Bidding and tendering supervision system for engineering project

Abstract

The invention relates to the field of bidding and tendering data supervision, and discloses a bidding and tendering supervision system for an engineering project, comprising a data acquisition module for acquiring bidding and tendering data, a bidding and tendering data supervision module comprising: a time sequence construction module for acquiring time sequence data of the bidding and tendering data, obtaining similarity ontology features based on the time sequence data; the time sequence management module is used for creating a time sequence feature matrix based on the similarity ontology features, extracting periodic data point features from the time sequence feature matrix, and separating the periodic data point features by using a time sequence decomposition algorithm to obtain periodic data points; and the anomaly analysis module is used for constructing an anomaly analysis model, inputting the periodic data points into the anomaly analysis model, obtaining an anomaly score of each periodic data point, automatically detecting the abnormal data points through a machine learning model, and distributing the anomaly score for each periodic data point, so that the workload of manual examination is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of bidding data supervision, and in particular to a bidding supervision system for engineering projects. Background Art

[0002] As competition in the project bidding market becomes increasingly fierce, a more fair and transparent supervision mechanism is needed. At present, the application of technologies such as big data, cloud computing and artificial intelligence has provided technical support for the construction and operation of the bidding supervision system. Problems such as bid rigging, false bidding and so on often occur in the construction of engineering projects. The existence of these problems has seriously damaged the fair competition environment in the market. While ensuring the quality of the project, reducing costs and improving benefits have become one of the important goals of engineering project management. At present, there are abnormal time series data in the bidding data obtained by the bidding supervision system. Abnormal time series data is often an attempt to cover up the signs of collusion. For example, the system records show that the bidding documents have been submitted hours or even days before the bidding announcement is issued. This will be regarded as abnormal time series data, indicating that there may be illegal activities. If they are not screened out, it will affect the fairness of the final bidding process. Therefore, a bidding supervision system for engineering projects is proposed. Summary of the invention

[0003] The purpose of the present invention is to solve the shortcoming in the prior art that abnormal time series data affects the fairness of the final bidding process, and to propose a bidding supervision system for engineering projects.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: A bidding supervision system for engineering projects includes a data acquisition module for acquiring bidding data, and the bidding data supervision module includes: Time series construction module: used to obtain time series data of bidding data and obtain similarity ontology features based on time series data; Time series management module: Create a time series feature matrix based on similarity ontology features, extract periodic data point features from the time series feature matrix, and use the time series decomposition algorithm to separate the periodic data point features to obtain periodic data points; Anomaly analysis module: used to build an anomaly analysis model, input periodic data points into the anomaly analysis model, and obtain the anomaly score of each periodic data point; Dynamic adjustment module: used to capture the dynamic characteristics of periodic data points, calculate the abnormal behavior threshold based on the dynamic characteristics of periodic data points, and clear the bidding data corresponding to the periodic data points whose abnormal scores are greater than the abnormal behavior threshold.

[0005] The process of obtaining the time series data is as follows: Collect bidding data, remove missing values ​​or outliers from the collected bidding data, unify the format, and extract the time series, which can be expressed as in, is the length of the time series, Indicates that the length of the time series is The time series data corresponding to .

[0006] The process of obtaining the similarity ontology feature is as follows: Define the time interval for selecting data points in the time series as and embedding dimension ; Time interval The definition process is: Based on time series data , define the time series and the time series delayed version The mutual information between them is: ;in, For time series data in time series and time series data Delayed version of The joint probability distribution function of For time series data The marginal probability distribution function of Delayed version of time series data The marginal probability distribution function of ; Based on mutual information Draw Different The mutual information curve of , find the first local minimum in the mutual information curve, and the first local minimum in the mutual information curve corresponding to As the optimal time interval; Embedding Dimension The definition process is: Defining the embedding dimension The initial dimension is 1, based on time series data Constructing embedding vectors , ,in, and For the embedding dimension When, time series data Two different embedding vectors constructed, and For two different indexes, is the time interval, and The index is kept between 1 and Between, calculate each pair of embedding vectors and The Euclidean distance between ,in, is the embedding vector index; Increase the embedding dimension arrive , construct a new embedding vector , ,in, and All are embedded in the dimension When, time series data The two different embedding vectors constructed are and The Euclidean distance between , ,definition and The Euclidean distance between and The ratio of the Euclidean distance between ,in, express The norm of express The norm of , defines the embedding threshold as 0, and selects The smallest value closest to the embedding threshold As the final embedding dimension; Based on time interval and embedding dimension , the time series Mapping to the embedding dimension In the dimensional phase space, create two different Dimensional vector and , , Among them, the time point Limited to arrive In the definition and The direct similarity is , ,in is the standard deviation parameter, for and The square of the Euclidean distance between them, based on direct similarity, extracting time series data Similarity ontology features, obtain the average similarity measure ,in, Represents all vectors of different dimensions The average similarity ontology features are used to construct the similarity matrix, and the principal component analysis of the similarity matrix is ​​performed to obtain the similarity ontology features.

[0007] The process of obtaining the periodic data points is as follows: Based on the average similarity metric , construct the similarity time series matrix , expressed as ,in The embedding dimension is ; Based on the similarity time series matrix Perform spectrum analysis and define similarity time series matrix The spectrum matrix is , convert the similarity time series matrix to the frequency domain and obtain the spectrum matrix ,in, is the frequency index, is the base of the exponential function; Define the spectrum matrix The seasonal component of , based on the spectrum matrix application Decomposition, collect the seasonal components of the time series after each eigenvector decomposition ,in Represents a new sequence, based on each new sequence , for the new sequence Apply spectrum analysis to get the period , use cycle , locate periodic data points in the time series.

[0008] The training process of the anomaly analysis model is: Convert the periodic data points into a transaction data item format to obtain the transaction items of the periodic data points; Generate candidate item sets for transaction items. The generation process of candidate item sets is as follows: Starting from a single transaction item, all item set combinations are generated based on the transaction data items. For each candidate item set, its support in the transaction database is calculated and the minimum support threshold is set. , each item set is represented as , calculate the support of each item set, discard the item sets whose support is lower than the minimum support threshold, and the remaining items are frequent item sets. The support formula is: ,in, Is the included item set The number of transactions, is the total number of transactions for the transaction data item, Representing Item Sets The support of , we get the frequent itemsets; Define the minimum confidence threshold , the two mutually exclusive items in the frequent itemset are and , based on frequent item sets, generate association rules, the form of association rules is , calculate the confidence of the association rule, confidence ,in, for and The proportion of transactions based on the minimum confidence threshold , for confidence levels below the low confidence threshold Itemset Screen out and get the strong association rule item set; Based on the strong association rule item set and periodic data points, define the first The characteristics of a periodic data point are expressed as , the first The item set feature is expressed as , periodic data point characteristics and item set features Compose a dataset , divide the data set into training set and test set, train the anomaly analysis model, define the minimization loss function, and the formula for minimizing the loss function is: ,in, is the weight vector, is the bias term, is a non-negative slack variable, is the penalty parameter, is the sample size; Define the kernel function, the formula of the kernel function is: ,in, is the parameter of the kernel function, is the output value of the kernel function, and For the dataset For any two vectors in express and The Euclidean distance between Define constraints and use them to optimize the abnormal analysis model. The optimization method is as follows: , ,in, is the transposed representation of the weight vector, is the feature vector The category labels include 1 or -1, = 1, the eigenvector Belongs to the positive category, = −1, the eigenvector Belongs to the negative class and obtains an anomaly analysis model.

[0009] The process of obtaining the abnormal score of the periodic data point is as follows: Apply the anomaly analysis model to the new periodic data points, calculate the anomaly score of each data point, and define the anomaly score. The formula for the anomaly score is: ,in, Indicates the input The output anomaly score of each periodic data point feature.

[0010] The dynamic feature capture method of the periodic data is: Based on periodic data points, the window size of the sliding window is defined as , the window step size is , calculate the statistics within each sliding window, the dynamic characteristics of periodic data, and the statistics within the window include the window mean and window standard deviation , window mean ,in, is not equal to within the window The periodic data point characteristics, window standard deviation .

[0011] The calculation process of the abnormal behavior threshold is: Set abnormal behavior thresholds based on the dynamic characteristics of periodic data points , abnormal behavior threshold ,in, For the window The average value of the periodic data point features, For the window The standard deviation of the periodic data point characteristics, is a hyperparameter; when hour, The corresponding periodic data point definition is normal, when hour, The corresponding periodic data points are defined as abnormal, and the bidding data corresponding to the periodic data points defined as abnormal are cleared.

[0012] The present invention has the following beneficial effects: 1. In the present invention, firstly, the time series data of each bidding data is obtained to ensure that the data is arranged in chronological order, which is convenient for subsequent analysis. The relationship between the data is captured through the similarity ontology feature, and a time series feature matrix that is easy to analyze is constructed. The periodic patterns in the data are identified through time series decomposition to help understand the trend and seasonal changes of the data. The abnormal data points are automatically detected through the machine learning model, and an abnormal score is assigned to each periodic data point, which reduces the workload of manual review. The abnormal threshold is set according to the dynamic characteristics of the periodic data point. By clearing the bidding data corresponding to the abnormal data points judged by the abnormal threshold, the project cost and benefit are more accurately evaluated to ensure the fairness of the final bidding process. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention provides a system block diagram of a bidding supervision system for engineering projects. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] like Figure 1 As shown, the present invention proposes a bidding supervision system for engineering projects, including a data acquisition module for acquiring bidding data, and the bidding data supervision module includes: Time series construction module: obtain the time series data of bidding data, and obtain similarity ontology features based on the time series data; Time series management module: Create a time series feature matrix based on similarity ontology features, extract periodic data point features from the time series feature matrix, and use the time series decomposition algorithm to separate the periodic data point features to obtain periodic data points; Anomaly analysis module: Use machine learning methods to train anomaly analysis models, input periodic data points into the anomaly analysis model, and obtain anomaly scores for each periodic data point; Dynamic adjustment module: Set the abnormal behavior threshold, mark the periodic data points with abnormal scores greater than the threshold as abnormal periodic data points, and clear the bidding data corresponding to the periodic data points.

[0016] The process of obtaining the time series data is as follows: Collect bidding data, remove missing values ​​or outliers from the collected bidding data, unify the format, and extract time series data, which can be expressed as ,in, is the length of the time series, Indicates that the length of the time series is The time series data corresponding to .

[0017] The similarity ontology feature acquisition process is: Define the time interval for selecting data points in the time series as and embedding dimension ; Time interval The acquisition process is: Based on time series data , construct the delayed energy moment , where the delayed energy moment Each column is the original time series relative to a different time interval Version, defining the time series and delayed version The mutual information between , the mutual information acquisition process is: ;in, For time series data in time series and time series data Delayed version of The joint probability distribution function of For time series data The marginal probability distribution function of Delayed version of time series data The marginal probability distribution function of ; It is the same time series that is delayed The corresponding observations after time units, plotting the mutual information About Difference The graph of mutual information Find the first local minimum in the curve, because as As increases, the mutual information initially decreases until it reaches a point where the information sharing between the time series and its delayed version is minimal, but still maintains a certain correlation. The first local minimum in the mutual information curve corresponds to As the optimal time interval, this Minimize the information redundancy between the time series and its delayed version while maintaining sufficient correlation, which is suitable for phase space reconstruction; Embedding Dimension The acquisition process is: Defining the embedding dimension The initial dimension is 1, based on time series data Constructing embedding vectors , ,in, and All are time series data Two different embedding vectors constructed, and is the index, is the time interval, maintained at arrive Between, calculate each pair of embedding vectors and The Euclidean distance between , ,in, is the embedding vector index, from arrive ; Increase the embedding dimension arrive , construct a new embedding vector , ,get and The Euclidean distance between ,definition and The Euclidean distance between and The ratio of the Euclidean distance between ,in, express The norm of express The norm of increase, observe changes, when When it is close to zero, it indicates that the embedding dimension is increased. No longer leads to significant changes in the distance between the nearest neighbors, which usually means that enough information about the time series has been included. The minimum value close to zero The value is taken as the final embedding dimension, which shows that under this dimension, the phase space reconstruction can fully capture the dynamic characteristics of the system; Based on time interval and embedding dimension , the time series Mapping to the embedding dimension In the dimensional phase space, create two different Dimensional vector and , , , where time point Limited to arrive In the definition and The direct similarity is , ,in is the standard deviation parameter, for and The square of the Euclidean distance between the two, based on similarity, extract time series data Similarity ontology features, defining time series data The average similarity measure of all vectors in is ,in, Represents all vectors The average similarity ontology features are obtained, the similarity matrix is ​​constructed using the similarity measure, and the similarity ontology features are obtained by performing principal component analysis on the similarity matrix.

[0018] The periodic data point acquisition process is as follows: Based on the similarity ontology feature , construct the similarity time series matrix , expressed as ,in The embedding dimension is ; Based on the similarity time series matrix Perform spectrum analysis and define similarity time series matrix The spectrum matrix is , convert the similarity time series to the frequency domain, expressed as: ,in, is the frequency index, and the spectrum matrix is ​​obtained , is the base of the exponential function; Define the spectrum matrix The seasonal component of , based on the spectrum matrix application Decomposition, collect the seasonal components of the time series after each eigenvector decomposition , based on each new sequence , where each column corresponds to the original feature matrix The seasonal component at the corresponding time point in Apply spectral analysis (such as Fourier transform) to each column of (i.e., the seasonal component at each time point) to identify the most significant periodic frequency. From the spectral analysis, the main periodic frequencies are determined. These frequencies represent recurring patterns in the time series. Once the periodic frequencies are determined, the period can be calculated. , use cycle , locate the periodic data points in the time series, that is, The data points at multiple positions of the original time series Extract the cycle The data points at multiple positions constitute a set of periodic data points.

[0019] The training process of the anomaly analysis model is as follows: Convert periodic data points into transactional data item format; The transaction data item format is a commonly used data representation in data mining and machine learning, especially in association rule learning. This format is usually used to process transaction data, but it is also applicable to any type of data as long as the data can be viewed as a series of discrete events or "transactions". In a transaction database, each record represents an independent transaction, and a transaction is a collection of items. In a bidding supervision system, each "transaction" can be regarded as an independent link or event in a bidding process. Each transaction item can be a feature vector of a certain bidding. A candidate item set is generated for the transaction item. The generation process of the candidate item set is as follows: Starting from a single transaction item, all item set combinations are generated based on the transaction data items. For each candidate item set, its support in the transaction database is calculated and the minimum support threshold is set. , each item set is represented as , that is, the proportion of transactions containing this item set. If the support is lower than the minimum support threshold, the item set will be discarded, and the remaining items are frequent itemsets. The formula is: ,in, Is the included item set The number of transactions, is the total number of transactions for the transaction data item, Representing Item Sets The support of , we get the frequent itemsets; Define the minimum confidence threshold , the two mutually exclusive items in the frequent itemset are and , based on frequent item sets, generate association rules, the form of association rules is , calculate the confidence of the rule and determine whether it meets the minimum confidence threshold. The confidence is defined as: ,in, for and The proportion of transactions based on support and confidence Screening is performed based on the lowest confidence threshold , for confidence levels below the low confidence threshold Itemset Screen out and get the strong association rule item set; Based on the strong association rule item set and periodic data points, define the first The characteristics of a periodic data point are expressed as , the first The item set feature is expressed as , periodic data point characteristics and item set features Compose a dataset , divide the data set into training set and test set, based on support vector machine Train the anomaly analysis model. The model training process includes minimizing the loss function, defining constraints and kernel functions; Support vector machine Minimize the loss function, and the minimization loss function is expressed as: ,in, is the weight vector, is the bias term, is a non-negative slack variable, is the penalty parameter, is the sample size; Define the constraints as: ,in, It is Datasets The eigenvectors in is the label corresponding to the feature vector; The kernel function is: ,in, is the parameter of the kernel function, is the kernel function output value, and For the dataset For any two vectors in express and The Euclidean distance between Overall, it is expressed as The exponential function with base is applied to , the result remains a value between 0 and 1, indicating the similarity between the two vectors; After the model training process is completed, the model is evaluated. Further, the model evaluation process is to evaluate the performance of the model on the test set, calculate the accuracy, recall rate, F1 score, and then use grid search or random search to optimize Parameters and , and obtain the abnormal analysis model.

[0020] The process of obtaining the abnormal periodic data points is as follows: The trained The model is applied to new periodic data points, and the decision function value of each data point is calculated. The decision function value represents the distance of the data point from the classification boundary, which serves as the main basis for the anomaly score, or is equal to the anomaly score. The anomaly score is defined , the farther the decision function value is from the classification boundary (that is, the larger the absolute value), the more likely the data point belongs to the abnormal category. The abnormal score is defined as the absolute value of the decision function value or its converted form. The process is: ,in, Indicates the input The output anomaly score of each periodic data point feature.

[0021] The dynamic feature capture process of the periodic data is as follows: Based on periodic data points, the window size of the sliding window is defined as , the window step size is , calculate the statistics within each sliding window, the statistics within the window include the window mean and window standard deviation , window mean ,in, is not equal to within the window The periodic data point characteristics, window standard deviation ; The statistics within the sliding window can represent the dynamic characteristics of periodic data, because these statistics can capture the changes in data within different time windows. By calculating the window mean and window standard deviation of the statistics within the sliding window, we can observe the local trends and local fluctuations of periodic data in each time period. Furthermore, the window mean reflects the local trend, and the window mean reflects the average state of the data within a certain time window. For periodic data, the window mean can capture the average behavior of the data in a shorter time range, thereby revealing local trends and periodic patterns. The window standard deviation reflects local fluctuations. For periodic data, the window standard deviation can capture the fluctuation amplitude of the data within the local time range, thereby revealing the intensity of the periodic pattern. If the data fluctuates greatly in certain time periods, the window standard deviation will increase accordingly; conversely, if the data is relatively stable, the window standard deviation will be smaller, thereby obtaining the dynamic characteristics of periodic data.

[0022] The process of setting the abnormal behavior threshold is as follows: Set abnormal behavior thresholds based on periodic characteristics and anomaly scores , abnormal behavior threshold The acquisition process is: ,in, For the window The average value of the periodic data point features, For the window The standard deviation of the periodic data point characteristics, is a hyperparameter; Compare the anomaly score with the abnormal behavior threshold. hour, The corresponding periodic data point definition is normal, when hour, The corresponding periodic data points are defined as abnormal, and the bidding data corresponding to the periodic data points defined as abnormal are cleared.

[0023] Furthermore, if it is confirmed that the abnormal data point does represent an abnormal situation, the abnormal threshold can be adjusted according to the actual situation. This may involve the following situations: Increase Value: If the threshold is too low, causing many normal data points to be misjudged as abnormal, you can increase The value of makes the threshold higher and reduces false positives; Reduce Value: If the threshold is too high, causing some truly abnormal data points to not be identified, you can reduce The value of makes the threshold lower and improves the sensitivity of anomaly detection.

[0024] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A bidding supervision system for engineering projects, comprising a data acquisition module for acquiring bidding data, It is characterized in that The bidding data supervision module includes: Time series construction module: used to obtain time series data of bidding data and obtain similarity ontology features based on time series data; Time series management module: Create a time series feature matrix based on similarity ontology features, extract periodic data point features from the time series feature matrix, and use the time series decomposition algorithm to separate the periodic data point features to obtain periodic data points; Anomaly analysis module: used to build an anomaly analysis model, input periodic data points into the anomaly analysis model, and obtain the anomaly score of each periodic data point; Dynamic adjustment module: used to capture the dynamic characteristics of periodic data points, calculate the abnormal behavior threshold based on the dynamic characteristics of periodic data points, and clear the bidding data corresponding to the periodic data points whose abnormal scores are greater than the abnormal behavior threshold.

2. A bidding supervision system for engineering projects according to claim 1, It is characterized in that The process of obtaining the time series data is as follows: Collect bidding data, remove missing values ​​or outliers from the collected bidding data, unify the format, and extract the time series, which can be expressed as ,in, is the length of the time series, Indicates that the length of the time series is The time series data corresponding to .

3. A bidding supervision system for engineering projects according to claim 1, It is characterized in that The process of obtaining the similarity ontology feature is as follows: Define the time interval for selecting data points in the time series as and embedding dimension ; Time interval The definition process is: Based on time series data , define the time series and the time series delayed version The mutual information between them is: ； in, For time series data in time series and time series data Delayed version of The joint probability distribution function of For time series data The marginal probability distribution function of Delayed version of time series data The marginal probability distribution function of ; Based on mutual information Draw Different The mutual information curve of , find the first local minimum in the mutual information curve, and the first local minimum in the mutual information curve corresponding to As the optimal time interval; Embedding Dimension The definition process is: Defining the embedding dimension The initial dimension is 1, based on time series data Constructing embedding vectors , ,in, and For the embedding dimension When, time series data Two different embedding vectors constructed, and For two different indexes, is the time interval, and The index is kept between 1 and Between, calculate each pair of embedding vectors and The Euclidean distance between , ,in, is the embedding vector index; Increase the embedding dimension arrive , construct a new embedding vector , ,in, and All are embedded in the dimension When, time series data Two different embedding vectors are constructed, and we get and The Euclidean distance between , ,definition and The Euclidean distance between and The ratio of the Euclidean distance between ,in, express The norm of express The norm of , defines the embedding threshold as 0, and selects The smallest value closest to the embedding threshold As the final embedding dimension; Based on time interval and embedding dimension , the time series Mapping to the embedding dimension In the dimensional phase space, create two different Dimensional vector and , , Among them, the time point Limited to arrive In the definition and The direct similarity is , ,in is the standard deviation parameter, for and The square of the Euclidean distance between them, based on direct similarity, extracting time series data Similarity ontology features, obtain the average similarity measure ,in, Represents all vectors of different dimensions The average similarity ontology features are used to construct the similarity matrix, and the principal component analysis of the similarity matrix is ​​performed to obtain the similarity ontology features.

4. A bidding supervision system for engineering projects according to claim 1, It is characterized in that The process of obtaining the periodic data points is as follows: Based on the average similarity metric , construct the similarity time series matrix , expressed as ,in The embedding dimension is ; Based on the similarity time series matrix Perform spectrum analysis and define similarity time series matrix The spectrum matrix is , convert the similarity time series matrix to the frequency domain and obtain the spectrum matrix ,in, is the frequency index, is the base of the exponential function; Define the spectrum matrix The seasonal component of , based on the spectrum matrix application Decomposition, collect the seasonal components of the time series after each eigenvector decomposition ,in Represents a new sequence, based on each new sequence , for the new sequence Apply spectrum analysis to obtain the period , use cycle , locate periodic data points in the time series.

5. A bidding supervision system for engineering projects according to claim 1, It is characterized in that The training process of the anomaly analysis model is: Convert the periodic data points into a transaction data item format to obtain the transaction items of the periodic data points; Generate candidate item sets for transaction items. The generation process of candidate item sets is as follows: Starting from a single transaction item, all item set combinations are generated based on the transaction data items. For each candidate item set, its support in the transaction database is calculated and the minimum support threshold is set. , each item set is represented as , calculate the support of each item set, discard the item sets whose support is lower than the minimum support threshold, and the remaining items are frequent item sets. The support formula is: ,in, Is the included item set The number of transactions, is the total number of transactions for the transaction data item, Representing Item Sets The support of , get the frequent itemsets; Define the minimum confidence threshold , the two mutually exclusive items in the frequent itemset are and , based on frequent item sets, generate association rules, the form of association rules is , calculate the confidence of the association rule, confidence ,in, for and The proportion of transactions based on the minimum confidence threshold , for confidence levels below the low confidence threshold Itemset Screen out and get the strong association rule item set; Based on the strong association rule item set and periodic data points, define the first The characteristics of a periodic data point are expressed as , the first The item set feature is expressed as , periodic data point characteristics and item set features Compose a dataset , divide the data set into training set and test set, train the anomaly analysis model, define the minimization loss function, and the formula for minimizing the loss function is: ,in, is the weight vector, is the bias term, is a non-negative slack variable, is the penalty parameter, is the sample size; Define the kernel function, the formula of the kernel function is: ,in, is the parameter of the kernel function, is the output value of the kernel function, and For the dataset For any two vectors in express and The Euclidean distance between Define constraints and use them to optimize the abnormal analysis model. The optimization method is as follows: , ,in, is the transposed representation of the weight vector, is the feature vector The category labels include 1 or -1, = 1, the eigenvector Belongs to the positive category, = −1, the eigenvector Belongs to the negative class and obtains an anomaly analysis model.

6. A bidding supervision system for engineering projects according to claim 1, It is characterized in that The process of obtaining the abnormal score of the periodic data point is as follows: Apply the anomaly analysis model to the new periodic data points, calculate the anomaly score of each data point, and define the anomaly score. The formula for the anomaly score is: ,in, Indicates the input The output anomaly score of each periodic data point feature.

7. A bidding supervision system for engineering projects according to claim 1, It is characterized in that The dynamic feature capture method of the periodic data is: Based on periodic data points, the window size of the sliding window is defined as , the window step size is , calculate the statistics within each sliding window, the statistics within the window include the window mean and window standard deviation , window mean ,in, is not equal to within the window The periodic data point characteristics, window standard deviation , the dynamic characteristics of periodic data are obtained through statistics within the window.

8. A bidding supervision system for engineering projects according to claim 1, It is characterized in that The calculation process of the abnormal behavior threshold is: Set abnormal behavior thresholds based on the dynamic characteristics of periodic data points , abnormal behavior threshold ,in, For the window The average value of the periodic data point features, For the window The standard deviation of the periodic data point characteristics, is a hyperparameter; when hour, The corresponding periodic data point definition is normal, when hour, The corresponding periodic data points are defined as abnormal, and the bidding data corresponding to the periodic data points defined as abnormal are cleared.

Images