large-scale evaluation system for distribution network projects, such as Figure 1 As shown. It includes a data connection module 10, a problem library module 20, a demand library module 30, a big data review module 40, an appeal module 50 and a manual review module 60. The data connection module 10 is used to connect the business system and the data center, obtain the required data information, and push it to the problem library module 20 and the big data review module 40; the problem library module 20 calculates and generates various problem events based on basic data and operation data, and pushes them to the demand library module after manual filling; the demand library module 30 is filled with demand content by planners, and a project demand list is formed after review by the superior administrator, and pushed to the big data review module 40; the big data review module 40 trains the review model based on historical data and review rules, and forms a big data review result for the input project demand list; the appeal module 50 provides an appeal channel for some projects that do not pass the review, and returns the problem library module through the appeal channel to resubmit the information; the manual review module 60 extracts projects according to the expert-defined sampling rate, and automatically outputs the review opinions according to the big data review results and expert review results.

Claims

1. A large-scale evaluation method for distribution network projects based on big data, characterized in that: The following steps are involved: S1. Collect basic data and operation data of distribution network from business system and data center through data connection module; S2. Build a problem library in the problem library module, and collect problems according to the two types of equipment: distribution lines and distribution transformers. Before submitting each batch of projects, planning and design personnel will fill in the corresponding solutions in the problem library based on the on-site survey and current situation, that is, the problem will be solved through operation and maintenance or project establishment. The problems solved through operation and maintenance will be pushed to the early warning work order and emergency repair work order of the intelligent power supply service command system, forming a closed loop in the early warning work order and emergency repair work order. Push the issues solved through project establishment to the demand library; When all problems on a device are pushed to the next stage, a problem cycle of the device is completed. If a new problem occurs, it will be recalculated; S3. Generate a demand library through the demand library module. For the equipment and corresponding issues that need to be established, planning and design personnel fill in the project demand content. The status of the completed project demand is "pending approval". If the demand library accumulates multiple demands for the same equipment, they are automatically bundled according to the equipment after the demand content is completed, forming a "pending approval" project demand; The superior administrator account has approval authority. The project requirements approved by the superior administrator account are automatically pushed to the project review stage and input into the big data review in the form of a project requirement list. When all requirements on a device are pushed to the next stage, a requirement cycle of the device is completed. If new requirements arise, they will be recalculated. S4, data preprocessing; S5. Establish a big data review model based on historical data, input the list of project requirements to be reviewed into the big data review model, and automatically calculate the review results through the big data review model; S6. Appeal process: For projects that fail the big data review in steps S4~S5, they are divided into "incomplete information", "no necessity for project establishment" and "unreasonable project establishment" according to the review steps. Among them, there is no appeal channel for projects with "no necessity for project establishment" and it is not recommended to establish the project; for projects with "incomplete information" and "unreasonable project establishment", an appeal channel is provided. After rejection, the project can re-initiate the process from the problem library; S7. Expert manual review: Projects that pass the big data review will enter the expert manual review process. A spot check mechanism will be set up for manual review. From projects with a big data review result of "recommended for manual review", a number of items that need to be spot checked will be selected, accounting for g%; from projects with a big data review result of "passed", a number of items that need to be spot checked will be selected, accounting for (100-g)%; the value g is determined by the review experts; after the corresponding number is determined, a list of spot check items will be automatically generated. The items to be spot checked will be based on the manual review results, and the items not to be spot checked will be based on the big data review results; Step S5 includes the following process: S5-1, sample data set selection; S5-2. Text processing: The sample data set stores unstructured data. To facilitate subsequent analysis, the fields used to describe the items are standardized. S5-3, Project feature extraction: Step a. Taking 10kV line equipment and distribution transformer equipment as the main body, establish a feature equipment library, which is a project keyword set based on the standard equipment name; Step b. Use machine learning methods to extract and supplement keywords to obtain a keyword set based on the actual project library; Step c. Keyword classification and aggregation: Match the keyword set based on the standard equipment name in step a with the keyword set based on the actual project library in step b, and classify the keywords that can be used to describe the necessity of project establishment as standard equipment names; Step d. Feature structuring: Use the project name to associate with the main equipment to which it belongs, and count all the problems that occurred in the equipment during the evaluation time, including type, duration and number of times. The same type of problems are divided into repeated, long-term and temporary problems, and the feature items are counted according to the project establishment orientation of each type of problem; Combine the project type classification in the project data to combine the equipment and problems into second-order features; S5-4, evaluation of the necessity of project establishment; S5-41. Construct a decision tree model for evaluating the necessity of project establishment: Count the project establishment results under different combinations of equipment and problems, and construct a decision tree diagram based on the project type-equipment type-problem combination rules; S5-42. Pruning the decision tree; the pruning of the decision tree refers to: When the result of a branch is underfitting, the branch should be pruned. The project approval rate for all equipment and problem combinations is calculated based on the project sample. When the project approval rate is lower than u1, it means that the branch has significant features that cannot pass the review, and the result is marked as failed; when the project approval rate is in the interval (u1, u2), it means that the branch does not improve the generalization performance of the model, that is, it does not have the ability to significantly distinguish the results. Applying the results of this branch will cause the model to underfit, and the branch should be pruned; when the project approval rate reaches u2, it means that the branch has significant features that can pass the review, and the result is marked as passed; u1 is the lower limit of the project approval rate, and u2 is the upper limit of the project approval rate; When the sample size is too small, if the number of passed and failed items in a branch is less than the preset percentage threshold of the total number of items, the branch is considered to have no significance in affecting the review results and is pruned; S5-43. When the output results of the big data review model for each device are different, the combined output results are as follows: ① When the output results of all devices included in the project are "passed", the combined output result is "passed review"; ② When the output results of each device included in the project have both "pass" and "fail", the combined output result is "recommended for manual review"; ③ When the output results of all devices included in the project are "failed", the combined output result is "failed review"; S5-44. Extract projects from the historical project library and input them into the big data review model, and verify and correct the big data review model based on professional experience; S5-5, evaluation of project rationality; S5-6. Input the project to be reviewed into the big data review model, and obtain the big data review result through steps S5-2 to S5-5.

2. The method for large-scale evaluation of distribution network projects based on big data according to claim 1 is characterized in that: The project rationality evaluation includes duplicate project evaluation and unreasonable expected results evaluation. The four additional items of project cost rationality, implementation carrying capacity, investment orientation fit, and overall investment controllability are used as optional parameters of the big data review model. They are not used as the direct basis for whether the review is passed. Weights are configured and points are accumulated according to the overall planning direction of the project to be reviewed. When the points of the project additional items exceed the threshold, the output result "manual review recommended" is output.

3. The method for large-scale evaluation of distribution network projects based on big data according to claim 2 is characterized in that: Repeated project evaluation includes the following situations: Complete duplication of project names: Compare the projects to be reviewed with those before the review by the same unit k For the historical project database that was officially approved in 2017, if there is a result in which the project name is completely repeated, the review result of the project will be "failed"; Partial duplication of project names: Compare the projects to be reviewed with those before the review by the same unit k Historical project database officially approved in the year, calculate the projects to be reviewed r and historical projects l The edit distance of the project to be reviewed is calculated r Percentage of string length η %, string similarity is 100- η When the similarity is greater than 95, the match is successful, that is, there is a suspected duplicate project. The equipment features of the project to be reviewed and the matching historical project are calculated through steps S5-2 and S5-3. If the feature overlap is greater than 50%, it is judged that there is a duplicate project, and the project review result is "failed".

4. The method for large-scale evaluation of distribution network projects based on big data according to claim 1 is characterized in that: The business system includes an intelligent power supply service command system, a power consumption information collection system, a distribution automation master station system, and a PMS2.0 system.

5. The method for large-scale evaluation of distribution network projects based on big data according to claim 1 is characterized in that: The problem categories of the problem database include distribution line tripping events, distribution line branch line failure shutdown events, distribution line overload events, distribution line failure to meet communication requirements, unreasonable distribution line segmentation, distribution transformer overload events, low voltage events at the substation outlet, low voltage events for distribution transformer users, and old high-loss transformers; And set up manually created question categories.

6. The method for large-scale evaluation of distribution network projects based on big data according to claim 1 is characterized in that: The data preprocessing in step S4 is: the list of project requirements to be reviewed pushed to the project review stage through the problem library and the requirements library should contain fields such as project name, project type, project content, equipment name, and problem details; For renovation projects, it is stipulated that the fields in the list of project requirements to be reviewed cannot be empty. If one field is empty, the big data review result of the project will be "failed the review"; for new projects, it is stipulated that the fields other than the equipment name cannot be empty. If one field is empty, the big data review result of the project will be "failed the review"; if the reason for failure to pass the review is "incomplete information", the information will be resubmitted in the problem library through the appeal process.

7. A large-scale evaluation system for distribution network projects based on big data, used to implement the large-scale evaluation method for distribution network projects based on big data according to any one of claims 1 to 6, characterized in that: It includes data connection module, problem library module, demand library module, big data review module, appeal module and manual review module; the data connection module is used to connect the business system and the data center, obtain the required data information, and push it to the problem library module and the big data review module; the problem library module calculates and generates various problem events based on basic data and operation data, and pushes them to the demand library module after manual reporting; the demand library module is filled in by planners with demand content, and a project demand list is formed after review by the superior administrator and pushed to the big data review module; the big data review module trains the review model based on historical data and review rules, and forms a review result for the input project demand list; the appeal module provides an appeal channel for some projects that fail the review, and returns the materials to the problem library module through the appeal channel for resubmission; The manual review module extracts items according to the expert-defined sampling rate, and automatically outputs review opinions based on the big data review results and expert review results.

Images