Neural network-based auditing method and device for subsection settlement of power transmission and transformation project

Abstract

The invention discloses a power transmission and transformation project subsection settlement auditing method and device based on a neural network, and belongs to the technical field of power transmission and transformation project auditing management. The method comprises the following steps: obtaining settlement data of a power transmission and transformation project; performing data cleaning processing on the settlement data, and performing text analysis; converting characters in the settlement data into a vector form by using a word embedding technology, and expressing in a sequence data form to obtain sequence settlement data; dividing the sequence settlement data into a training set and a verification set, training a preset initialized power transmission and transformation project review model by using the training set, and performing model optimization iteration by taking the accuracy of the maximum settlement review task as a target to obtain a trained power transmission and transformation project review model; and using the trained power transmission and transformation project review model to intelligently review the new power transmission and transformation project settlement document, and outputting abnormal data to a review report. According to the method, the subsection settlement auditing efficiency of the power transmission and transformation project is remarkably improved, and a large amount of labor cost is saved.

Description

Technical Field

[0001] The invention relates to a method and device for auditing the divisional settlement of a power transmission and transformation project based on a neural network, and belongs to the technical field of audit management of power transmission and transformation projects. Background Art

[0002] In the traditional settlement audit method of power transmission and transformation projects, manual audit is a common method, which has problems such as low audit efficiency and easy errors. The current on-site engineering quantity collection data cannot be effectively compared with the settlement data in real time, making the audit work cumbersome and time-consuming.

[0003] The existing technology usually collects on-site construction data and settlement data manually, and then conducts settlement review through manual comparison and calculation. This method has the following shortcomings:

[0004] 1. Low efficiency of manual review: Traditional methods require auditors to do a lot of manual comparison and calculation, lack the support of AI technology, consume time and energy, and are prone to omissions and errors.

[0005] 2. Errors are prone to occur: Manual review is easily affected by subjective factors and negligence, resulting in inaccurate settlement results and even errors.

[0006] 3. The comparison between on-site engineering quantity collection data and settlement data is not timely: On-site engineering quantity collection data and settlement data cannot be compared in real time, and abnormal situations cannot be discovered and handled in time, resulting in untimely and inaccurate audit work.

[0007] 4. Audit work is cumbersome and time-consuming: Traditional methods require a lot of manpower and time for data processing and auditing, which is inefficient and does not meet the needs of modern engineering management.

[0008] In order to realize the rapid comparison and automatic review of on-site engineering quantity data and settlement data, an intelligent review method needs to be proposed. Summary of the invention

[0009] In order to solve the above problems, the present invention proposes a method and device for auditing the settlement of power transmission and transformation projects based on neural networks, which can improve the efficiency of auditing the settlement of power transmission and transformation projects and reduce errors caused by human factors.

[0010] The technical solution adopted by the present invention to solve the technical problem is:

[0011] In a first aspect, an embodiment of the present invention provides a method for reviewing the settlement of power transmission and transformation projects based on a neural network, comprising the following steps:

[0012] Obtaining settlement data of the power transmission and transformation project, the settlement data including bill of quantities, construction progress and settlement documents;

[0013] Perform data cleaning on the settlement data and perform text analysis on the cleaned settlement data;

[0014] Use word embedding technology to convert the text in the settlement data into vector form and express it in the form of sequence data to obtain sequence settlement data;

[0015] The sequence settlement data is divided into a training set and a validation set. The preset initialization transmission and transformation project review model is trained using the training set. The model is optimized and iterated with the goal of maximizing the accuracy of the settlement review task to obtain a trained transmission and transformation project review model.

[0016] Use the trained power transmission and transformation project review model to intelligently review new power transmission and transformation project settlement documents and output abnormal data to the review report.

[0017] As a possible implementation of this embodiment, the bill of quantities includes:

[0018] Project name and number, construction unit and contact information;

[0019] Quantity Items: Lists each specific work item in the project;

[0020] Quantity and unit: the specific quantity and measurement unit of each engineering quantity item;

[0021] Unit price and total amount: unit price and total amount of each project quantity;

[0022] The construction progress includes:

[0023] Project phase: Divide the project into different phases or nodes;

[0024] Completion status: the actual completion status of each stage;

[0025] Problems and Challenges: Record problems encountered, delays or other factors that affect the project progress;

[0026] The settlement documents include:

[0027] Contract information: confirm the contract terms, conditions and agreements of the project;

[0028] Payment terms: stipulates the payment methods and conditions during the settlement process;

[0029] Cost details: Specific cost details for each engineering quantity item;

[0030] Final settlement amount: the final amount payable after review and confirmation.

[0031] As a possible implementation of this embodiment, the data cleaning process for the settlement data includes the following steps:

[0032] Remove stop words and punctuation marks: Remove common stop words and punctuation marks in settlement data to retain key information;

[0033] Standardize the format of settlement data and format the settlement data into a standard form;

[0034] Detect and address outliers or erroneous data in settlement data.

[0035] As a possible implementation of this embodiment, the text parsing of the settlement data after data cleaning includes the following steps:

[0036] Use natural language processing techniques to identify and extract important fields from settlement data;

[0037] The settlement data is segmented into paragraphs or preset structures, including the bill of quantities part, the construction progress part and the settlement terms part.

[0038] As a possible implementation of this embodiment, the method of converting the text in the settlement data into a vector form by using word embedding technology and expressing it in a sequence data form to obtain sequence settlement data includes the following steps:

[0039] The settlement data is trained through the CBOW model to obtain word embedding vectors;

[0040] After obtaining the word embedding vector, each word in the settlement data is represented as a corresponding word embedding vector, and then the entire settlement data document is represented as a word embedding matrix or sequence data;

[0041] The settlement data document is represented as a word embedding matrix, where each row represents the word embedding vector of a word, forming a sequence to obtain sequence settlement data.

[0042] As a possible implementation of this embodiment, in the process of training the settlement data through the CBOW model to obtain the word vector, the vector of each word is learned by predicting the surrounding words:

[0043] The CBOW (Continuous Bag-of-Words Model) model is a Word2Vec model that uses the CBOW model to maximize the conditional probability of a given central word and its context words:

[0044]

[0045] Where T is the total number of words in the document, W tAs the central word, W t+j is the context word, c is the window size; P(W t+j |W t ) is the specific conditional probability; V is all the words in the vocabulary, socre(W O , W I ) indicates W O With W I The vector inner product, W O and W I are the word counts of word O and word I respectively, and both O and I belong to V.

[0046] As a possible implementation of this embodiment, the sequence settlement data is divided into a training set and a validation set, and a preset initialization power transmission and transformation project review model is trained using the training set. The model optimization iteration is performed with the accuracy of the maximum settlement review task as the goal to obtain a trained power transmission and transformation project review model, including the following steps:

[0047] The sequence settlement data is divided into a training set and a validation set, wherein the training set is used to train the parameters of the model, and the validation set is used to evaluate the performance of the model and adjust the hyperparameters;

[0048] Initialize the model according to the preset power transmission and transformation project review model, and initialize and train the model based on the Adam optimizer. At the same time, combine the preset cross entropy loss function to judge the performance of the model.

[0049] Calculate the mean square error loss between the predicted value and the actual value. When it is less than the preset convergence threshold, stop the iteration and obtain the initial power transmission and transformation project settlement audit model.

[0050] The training set is input into the initial power transmission and transformation project review model, and the iterative weights are optimized through the preset ADam optimizer. When the loss function converges or the maximum number of iterations is reached, the iteration is stopped to obtain the final optimized weights, thereby obtaining the final power transmission and transformation project review model;

[0051] Among them, the preset initialization transmission and transformation project review model is as follows;

[0052]

[0053] In the formula, i t 、f t 、a t They are input gate, output gate and forget gate respectively; c t represents the long-term memory of the cell state, η and tanh are activation functions, and θ t-1 is the output of the previous stage, w ix 、w fx 、w cx 、w ax 、wih 、w fh 、w ch 、w ah are the corresponding weights to be trained, b i 、b f 、b c 、b a is the bias term;

[0054] The preset ADam optimizer is:

[0055]

[0056]

[0057] Where β1 and β2 are hyperparameters that control the exponential decay rate of the gradient, θ t+1 is the output of layer t+1, and are the two feature outputs of the t+1 layer of the forget gate, m t+1 and V t+1 It is the input of the forget gate;

[0058] The cross entropy loss function is:

[0059]

[0060] In the formula, is the cross entropy loss function, is the actual predicted value in the training set, y is the model predicted value, and n is the number of data in the training set;

[0061] The calculation formula of the mean square error loss is:

[0062]

[0063] As a possible implementation of this embodiment, the method of using the trained power transmission and transformation project review model to intelligently review the new power transmission and transformation project settlement document and outputting abnormal data to the review report includes the following steps:

[0064] Perform data cleaning, text parsing and text conversion on the new power transmission and transformation project settlement documents, and express them in the form of sequence data to obtain new sequence settlement data;

[0065] The new sequence settlement data is input into the trained power transmission and transformation project review model for prediction, and a review report is generated according to the prediction results output by the trained power transmission and transformation project review model.

[0066] As a possible implementation of this embodiment, the review report pays special attention to identifying and describing abnormal data considered by the trained power transmission and transformation project review model; the review report includes:

[0067] Detailed description of abnormal data: For each data point or quantity item identified as abnormal by the model, provide a detailed description and problem;

[0068] Suggested corrective actions: For each abnormal data, provide correction suggestions or improvement opinions to help correct errors and improve settlement accuracy;

[0069] Summary of audit results: Summarize the audit results of the overall settlement documents, including items that have passed and items that require further review or revision.

[0070] In a second aspect, an embodiment of the present invention provides a neural network-based power transmission and transformation project division settlement audit device, comprising:

[0071] A data acquisition module is used to obtain the settlement data of the power transmission and transformation project, wherein the settlement data includes the bill of quantities, construction progress and settlement documents;

[0072] A data processing module is used to perform data cleaning on the settlement data and perform text analysis on the settlement data after data cleaning;

[0073] A data conversion module is used to convert the text in the settlement data into a vector form using word embedding technology, and express it in a sequence data form to obtain sequence settlement data;

[0074] The model training module is used to divide the sequence settlement data into a training set and a validation set, and use the training set to train the preset initialization power transmission and transformation project review model. With the maximum accuracy of the settlement review task as the goal, the model is optimized and iterated to obtain a trained power transmission and transformation project review model;

[0075] The intelligent audit module is used to use the trained power transmission and transformation project review model to perform intelligent audits on new power transmission and transformation project settlement documents and output abnormal data to the review report.

[0076] The beneficial effects of the technical solution of the embodiment of the present invention are as follows:

[0077] The technical solution of the embodiment of the present invention is a method for auditing the settlement of power transmission and transformation engineering divisions based on a neural network, comprising the following steps: obtaining the settlement data of the power transmission and transformation engineering, the settlement data including the bill of quantities, the construction progress and the settlement documents; performing data cleaning processing on the settlement data, and performing text parsing on the settlement data after the data cleaning processing; using word embedding technology to convert the text in the settlement data into a vector form, and expressing it in the form of sequence data to obtain sequence settlement data; dividing the sequence settlement data into a training set and a validation set, using the training set to train the preset initialization power transmission and transformation engineering review model, with the accuracy of the maximum settlement audit task as the goal, performing model optimization iteration to obtain a trained power transmission and transformation engineering review model; using the trained power transmission and transformation engineering review model to intelligently audit the new power transmission and transformation engineering settlement documents, and outputting abnormal data to the review report. The power transmission and transformation engineering review model of the present invention can also continuously improve accuracy and efficiency through continuous learning and optimization; at the same time, through real-time data collection and automatic comparison, the quality and accuracy of the data are also improved, providing more reliable data support for the management and decision-making of power transmission and transformation engineering. The present invention significantly improves the efficiency of the settlement and audit of power transmission and transformation engineering sub-divisions, saves a lot of manpower costs, and saves a lot of resources for the power industry.

[0078] The technical solution of an embodiment of the present invention, a neural network-based audit device for the partial settlement of a power transmission and transformation project, has the same beneficial effects as the technical solution of an embodiment of the present invention, a neural network-based audit method for the partial settlement of a power transmission and transformation project. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a flow chart of a method for reviewing the settlement of power transmission and transformation projects based on a neural network according to an exemplary embodiment;

[0080] Figure 2 It is a schematic diagram of the structure of an audit device for partial settlement of power transmission and transformation projects based on a neural network according to an exemplary embodiment. DETAILED DESCRIPTION

[0081] In order to more clearly illustrate the technical features of the solution of the present invention, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0082] like Figure 1 As shown, an embodiment of the present invention provides a method for reviewing the settlement of power transmission and transformation project divisions based on a neural network, comprising the following steps:

[0083] Obtaining settlement data of the power transmission and transformation project, the settlement data including bill of quantities, construction progress and settlement documents;

[0084] Perform data cleaning on the settlement data and perform text analysis on the settlement data after data cleaning;

[0085] Use word embedding technology to convert the text in the settlement data into vector form and express it in the form of sequence data to obtain sequence settlement data;

[0086] The sequence settlement data is divided into a training set and a validation set. The preset initialization transmission and transformation project review model is trained using the training set. The model is optimized and iterated with the goal of maximizing the accuracy of the settlement review task to obtain a trained transmission and transformation project review model.

[0087] Use the trained power transmission and transformation project review model to intelligently review new power transmission and transformation project settlement documents and output abnormal data to the review report.

[0088] As a possible implementation of this embodiment, the bill of quantities includes:

[0089] Project name and number, construction unit and contact information;

[0090] Quantity Items: Lists each specific work item in the project;

[0091] Quantity and unit: the specific quantity and measurement unit of each engineering quantity item;

[0092] Unit price and total amount: unit price and total amount of each project quantity;

[0093] The construction progress includes:

[0094] Project phase: Divide the project into different phases or nodes;

[0095] Completion status: the actual completion status of each stage;

[0096] Problems and Challenges: Record problems encountered, delays or other factors that affect the project progress;

[0097] The settlement documents include:

[0098] Contract information: confirm the contract terms, conditions and agreements of the project;

[0099] Payment terms: stipulates the payment methods and conditions during the settlement process;

[0100] Cost details: Specific cost details for each engineering quantity item;

[0101] Final settlement amount: the final amount payable confirmed after review.

[0102] As a possible implementation of this embodiment, the data cleaning process for the settlement data includes the following steps:

[0103] Remove stop words and punctuation marks: Remove common stop words and punctuation marks in settlement data to retain key information;

[0104] Standardize the format of settlement data and format the settlement data into a standard form;

[0105] Detect and address outliers or erroneous data in settlement data.

[0106] As a possible implementation of this embodiment, the text parsing of the settlement data after data cleaning includes the following steps:

[0107] Use natural language processing techniques to identify and extract important fields from settlement data;

[0108] The settlement data is segmented into paragraphs or preset structures, including the bill of quantities part, the construction progress part and the settlement terms part.

[0109] As a possible implementation of this embodiment, the method of converting the text in the settlement data into a vector form by using word embedding technology and expressing it in a sequence data form to obtain sequence settlement data includes the following steps:

[0110] The settlement data is trained through the CBOW model to obtain word embedding vectors;

[0111] After obtaining the word embedding vector, each word in the settlement data is represented as a corresponding word embedding vector, and then the entire settlement data document is represented as a word embedding matrix or sequence data;

[0112] The settlement data document is represented as a word embedding matrix, where each row represents the word embedding vector of a word, forming a sequence to obtain sequence settlement data.

[0113] As a possible implementation of this embodiment, in the process of training the settlement data through the CBOW model to obtain the word vector, the vector of each word is learned by predicting the surrounding words:

[0114] Use the CBOW model to maximize the conditional probability of a given central word and its context words:

[0115]

[0116] Where T is the total number of words in the document, W t As the central word, W t+j is the context word, c is the window size; P(W t+j |W t ) is the specific conditional probability; V is all the words in the vocabulary, socre(W O , WI ) indicates W O With W I The vector inner product, W O and W I are the word counts of word O and word I respectively, and both O and I belong to V.

[0117] As a possible implementation of this embodiment, the sequence settlement data is divided into a training set and a validation set, and a preset initialization power transmission and transformation project review model is trained using the training set. The model optimization iteration is performed with the accuracy of the maximum settlement review task as the goal to obtain a trained power transmission and transformation project review model, including the following steps:

[0118] The sequence settlement data is divided into a training set and a validation set, wherein the training set is used to train the parameters of the model, and the validation set is used to evaluate the performance of the model and adjust the hyperparameters;

[0119] Initialize the model according to the preset power transmission and transformation project review model, and initialize and train the model based on the Adam optimizer. At the same time, combine the preset cross entropy loss function to judge the performance of the model.

[0120] Calculate the mean square error loss between the predicted value and the actual value. When it is less than the preset convergence threshold, stop the iteration and obtain the initial power transmission and transformation project settlement audit model.

[0121] The training set is input into the initial power transmission and transformation project review model, and the iterative weights are optimized through the preset ADam optimizer. When the loss function converges or the maximum number of iterations is reached, the iteration is stopped to obtain the final optimized weights, thereby obtaining the final power transmission and transformation project review model;

[0122] Among them, the preset initialization transmission and transformation project review model is as follows;

[0123]

[0124] In the formula, i t 、f t 、a t They are input gate, output gate and forget gate respectively; c t represents the long-term memory of the cell state, η and tanh are activation functions, and θ t-1 is the output of the previous stage, w ix 、w fx 、w cx 、w ax 、w ih 、w fh 、w ch 、w ah are the corresponding weights to be trained, b i 、b f 、bc 、b a is the bias term;

[0125] The preset ADam optimizer is:

[0126]

[0127] Where β1 and β2 are hyperparameters that control the exponential decay rate of the gradient, θ t+1 is the output of layer t+1, and are the two feature outputs of the t+1 layer of the forget gate, m t+1 and V t+1 is the input of the forget gate;

[0128] The cross entropy loss function is:

[0129]

[0130] In the formula, is the cross entropy loss function, is the actual predicted value in the training set, y is the model predicted value, and n is the number of data in the training set;

[0131] The calculation formula of the mean square error loss is:

[0132]

[0133] As a possible implementation of this embodiment, the method of using the trained power transmission and transformation project review model to intelligently review the new power transmission and transformation project settlement document and outputting abnormal data to the review report includes the following steps:

[0134] Perform data cleaning, text parsing and text conversion on the new power transmission and transformation project settlement documents, and express them in the form of sequence data to obtain new sequence settlement data;

[0135] The new sequence settlement data is input into the trained power transmission and transformation project review model for prediction, and a review report is generated according to the prediction results output by the trained power transmission and transformation project review model.

[0136] As a possible implementation of this embodiment, the review report pays special attention to identifying and describing abnormal data considered by the trained power transmission and transformation project review model; the review report includes:

[0137] Detailed description of abnormal data: For each data point or quantity item identified as abnormal by the model, provide a detailed description and problem;

[0138] Suggested corrective actions: For each abnormal data, provide correction suggestions or improvement opinions to help correct errors and improve settlement accuracy;

[0139] Summary of audit results: Summarize the audit results of the overall settlement documents, including items that have passed and items that require further review or revision.

[0140] like Figure 2 As shown, an embodiment of the present invention provides a neural network-based power transmission and transformation project division settlement audit device, comprising:

[0141] A data acquisition module is used to obtain the settlement data of the power transmission and transformation project, wherein the settlement data includes the bill of quantities, construction progress and settlement documents;

[0142] A data processing module is used to perform data cleaning on the settlement data and perform text analysis on the settlement data after data cleaning;

[0143] A data conversion module is used to convert the text in the settlement data into a vector form using word embedding technology, and express it in a sequence data form to obtain sequence settlement data;

[0144] The model training module is used to divide the sequence settlement data into a training set and a validation set, and use the training set to train the preset initialization power transmission and transformation project review model. With the maximum accuracy of the settlement review task as the goal, the model is optimized and iterated to obtain a trained power transmission and transformation project review model;

[0145] The intelligent audit module is used to use the trained power transmission and transformation project review model to perform intelligent audits on new power transmission and transformation project settlement documents and output abnormal data to the review report.

[0146] The specific process of using the audit device of the present invention to audit the settlement of power transmission and transformation project sections is as follows.

[0147] S1, obtain the settlement data of the power transmission and transformation project.

[0148] The data includes bill of quantities, construction progress, and settlement documents.

[0149] The bill of quantities is an important part of the settlement of power transmission and transformation projects, and usually includes the following information:

[0150] Project name and number, construction unit and contact information; Project quantity items: lists each specific work item in the project, such as civil engineering, electrical equipment installation, and line installation; Quantity and unit: the specific quantity and measurement unit of each project quantity item, such as meters, cubic meters, and units; Unit price and total amount: the unit price and total amount of each project quantity, which are used to calculate the total cost of the project.

[0151] The construction progress refers to: The construction progress report records the completion status and progress of the power transmission and transformation project at different stages, including:

[0152] Project stages: The project is divided into different stages or nodes, such as the design stage, construction stage, commissioning stage, etc.;

[0153] Completion status: The actual completion status of each stage, usually expressed as a percentage or a specific description of the completed work;

[0154] Problems and challenges: Record the problems, delays or other factors affecting the project progress encountered.

[0155] The settlement document contains the detailed information of the final settlement of the power transmission and transformation project, mainly including:

[0156] Contract information: Confirm the contract terms, conditions and agreements of the project;

[0157] Payment terms: Specify the payment methods and conditions during the settlement process;

[0158] Cost details: The specific cost details of each project quantity item, including the original contract amount, change amount, deduction amount, etc.;

[0159] Final settlement amount: The final payable amount confirmed after review, including the comprehensive total of all contract terms and changes.

[0160] S2, for the settlement data of the power transmission and transformation project, based on text parsing and data cleaning.

[0161] In the text parsing stage, the main goal is to extract and understand the key information from the settlement document. The specific steps include:

[0162] Identify key fields: Use natural language processing techniques (such as regular expressions, named entity recognition, etc.) to identify and extract important fields in the settlement document, such as project quantity items, quantities, unit prices, etc.;

[0163] Segmentation and classification: Segment the document by paragraphs or a specific structure for subsequent processing. For example, divide the document into parts such as the bill of quantities section, construction progress section, and settlement terms section.

[0164] The purpose of data cleaning is to process the raw data obtained from the text parsing stage to ensure the quality and consistency of the data, such as removing error or redundant information. The specific steps include:

[0165] Remove stop words and punctuation marks: Remove common stop words (such as "of", "and", "with", etc.) and punctuation marks in the settlement document to retain the key information;

[0166] Data format standardization: Format the data into a standard form that the system can process. For example, convert quantity and unit price into digital format;

[0167] Outlier processing: Detect and process outliers or erroneous data in settlement data. For example, check whether there are quantities or unit prices that are outside the reasonable range.

[0168] S3, uses word embedding technology to convert the text in the settlement document into vector form; and expresses it in the form of sequence data to obtain the preprocessed settlement data.

[0169] Word embedding is a technique that maps words or phrases to real-domain vectors, which can capture the semantic relationship between words. Common word embedding methods include Word2Vec, GloVe, and FastText.

[0170] Word2Vec is a neural network-based word embedding model that learns the vector representation of each word by predicting surrounding words;

[0171] Given a central word and its context words, use the CBOW model to maximize the conditional probability:

[0172] The specific calculation formula for maximizing conditional probability is as follows:

[0173]

[0174] Where T is the total number of words in the document, W t As the central word, W t+j is the context word, c is the window size; P(W t+j |W t ) is the specific conditional probability;

[0175] The specific calculation formula for specific conditional probability is as follows:

[0176]

[0177] Where V is all the words in the vocabulary, socre(W O , W I ) indicates W O With W I The vector inner product, W O and W I are the word counts of word O and word I respectively, both O and I belong to V;

[0178] After obtaining the word embedding vector, each word in the document is represented as the corresponding word embedding vector, and then the entire document is represented as a word embedding matrix or sequence data.

[0179] Sequence data form: The document is represented as a word embedding matrix, where each row represents the word embedding vector of a word, forming a sequence.

[0180] S4, divide the preprocessed settlement data into a training set and a validation set, use the training set to train the preset initialization transmission and transformation project review model, and optimize the iterative model with the goal of maximizing the accuracy of the settlement review task.

[0181] First, the preprocessed settlement data is divided into a training set and a validation set. The training set is used to train the parameters of the model, and the validation set is used to evaluate the performance of the model and adjust the hyperparameters.

[0182] The training set is used to train the model, that is, to adjust the model parameters through the optimization algorithm to minimize the loss function.

[0183] The validation set is used to select the best model hyperparameters, such as learning rate, regularization parameters, etc., and to evaluate the generalization ability of the model on unseen data.

[0184] Initialize the model according to the preset power transmission and transformation project review model, and initialize and train the model based on the Adam optimizer. At the same time, combine the preset cross entropy loss function to judge the performance of the model.

[0185] The mean square error loss between the predicted value and the actual value is calculated. When it is less than the preset convergence threshold, the iteration is stopped to obtain the settlement audit model of the power transmission and transformation project.

[0186] The specific formula of the cross entropy loss function is as follows:

[0187]

[0188] In the formula, is the cross entropy loss function, is the actual predicted value in the training set, y is the model predicted value, and n is the number of data in the training set;

[0189] The specific calculation formula of the mean square error loss is as follows:

[0190]

[0191] Among them, the preset initialization transmission and transformation project review model is as follows;

[0192]

[0193] In the formula, i t 、f t 、a t They are input gate, output gate and forget gate respectively; c t represents the long-term memory of the cell state, η and tanh are activation functions, and θt-1 is the output of the previous stage, w ix 、w fx 、w cx 、w ax 、w ih 、w fh 、w ch 、w ah are the corresponding weights to be trained, b i 、b f 、b c 、b a is the bias term.

[0194] The preset ADam optimizer is as follows:

[0195]

[0196] Where β1 and β2 are hyperparameters that control the exponential decay rate of the gradient, θ t+1 is the output of layer t+1, and are the two feature outputs of the t+1 layer of the forget gate, m t+1 and V t+1 It is the input of the forget gate;

[0197] The optimization iteration model is specifically:

[0198] The training set is input into the initial power transmission and transformation project review model, and the iterative weights are optimized through the preset ADam optimizer. When the loss function converges or the maximum number of iterations is reached, the iteration is stopped to obtain the final optimized weights, thereby obtaining the final power transmission and transformation project review model.

[0199] S5, uses the trained power transmission and transformation project review model to intelligently review the new power transmission and transformation project settlement documents, and outputs abnormal data to the review report.

[0200] First, the new power transmission and transformation project settlement documents are preprocessed in a similar way to the training data, including text parsing, data cleaning, and field identification. These steps are designed to ensure that the model can correctly understand and process the input data.

[0201] The model makes predictions for each input document and generates review reports based on the model's predictions, with a particular focus on identifying and describing data that the model considers to be anomalous. These reports can include the following:

[0202] Detailed description of abnormal data: For each data point or quantity item identified as abnormal by the model, provide a detailed description and problem;

[0203] Suggested corrective actions: For each abnormal data, provide correction suggestions or improvement opinions to help correct errors and improve settlement accuracy;

[0204] Summary of audit results: Summarize the audit results of the overall settlement documents, including items that have passed and items that require further review or revision.

[0205] Compared with the prior art, the present invention has the following characteristics:

[0206] 1. Automatic matching and comparison: The present invention uses an intelligent power transmission and transformation project review model to automatically match and compare the collected actual engineering quantity data with the settlement data. This is not only fast, but also reduces the need for manual intervention and avoids errors caused by human factors. The intelligent algorithm can quickly identify the difference between the settlement data and the actual data, improving the accuracy of the review.

[0207] 2. Real-time data collection: Through on-site engineering quantity collection equipment, the present invention realizes real-time data collection of power transmission and transformation projects, which ensures the timeliness and accuracy of settlement data and actual data, avoids the periodic data collection and calculation in traditional methods, and thus improves audit efficiency.

[0208] 3. Intelligent power transmission and transformation project review model: The present invention establishes an intelligent power transmission and transformation project review model, and uses machine learning and other technologies to continuously optimize and adjust the review rules. It can quickly and accurately detect abnormal items and differences, avoiding the tedious manual review process in traditional methods and reducing the review error rate.

[0209] The intelligent power transmission and transformation project review model of the present invention is not only a one-time application, but also can continuously improve accuracy and efficiency through continuous learning and optimization. At the same time, through real-time data collection and automatic comparison, the quality and accuracy of data are also improved, providing more reliable data support for the management and decision-making of power transmission and transformation projects.

[0210] The present invention can significantly improve the efficiency of the settlement and audit of power transmission and transformation projects, save a lot of manpower costs, and save a lot of resources for the power industry. For example, in the settlement of a power transmission and transformation project, traditional manual auditing may take several days or even weeks, while the present invention can complete the audit within a few hours, greatly shortening the audit cycle.

[0211] The present invention brings huge economic benefits by saving labor costs and audit time. Taking a power transmission and transformation project in a certain area as an example, the traditional manual audit cost may reach hundreds of thousands of yuan, while the intelligent audit method of the present invention only requires a small part of the cost. In addition, since audit errors are reduced, disputes and compensation situations are avoided, thereby saving more costs.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for auditing the settlement of power transmission and transformation projects based on neural networks, characterized in that: The steps include: Obtaining settlement data of the power transmission and transformation project, the settlement data including bill of quantities, construction progress and settlement documents; Perform data cleaning on the settlement data and perform text analysis on the cleaned settlement data; Use word embedding technology to convert the text in the settlement data into vector form and express it in the form of sequence data to obtain sequence settlement data; The sequence settlement data is divided into a training set and a validation set. The preset initialization transmission and transformation project review model is trained using the training set. The model is optimized and iterated with the goal of maximizing the accuracy of the settlement review task to obtain a trained transmission and transformation project review model. Use the trained power transmission and transformation project review model to intelligently review new power transmission and transformation project settlement documents and output abnormal data to the review report.

2. The audit method for the settlement of power transmission and transformation projects based on neural network according to claim 1 is characterized in that: The bill of quantities includes: Project name and number, construction unit and contact information; Quantity Items: Lists each specific work item in the project; Quantity and unit: the specific quantity and measurement unit of each engineering quantity item; Unit price and total amount: unit price and total amount of each project quantity; The construction progress includes: Project phase: Divide the project into different phases or nodes; Completion status: the actual completion status of each stage; Problems and Challenges: Record problems encountered, delays or other factors that affect the project progress; The settlement documents include: Contract information: confirm the contract terms, conditions and agreements of the project; Payment terms: stipulates the payment methods and conditions during the settlement process; Cost details: Specific cost details for each engineering quantity item; Final settlement amount: the final amount payable after review and confirmation.

3. The audit method for the settlement of power transmission and transformation projects based on neural network according to claim 1 is characterized in that: The data cleaning process for the settlement data comprises the following steps: Remove stop words and punctuation marks: Remove common stop words and punctuation marks in settlement data to retain key information; Standardize the format of settlement data and format the settlement data into a standard form; Detect and address outliers or erroneous data in settlement data.

4. The audit method for the settlement of power transmission and transformation projects based on neural network according to claim 1 is characterized in that: The text parsing of the settlement data after data cleaning includes the following steps: Use natural language processing techniques to identify and extract important fields from settlement data; The settlement data is segmented into paragraphs or preset structures, including the bill of quantities part, the construction progress part and the settlement terms part.

5. The audit method for the settlement of power transmission and transformation projects based on neural network according to claim 1 is characterized in that: The method of converting the text in the settlement data into a vector form by using word embedding technology and expressing it in a sequence data form to obtain sequence settlement data includes the following steps: The settlement data is trained through the CBOW model to obtain word embedding vectors; After obtaining the word embedding vector, each word in the settlement data is represented as a corresponding word embedding vector, and then the entire settlement data document is represented as a word embedding matrix or sequence data; The settlement data document is represented as a word embedding matrix, where each row represents the word embedding vector of a word, forming a sequence to obtain sequence settlement data.

6. The audit method for the settlement of power transmission and transformation projects based on neural network according to claim 5 is characterized in that: In the process of training the settlement data through the CBOW model to obtain word vectors, the vector of each word is learned by predicting the surrounding words: Use the CBOW model to maximize the conditional probability of a given central word and its context words: Where T is the total number of words in the document, W t As the central word, W t+j is the context word, c is the window size; P(W t+j |W t ) is the specific conditional probability; V is all the words in the vocabulary, socre(W O , W I ) indicates W O With W I The vector inner product, W O and W I are the word counts of word O and word I respectively, and both O and I belong to V.

7. The audit method for the settlement of power transmission and transformation projects based on neural network according to claim 1 is characterized in that: The method divides the sequence settlement data into a training set and a validation set, uses the training set to train the preset initialization power transmission and transformation project review model, and performs model optimization iteration with the goal of maximizing the accuracy of the settlement review task to obtain a trained power transmission and transformation project review model, including the following steps: The sequence settlement data is divided into a training set and a validation set, wherein the training set is used to train the parameters of the model, and the validation set is used to evaluate the performance of the model and adjust the hyperparameters; Initialize the model according to the preset power transmission and transformation project review model, and initialize and train the model based on the Adam optimizer. At the same time, combine the preset cross entropy loss function to judge the performance of the model. Calculate the mean square error loss between the predicted value and the actual value. When it is less than the preset convergence threshold, stop the iteration and obtain the initial power transmission and transformation project settlement audit model. The training set is input into the initial power transmission and transformation project review model, and the iterative weights are optimized through the preset ADam optimizer. When the loss function converges or the maximum number of iterations is reached, the iteration is stopped to obtain the final optimized weights, thereby obtaining the final power transmission and transformation project review model; Among them, the preset initialization transmission and transformation project review model is as follows; In the formula, i t 、f t 、a t They are input gate, output gate and forget gate respectively; c t represents the long-term memory of the cell state, η and tanh are activation functions, θ t-1 is the output of the previous stage, w ix 、w fx 、w cx 、w ax 、w ih 、w fh 、w ch 、w ah are the corresponding weights to be trained, b i , b f , b c , b a is the bias term; The preset ADam optimizer is: Where β1 and β2 are hyperparameters that control the exponential decay rate of the gradient, θ t+1 is the output of layer t+1, and are the two feature outputs of the t+1 layer of the forget gate, m t+1 and V t+1 It is the input of the forget gate; The cross entropy loss function is: In the formula, is the cross entropy loss function, is the actual predicted value in the training set, y is the model predicted value, and n is the number of data in the training set; The calculation formula of the mean square error loss is:

8. The neural network-based audit method for power transmission and transformation project divisional settlement according to any one of claims 1 to 7, characterized in that: The method of using the trained power transmission and transformation project review model to intelligently review the new power transmission and transformation project settlement documents and outputting abnormal data to the review report includes the following steps: Perform data cleaning, text parsing and text conversion on the new power transmission and transformation project settlement documents, and express them in the form of sequence data to obtain new sequence settlement data; The new sequence settlement data is input into the trained power transmission and transformation project review model for prediction, and a review report is generated according to the prediction results output by the trained power transmission and transformation project review model.

9. The neural network-based power transmission and transformation project divisional settlement audit method according to claim 8 is characterized in that: The review report includes: Detailed description of abnormal data: For each data point or quantity item identified as abnormal by the model, provide a detailed description and problem; Suggested corrective actions: For each abnormal data, provide correction suggestions or improvement opinions to help correct errors and improve settlement accuracy; Summary of audit results: Summarize the audit results of the overall settlement documents, including items that have passed and items that require further review or revision.

10. A neural network-based audit device for power transmission and transformation project divisional settlement, characterized in that: include: A data acquisition module is used to obtain the settlement data of the power transmission and transformation project, wherein the settlement data includes the bill of quantities, construction progress and settlement documents; A data processing module is used to perform data cleaning on the settlement data and perform text analysis on the settlement data after data cleaning; A data conversion module is used to convert the text in the settlement data into a vector form using word embedding technology, and express it in a sequence data form to obtain sequence settlement data; The model training module is used to divide the sequence settlement data into a training set and a validation set, and use the training set to train the preset initialization power transmission and transformation project review model. With the maximum accuracy of the settlement review task as the goal, the model is optimized and iterated to obtain a trained power transmission and transformation project review model; The intelligent audit module is used to use the trained power transmission and transformation project review model to perform intelligent audits on new power transmission and transformation project settlement documents and output abnormal data to the review report.

Images