Automatic error correction method for filling and reporting data acquisition items of special equipment

Abstract

The invention relates to the field of special equipment data processing, in particular to a special equipment data acquisition item filling and reporting automatic error correction method. The invention aims to solve the problems of low manual auditing efficiency, strong subjectivity and high error-prone performance in the special equipment data acquisition item filling and reporting process, and through the automatic error correction method, the data auditing efficiency and accuracy are improved, the labor cost is reduced, the data normalization and reliability are ensured, and the data auditing efficiency and accuracy are improved. And more powerful data support is provided for management and safe operation of special equipment. According to the technical scheme, the special equipment data acquisition item filling and reporting automatic error correction method comprises the following steps of 1, preparing data; 2, converting the data into semantic vectors; step 3, calculating semantic similarity; 4, converting the data into font vectors; 5, calculating font similarity; 6, calculating the total similarity; and 7, automatically correcting errors.

Description

Technical Field

[0001] The present invention relates to the field of special equipment data processing, and in particular to an automatic error correction method for filling in special equipment data collection items. Background Art

[0002] In terms of filling in special equipment data collection items, manual review is currently usually relied on to judge the accuracy and standardization of the reported data. The general process is for inspectors or relevant staff to check the collected data one by one against the established standards and specifications to determine whether it meets the requirements.

[0003] This method has the following problems and disadvantages:

[0004] 1. Inefficiency: Manual review takes a lot of time and manpower, especially when faced with a large number of special equipment data collection items. The review work may be backlogged, resulting in untimely data processing and affecting the overall work progress;

[0005] 2. High subjectivity: Different auditors may have different understandings of standards and specifications, which can easily lead to inconsistency in audit results and reduce the accuracy and reliability of data audits;

[0006] 3. High error-proneness: Long-term manual review can easily make auditors tired, thereby increasing the risk of misjudgment and omission, and may miss some incorrectly reported data or issue false warnings for correct data. Summary of the invention

[0007] The purpose of the present invention is to overcome the deficiencies in the above-mentioned background technology and to provide a method for automatic error correction in filling in special equipment data collection items, which method should be able to improve the efficiency and accuracy of special equipment data review.

[0008] The technical solution of the present invention is:

[0009] A method for automatically correcting errors in filling in special equipment data collection items comprises the following steps:

[0010] Step 1: Prepare data;

[0011] Step 2: Convert data into semantic vectors;

[0012] Step 3: Calculate semantic similarity;

[0013] Step 4: Convert data into glyph vectors;

[0014] Step 5: Calculate the glyph similarity;

[0015] Step 6: Calculate the total similarity;

[0016] Step 7: Automatic error correction.

[0017] The step 1 includes: 1) setting collection items: the collection items are the names of the data that need to be filled in for special equipment data and are prone to filling errors; 2) setting early warning group data: the early warning group data is a group of data that corresponds to the collection items and is filled in correctly; 3) extracting the data to be warned: the data filled in by the collection and inspection personnel in the collection items are used as the data to be warned.

[0018] The step 2 includes: for the same collection item, converting the data to be warned into a vector form, and converting the warning group data into a vector form.

[0019] The step 3 includes: calculating the cosine similarity between the data to be warned and the warning group data.

[0020] The step 4 includes: 1) converting the character library into a picture, and then converting the picture into a vector form, and establishing a relationship dictionary between characters and vectors; 2) converting each character of the data to be warned and the warning group data into a picture vector according to the relationship dictionary.

[0021] The step 5 includes: 1) calculating the cosine similarity of each character of the data to be warned and the warning group data according to the picture vector; 2) integrating all cosine similarities to calculate the glyph similarity of the data to be warned and the warning group data.

[0022] The step 6 comprises: calculating the total similarity between the data to be warned and the data to be compared respectively according to the semantic similarity and the glyph similarity.

[0023] The step 7 includes: 1) setting a similarity threshold; 2) error correction judgment: if the total similarity is greater than the similarity threshold, it means that the collection item meets the reporting requirements; otherwise, it means that the collection item does not meet the reporting requirements, and select the warning group data with the highest total similarity or multiple warning group data with higher total similarities for prompting.

[0024] The beneficial effects of the present invention are:

[0025] 1. Improve work efficiency: Compared with manual review, the present invention can quickly process a large number of special equipment data collection items, greatly shorten the time of data review, improve work efficiency, and enable data to be processed and applied in a timely manner;

[0026] 2. Improve data accuracy: Through the calculation of semantic and glyph multi-dimensional similarity, timely warning and accurate recommendation can be achieved, which helps users to quickly correct erroneous data, reduce misjudgment and missed judgments, improve the accuracy of data judgment, ensure the standardization and reliability of data, and provide more accurate data support for the management and safe operation of special equipment.

[0027] 3. Reduce labor costs: Automated error correction and recommendation functions reduce reliance on manual review, saving a lot of labor costs and allowing staff to devote more energy to other important work. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of the present invention.

[0029] Figure 2 This is an error correction effect diagram of Example 1.

[0030] Figure 3 This is an error correction effect diagram of Example 2. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0032] like Figure 1 As shown, a method for automatically correcting errors in filling in special equipment data collection items includes the following steps:

[0033] Step 1: Prepare the data

[0034] The data includes data to be warned and warning group data.

[0035] 1) Set the collection items

[0036] The collection items are the names of data that require special equipment data to be filled in and are prone to errors in filling in.

[0037] 2) Set the warning group data

[0038] The early warning group data is a group of data that corresponds to the collection items and is correctly reported.

[0039] 3) Extract the data to be warned

[0040] The data filled in by the collection and inspection personnel in the collection items shall be used as data to be warned.

[0041] Since different special equipment has different collection items, the present invention only takes the collection item data that is easy to fill in errors as the data to be warned, and then prepares a group of correctly filled data for the data to be warned as the warning group data. The warning group data is determined based on past experience, standard specifications and possible reasonable data ranges. The warning group data is used for comparative analysis with the data to be warned. Both the data to be warned and the warning group data are string type.

[0042] Step 2: Convert data into semantic vectors

[0043] For the same collection item, the data to be warned is converted into vector form, and the warning group data is converted into vector form. The purpose is to convert character data into vector form that can be processed and calculated by computer, so as to facilitate the subsequent similarity calculation.

[0044] The data to be warned and the warning group data are converted into semantic vectors through the embedding model. The semantic vector of the data to be warned is recorded as A, and the semantic vector of the warning group data is recorded as B.

[0045] Step 3: Calculate semantic similarity

[0046] Calculate the cosine similarity between the data to be warned and the warning group data.

[0047] For the same collection item, the cosine similarity of the data to be warned and the warning group data to be compared is calculated based on the semantic vector. The cosine similarity is the semantic similarity, and the calculation formula is:

[0048]

[0049] Among them, consine similarity(A,B) is the cosine similarity, A is the semantic vector of the data to be warned, B is the semantic vector of the warning group data, A i are the components in the semantic vector of the data to be warned, B i are the components in the semantic vector of the warning group data, and i is the sequence number of the components in the semantic vector of a group of data to be warned and the warning group data.

[0050] Traverse the warning group data to obtain the semantic similarity between the warning data to be warned of each collection item and the warning group data to be compared, denoted as S ak , k = 1, 2…n, k is the data sequence number of the warning group data. Through this step, the semantic closeness between the data to be warned and the warning group data can be quantitatively evaluated.

[0051] Step 4: Convert data into glyph vectors

[0052] 1) Convert the character library into images, then convert the images into vectors, and establish a dictionary of the relationship between characters and vectors;

[0053] The font library includes Chinese characters, commonly used trademark symbols and English letters.

[0054] 2) Convert each character of the data to be warned and the warning group data into a picture vector according to the relational dictionary;

[0055] The image vector of the character to be warned is A j , the picture vector of the character of the warning group data is B j, j is the character position sequence number of the data to be warned and the warning group data.

[0056] Step 5: Calculate glyph similarity

[0057] Calculate the glyph similarity between the data to be warned and the warning group data.

[0058] For the same collection item, the cosine similarity of the data to be warned and the warning group data to be compared is calculated based on the image vector. The cosine similarity is the similarity of the glyphs. The calculation steps include:

[0059] 1) Calculate the cosine similarity of characters

[0060] According to the image vector, the cosine similarity of the characters at the same position in the warning data to be warned and the warning group data to be compared is calculated one by one;

[0061] If there is an empty character at the same position (the character at that position of the data to be warned is empty or the character at that position of the warning group data is empty), the cosine similarity of the character is 0;

[0062] 2) Calculate glyph similarity

[0063] By integrating the cosine similarity, we can obtain the glyph similarity between the data to be warned and the warning group data to be compared, that is, the glyph similarity. The calculation formula is:

[0064]

[0065] Where: S ABj is the cosine similarity of the characters, and j is the character position sequence number of the data to be warned and the warning group data.

[0066] Traverse and obtain the similarity of the shape of the data to be warned of each collection item and the data to be warned for comparison, denoted as S bk , k = 1, 2…n, k is the data sequence number in the warning group data. Through this step, the warning data and the warning group data can be compared and analyzed from the perspective of glyphs, further supplementing the judgment dimension of data similarity.

[0067] Step 6: Calculate the total similarity

[0068] According to the semantic similarity S ak Similarity with the glyph S bk Calculate the total similarity between the data to be warned and the warning group data to be compared.

[0069] The calculation formula is:

[0070] S k =W a S ak +W b Sbk

[0071] Where: S k is the total similarity, W a is the weight of semantic similarity, W b is the weight of the glyph similarity, W a +W b =1; k=1,2…n, k is the data sequence number in the early warning group data.

[0072] Step 7: Automatic error correction

[0073] 1) Set the similarity threshold S ′ ;

[0074] 2) Error correction judgment

[0075] If there is a total similarity S i Greater than the similarity threshold S ′ , indicating that the collection item meets the reporting requirements; otherwise, it means that the collection item does not meet the reporting requirements, and the warning group data with the highest total similarity or multiple warning group data with high total similarity are selected for prompting.

[0076] Example 1

[0077] The type of special equipment is a hoist cage, and one of the collection items that is easy to fill in incorrectly is "transmission system".

[0078] The warning group data of this collection item is set to {“single transmission”, “parallel double transmission”, “parallel triple transmission”}.

[0079] The data collection and inspection personnel shall fill in “Parallel transmission” in the collection item and take “Parallel transmission” as the data to be warned.

[0080] bge-large-zh-v1.5 is used to convert the data to be warned and the warning group data into semantic vectors.

[0081] Use pygame to convert a font library with a large number of Chinese characters (no less than 3500), commonly used trademark symbols and English letters into black and white pictures, and then use opencv to convert the pictures into graphic vectors.

[0082] The weight of semantic similarity W a =0.8, the weight of the glyph similarity is W b =0.2, similarity threshold S ′ =0.8.

[0083] The total similarity is as follows:

[0084] The total similarity between “parallel transmission” and “single transmission” is 0.52, which is less than the similarity threshold.

[0085] The total similarity between “parallel transmission” and “parallel dual transmission” is 0.71, which is less than the similarity threshold.

[0086] The total similarity between “parallel transmission” and “parallel triple transmission” is 0.72, which is less than the similarity threshold and has the highest similarity in the warning group.

[0087] like Figure 2 As shown, a prompt is given: the collection item does not meet the reporting requirements, warning! The recommended possible correct reporting value is "parallel three-drive".

[0088] Example 2

[0089] The type of special equipment is a cage, and one of the collection items that is easy to fill in incorrectly is "shell-side medium".

[0090] The warning group data of this collection item is set to {“oxygen”, “air”, “natural gas”, “mixed gas”}.

[0091] The collection and inspection personnel shall fill in "oxygen vapor" in the collection item and use "oxygen vapor" as the data to be warned.

[0092] The weight of semantic similarity W a =0.8, the weight of the glyph similarity is W b =0.2, similarity threshold S ′ =0.8.

[0093] The total similarity is as follows:

[0094] The total similarity between “oxygen vapor” and “oxygen” is 0.74, which is less than the similarity threshold and has the highest similarity in the warning group.

[0095] The total similarity between "oxygen vapor" and "air" is 0.63, which is less than the similarity threshold.

[0096] The total similarity between "oxygen vapor" and "natural gas" is 0.44, which is less than the similarity threshold.

[0097] The total similarity between "oxygen vapor" and "mixed gas" is 0.46, which is less than the similarity threshold.

[0098] like Figure 3 As shown, a prompt is given: the collection item does not meet the reporting requirements! It may be "oxygen".

[0099] The key technical points of the present invention are as follows:

[0100] 1. Application of data vector conversion model: Use a specific embedding model (such as bge-large-zh-v1.5) to convert text data into vector representation, which is the basis for subsequent similarity calculation. The selection and application of the model has an important impact on the accuracy of the entire automatic error correction method.

[0101] 2. Semantic and glyph similarity calculation method: The unique cosine similarity-based semantic and glyph similarity calculation method, especially the glyph similarity calculation through image vector conversion, and the method of combining the two to calculate the total similarity, can accurately quantify the similarity between data, thereby effectively judging the rationality of the data.

[0102] 3. Dynamic weight setting: Set the weights of semantic similarity and glyph similarity according to the different types of data items, so that the automatic error correction method can adapt to the characteristics of different types of data and improve the accuracy and flexibility of judgment.

[0103] The preferred embodiments of the present invention are shown in the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in this specification. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

Claims

1. A method for automatically correcting errors in filling in special equipment data collection items, comprising the following steps: Step 1: Prepare data; Step 2: Convert data into semantic vectors; Step 3: Calculate semantic similarity; Step 4: Convert data into glyph vectors; Step 5: Calculate the glyph similarity; Step 6: Calculate the total similarity; Step 7: Automatic error correction.

2. The automatic error correction method for filling in special equipment data collection items according to claim 1 is characterized by: The step 1 includes: 1) setting collection items: the collection items are the names of the data that need to be filled in for special equipment data and are prone to filling errors; 2) setting early warning group data: the early warning group data is a group of data that corresponds to the collection items and is filled in correctly; 3) extracting the data to be warned: the data filled in by the collection and inspection personnel in the collection items are used as the data to be warned.

3. The automatic error correction method for filling in special equipment data collection items according to claim 2 is characterized by: The step 2 includes: for the same collection item, converting the data to be warned into a vector form, and converting the warning group data into a vector form.

4. The automatic error correction method for filling in special equipment data collection items according to claim 3 is characterized by: The step 3 includes: calculating the cosine similarity between the data to be warned and the warning group data.

5. The automatic error correction method for filling in special equipment data collection items according to claim 4 is characterized by: The step 4 includes: 1) converting the character library into a picture, and then converting the picture into a vector form, and establishing a relationship dictionary between characters and vectors; 2) converting each character of the data to be warned and the warning group data into a picture vector according to the relationship dictionary.

6. The automatic error correction method for filling in special equipment data collection items according to claim 5 is characterized by: The step 5 includes: 1) calculating the cosine similarity of each character of the data to be warned and the warning group data according to the picture vector; 2) integrating all cosine similarities to calculate the glyph similarity of the data to be warned and the warning group data.

7. The automatic error correction method for filling in special equipment data collection items according to claim 6 is characterized by: The step 6 comprises: calculating the total similarity between the data to be warned and the data to be compared respectively according to the semantic similarity and the glyph similarity.

8. The automatic error correction method for filling in special equipment data collection items according to claim 7 is characterized by: The step 7 includes: 1) setting a similarity threshold; 2) error correction judgment: if the total similarity is greater than the similarity threshold, it means that the collection item meets the reporting requirements; otherwise, it means that the collection item does not meet the reporting requirements, and select the warning group data with the highest total similarity or multiple warning group data with higher total similarities for prompting.

Images