Construction unit price intelligent prediction method based on standard engineering quantity list

Abstract

The invention discloses a construction unit price intelligent prediction method based on a standard bill of quantities. The method comprises the following steps: S1, investigating and surveying influence factors on construction unit prices of the bill of quantities; s2, querying target data in a database, and cleaning the data according to rules; s3, data exploration; s4, performing feature screening; s5, carrying out statistics on all types of data containing each discrete feature, and coding each discrete feature by using a One-Hot method; s6, dividing a training set and a test set by adopting a random sampling method; s7, converting original data through feature engineering; s8, adjusting and optimizing the model; and S9, for the project quantity list of the construction unit price to be predicted, inputting the corresponding influence factors into the trained random forest model, and predicting to obtain a predicted value of the cost. According to the method, the problems of poor training effect and huge data training caused by not-in-place data cleaning are solved; and predicting the construction cost of the list granularity through a machine learning algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent construction unit price, and more particularly to an intelligent prediction method for construction unit price based on a standard bill of quantities. Background Art

[0002] In the field of data mining for intelligent construction unit price, its importance is becoming increasingly prominent with the development of technology and the evolution of industry needs. This field is at the intersection of computer science and engineering, involving multiple technologies such as data mining, statistics, and machine learning, aiming to bring more intelligent and efficient solutions to construction unit price management. With the increasing complexity of current engineering projects, the amount of data has shown a rapid growth trend, and traditional construction unit price calculation methods are no longer applicable to cost prediction for such large amounts of data. From a technical perspective, data mining technology comprehensively uses a variety of algorithms and analysis methods, including regression analysis, time series analysis, neural networks, clustering analysis, and association analysis. These algorithms can deeply process and analyze large amounts of engineering data to mine valuable information contained therein.

[0003] In terms of specific application scenarios, the first is engineering cost prediction. By mining a large amount of historical construction unit price data, this technology can extract various features of the project, such as project scale, building type, geographical area, material usage, construction time, etc., and use corresponding data mining algorithms to analyze the potential relationship between these features and costs. Based on this, a cost prediction model can be established for new engineering projects, so as to accurately predict their costs and provide strong support for project budgeting and decision-making. Secondly is cost control and deviation analysis. In the process of project implementation, the mining of real-time data is particularly important. With the help of data mining technology, the real-time change trend of project costs can be continuously monitored, which helps to detect cost deviations in a timely manner. At the same time, by deeply mining these data, various factors leading to cost deviations can be identified, including key factors such as fluctuations in material prices, delays in the construction period, and changes in construction technology. Furthermore, detailed cost control suggestions can be provided for project managers to ensure that project costs are within a controllable range and improve the cost performance of the project. Thirdly is the optimal allocation of resources. Mining the data of construction resources such as manpower, materials, and equipment involved in the construction process can help achieve the optimal allocation of resources. By analyzing the resource usage of different projects at different stages, data mining technology can be used to find the most economical and efficient resource allocation mode and the best cost-benefit ratio. For example, through clustering analysis, projects or resources can be classified, and resource allocation schemes matching the characteristics of different types of projects can be recommended, thus avoiding waste of resources and unreasonable increase in costs. Finally is supplier management. Data mining can comprehensively analyze various aspects of suppliers, including price, quality, delivery date, reputation, etc. Classification algorithms can be used to divide suppliers into different categories, screen out high-quality suppliers, and at the same time, through association analysis, identify the key attributes of high-quality suppliers, providing a reliable basis for selecting new suppliers. Moreover, by deeply mining the price data of suppliers, the future price trend can be predicted, which provides important information support for procurement negotiations and contract signing, and helps enterprises make more informed decisions in the procurement link.

[0004] The data sources required for data mining are extensive. Among them, historical project data is a very important part, which contains rich information such as cost details, design documents, construction records, change records, and settlement documents of completed engineering projects. These information constitute the basis of cost prediction and control models and provide rich samples for subsequent data mining. In addition, market data is also indispensable. It covers the market price information of various building materials, equipment, and labor. Mining these data can help master the fluctuation laws and trends of prices and provide necessary support for cost estimation and cost adjustment of engineering projects.

[0005] In summary, data mining technology plays an important role in the field of intelligent construction unit prices. By integrating various technical means and delving into multiple application scenarios, it promotes the development of construction unit price management towards a more intelligent, accurate, and efficient direction, providing strong guarantees for the smooth implementation of engineering projects and cost control.

[0006] The construction cost calculation methods involved in the prior art mainly include the following two:

[0007] Method 1: Intelligent construction cost calculation method based on the engineering bill:

[0008] Based on the obtained sample data, establish a standard engineering bill cost library. The standard engineering bill cost library should include project name, measurement unit, bill code, bill parameters, labor cost unit price, machinery cost unit price, material cost unit price, and management cost unit price;

[0009] Establish a matching rule library for non-standard names and standard names. The matching rule library should include the non-standard name, standard name, and bill location code; the non-standard name is the project name in the engineering bill of the proposed project; the standard name is the project name in the standard engineering bill cost library;

[0010] Obtain the engineering bill data of the proposed project;

[0011] According to the project name in the obtained engineering bill data, through the matching rule library, obtain all the data that meet the matching rules in the standard engineering bill cost library;

[0012] Extract the bill parameter values in the project characteristics of the engineering bill data according to the bill location code, and through calculation, select the only cost data that conforms in the standard engineering bill cost library;

[0013] If the result of step (5) is unique, it means the result is accurate; if the result is not unique or the result does not exist, check whether the matching rule library and the standard engineering bill cost library are established completely and supplement and improve them;

[0014] Calculate the construction cost according to the selected cost data;

[0015] Construction cost = quantity of work in the engineering bill data * [labor cost unit price + machinery cost unit price + material cost unit price * (1 + material loss rate) + management cost unit price].

[0016] Method 2: Construction cost prediction method for construction projects based on extreme learning machine:

[0017] Construct an index system that affects the construction cost of construction projects;

[0018] Adopt the principal component analysis method to extract the main influencing indicators in the index system;

[0019] The quantified values of the construction costs of construction projects and the corresponding main influencing indicators are combined into a sample data, and the quantified values of the construction costs and the corresponding main influencing indicators of several construction projects are collected to construct a sample data set;

[0020] Construct an extreme learning machine and use the sample data set to complete the network training of the extreme learning machine;

[0021] For a construction project whose construction cost is to be predicted, the quantified value of its main influencing indicator is input into the trained extreme learning machine to obtain the predicted value of the construction cost.

[0022] The existing intelligent construction unit price method based on the bill of quantities mainly relies on manual operations, and there are many significant drawbacks in this mode, which seriously restricts the efficient development of the construction unit price calculation work. First, manual operations lead to an extremely large workload and cumbersome operations in the calculation process. During the calculation process, staff need to manually process and calculate a large amount of data. Not only is the task heavy, but also human errors are prone to occur, greatly reducing work efficiency and thus affecting the progress of the entire construction unit price calculation. This may lead to project schedule delays, increasing the time cost and wasting resources for engineering projects. Second, this method lacks the ability of intelligent learning and automatic adjustment of matching rules. In the data recording of construction unit price calculation, there are serious problems of inconsistent terms. Different personnel or units may use different expressions. For example, for the same type of material or construction project, there may be multiple statements, and the existing cost calculation method cannot uniformly classify these diverse expressions through intelligent means. This chaotic state of data brings great difficulties to subsequent cost calculation and analysis, and it is difficult to ensure the accuracy and standardization of data. Third, when facing a huge amount of construction unit price data, the limitations of manual analysis are prominent. Due to the relatively slow speed of manual operations, it takes a large amount of manpower and time to complete the analysis of a large amount of data. Moreover, the influence of human factors on the results is relatively large, and different people may result in the difficulty of ensuring the accuracy and consistency of calculation results due to differences in experience, professional level, and subjective judgment. Finally, most of the existing methods only focus on the granularity of a single project and lack in-depth and intelligent analysis methods for the bill of quantities. This means that they can only conduct cost accounting for projects at a relatively simple level, unable to fully explore the information hidden in the bill of quantities, and it is also difficult to achieve comprehensive, systematic, and refined analysis of project costs. In view of the above problems, artificial intelligence technology is introduced to use its advantages to quickly and intelligently classify and analyze data, so as to significantly improve the efficiency and data quality of construction unit price calculation. By integrating artificial intelligence technology into the construction unit price calculation, it is expected to overcome many deficiencies of traditional manual methods, promote the construction unit price calculation work to be carried out more scientifically, efficiently, and accurately, and provide strong guarantee for the smooth progress and cost control of engineering projects.

[0023] The information disclosed in this background art section is only intended to deepen the understanding of the general background art of the present invention, and should not be regarded as an admission or any form of implication that this information constitutes the prior art known to those skilled in the art. Summary of the Invention

[0024] The object of the present invention is to provide a construction unit price intelligent prediction method based on a standard bill of quantities to solve the problems existing in the prior art.

[0025] To achieve the above object, the present invention adopts the following technical solutions:

[0026] The present invention provides an intelligent prediction method for construction unit prices based on a standard bill of quantities, comprising the following steps:

[0027] S1. Investigate the influencing factors on the construction unit price of the bill of quantities;

[0028] S2. Query the target data in the database and clean the data according to the rules;

[0029] S3. Data exploration: Draw a frequency distribution histogram and a box plot for each feature, and delete outliers according to the box plot;

[0030] S4. Feature screening: Screen the features according to the KDE distribution plots on the training set and the test set, and delete the features with large differences;

[0031] S5. Count the data containing all categories of each discrete feature, encode each discrete feature using the One-Hot method, and save the encoder;

[0032] S6. Adopt the method of random sampling to divide the training set and the test set, with a ratio of 7:3;

[0033] S7. Transform the original data through feature engineering, construct a random forest model, and train it using the incremental learning method;

[0034] S8. Model tuning: First, use the learning curve to determine the value range of the main parameters of the random forest, and then use the grid search method to determine the optimal parameters;

[0035] S9. For the bill of quantities for which the construction unit price is to be predicted, input the corresponding influencing factors into the trained random forest model to predict the predicted value of the cost.

[0036] Further, the specific implementation method of step S2 is as follows:

[0037] S2.1. Count all the data associated with the standard bill of quantities;

[0038] S2.2. Screen the data. The bill of quantities data associated with each standard bill of quantities should meet the following conditions: the number of bills > 400 and the number of covered projects > 50;

[0039] S2.3. Clean the bill of quantities associated with each standard bill of quantities. The cleaning rules are as follows:

[0040] 1) Clean unreasonable price data; 2) Data with units that do not conform to the standard bill of quantities units; 3) Data with empty quantity or empty unit price; 4) Data with incorrect standard list attachment; 5) Number of lists > 400; 6) Number of covered projects > 50.

[0041] Further, the specific implementation method of step S3 is as follows:

[0042] Draw a histogram and a box plot, and observe whether the distribution of the frequency histogram conforms to the business situation; for the box plot, if there are values in the data that are less than Q1 - 1.5×IQR or greater than Q3 + 1.5×IQR, they are considered outliers and need to be removed;

[0043] Among them: the lower boundary of the box is Q1, the upper boundary is Q3, the line in the middle of the box represents the median Q2, and IQR is the length of the box: IQR = Q3 - Q1.

[0044] Further, the specific implementation method of step S4 is as follows:

[0045] S4.1. Screen features according to the KDE distribution diagrams of the training set and the test set, and delete features with large differences;

[0046] S4.2. Find k features most relevant to the target, and feature variables with a Pearson correlation coefficient greater than 0.5 with the target;

[0047] Calculation formula: Suppose there are two variables X = (x1, x2,... x n ) and Y = (y1, y2,... y n ), and their Pearson correlation coefficient ρ X , Y The calculation formula is:

[0048]

[0049] Among them is the mean of variable X, is the mean of variable Y.

[0050] Further, the specific implementation method of step S5 is as follows:

[0051] For each discrete feature, perform One-Hot encoding to obtain an encoder and save the encoder;

[0052] The One-Hot encoding method is as follows:

[0053] For each observation value x " (i = 1, 2,..., n, where n is the number of observation values) in the categorical variable X, construct a vector y of length k " = (y"1 , y "2 ,..., y "k ) to perform One - Hot encoding;

[0054] If x " belongs to category C ％ , then y "％ = 1, and for all l ≠ j, y "％ = 0;

[0055] where j = 1, 2,... k..

[0056] Furthermore, the specific implementation method of step S7 is as follows:

[0057] S7.1. Read 100,000 pieces of data;

[0058] S7.2. Separate features and labels;

[0059] S7.3. Perform min - max normalization on continuous features. For a given set of data x = 0x1, x2,..., x n 1, where x 2"n is the minimum value of this set of data, and x 234 is the maximum value of this set of data. After min - max normalization, the data is x n562 ;

[0060] The mathematical formula for min - max normalization is:

[0061] S7.4. Perform Yeo - Johnson transformation on continuous features. The purpose is to make the data closer to a normal distribution. The calculation formula is as follows:

[0062] When x >= 0:

[0063]

[0064] When x < 0:

[0065]

[0066] λ is the parameter of the transformation, and its optimal value is determined by a certain method to make the transformed data closest to a normal distribution;

[0067] S7.5. Use the encoder in step S5 to transform the discrete features in the data;

[0068] S7.6. Use PCA (Principal Component Analysis) for dimensionality reduction;

[0069] S7.7. Perform model training;

[0070] S7.8. Increase the number of n_estimators.

[0071] Furthermore, the specific implementation method of step S8 is as follows:

[0072] S8.1. Initially determine the parameter range through the learning curve:

[0073] n_estimators - the number of trees, max_depth - the maximum depth of the tree, min_samples_split - the minimum number of samples required for internal node re - division, min_samples_leaf - the minimum number of samples in a leaf node, max_features - the maximum number of features considered;

[0074] S8.2. Select the evaluation metric:

[0075] Coefficient of determination (R 2 ): The calculation formula is where is the predicted value of the model, is the mean of the true values, and n is the number of samples; The value range of R 2 is (-∞, 1), and the closer it is to 1, the better the model fitting effect;

[0076] S8.3. Parameter search method:

[0077] Grid search: Grid search will exhaust all possible parameter combinations;

[0078] S8.4. Cross - validation to evaluate the model performance:

[0079] Adopt k - fold cross - validation, set k = 10; Divide the dataset into k parts, each time use k - 1 parts as the training set, and the remaining 1 part as the validation set to obtain k model performance evaluation results;

[0080] S8.5. Select the best parameter combination:

[0081] Compare the evaluation metrics of different parameter combinations after cross - validation, and select the parameter combination with the best performance as the final model parameters.

[0082] Adopting the above technical solution, the present invention has the following beneficial effects:

[0083] The present invention solves the problems of poor training effect caused by insufficient data cleaning and training of huge amounts of data; realizes the prediction of construction cost at the list granularity through machine learning algorithms. Description of the Drawings

[0084] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0085] Figure 1 This is the flowchart of the intelligent prediction method for construction unit prices based on the standard bill of quantities of the present invention. Specific embodiments

[0086] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the drawings. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0087] The following will detail the specific embodiments of the present invention in conjunction with the drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0088] Combined with Figure 1 As shown, the present invention provides an intelligent prediction method for construction unit prices based on the standard bill of quantities, and the steps are as follows:

[0089] Step 1: Investigate the influencing factors of the construction unit price of the bill of quantities.

[0090] Step 2: Query the target data in the database and clean the data according to the rules.

[0091] Step 3: Data exploration, draw the frequency distribution histogram and box plot of each feature, and delete outliers according to the box plot.

[0092] Step 4: Feature screening, screen the features according to the KDE distribution diagrams on the training set and the test set, and delete the features with large differences.

[0093] Step 5: Count the data containing all categories of each discrete feature, encode each discrete feature using the One - Hot method, and save the encoder.

[0094] Step 6: Use the method of random sampling to divide the training set and the test set, with a ratio of 7:3.

[0095] Step 7: Transform the original data through feature engineering, construct a random forest model, and train it using the incremental learning method.

[0096] Step 8: Model Tuning. First, use the learning curve to determine the value range of the main parameters of the random forest, and then use the grid search method to determine the optimal parameters.

[0097] Step 9: For the bill of quantities for which the construction unit price is to be predicted, input the corresponding influencing factors into the trained random forest model to predict the predicted value of the cost.

[0098] Detailed process of Step 2:

[0099] 1. Statistically analyze all the data linked to the standard bill of quantities.

[0100] 2. Screen the data. The bill of quantities data linked to each standard bill of quantities should meet the following conditions: the number of bills > 400 and the number of covered projects > 50.

[0101] 3. For the bill of quantities linked to each standard bill of quantities, cleaning is required. The cleaning rules are as follows:

[0102] 1) Clean unreasonable price data; 2) Data with units that do not conform to the units of the standard bill of quantities; 3) Data with empty quantities or empty unit prices; 4) Data with incorrect linked standard bills; 5) The number of bills > 400; 6) The number of covered projects > 50.

[0103] Detailed process of Step 3:

[0104] Draw a histogram and a box plot to observe whether the distribution of the frequency histogram conforms to the business situation; for the box plot, values less than Q1 - 1.5×IQR or greater than Q3 + 1.5×IQR in the data are considered outliers and need to be removed.

[0105] Among them: the lower boundary of the box is Q1, the upper boundary is Q3, the line in the middle of the box represents the median Q2, and IQR is the length of the box: IQR = Q3 - Q1.

[0106] Detailed process of Step 4:

[0107] 1. Screen the features based on the KDE distribution diagrams of the training set and the test set. Features with large differences need to be deleted;

[0108] 2. Find k features most relevant to the target, that is, feature variables with a Pearson correlation coefficient greater than 0.5 with the target;

[0109] Calculation formula: Suppose there are two variables X = (x1, x2,... x n ) and Y = (y1, y2,... y n ), and their Pearson correlation coefficient ρ X , Y The calculation formula is:

[0110]

[0111] wherein is the mean value of variable X, is the mean value of variable Y

[0112] Detailed process of step five:

[0113] Perform One-Hot encoding on each discrete feature to obtain an encoder and save the encoder.

[0114] The One-Hot encoding method is as follows:

[0115] For each observation value x in the categorical variable X " (i = 1, 2,..., n, where n is the number of observation values), we construct a vector y of length k " =(y "1 , y "2 ,..., y "k ) for One-Hot encoding

[0116] If x " belongs to category C ％ , then y "％ = 1, and for all l ≠ j, y "％ = 0.

[0117] where j = 1, 2,... k.

[0118] Detailed process of step seven:

[0119] 1. Read 100,000 pieces of data.

[0120] 2. Separate the features and labels.

[0121] 3. Perform min-max normalization on continuous features. For a given set of data x = 0x1, x2,..., x n 1, where x 2"n is the minimum value of this set of data, and x 234 is the maximum value of this set of data. After min-max normalization, the data is x n562 .

[0122] The mathematical formula for min-max normalization is:

[0123] Perform Yeo-Johnson transformation on continuous features to make the data closer to a normal distribution. The calculation formula is as follows:

[0124] When x >= 0:

[0125]

[0126] When x<0:

[0127]

[0128] λ is the parameter of the transformation, and its optimal value needs to be determined by certain methods (such as maximum likelihood estimation) to make the transformed data closest to the normal distribution.

[0129] 5. Use the encoder in step 5 to convert the discrete features in the data.

[0130] 6. Use PCA (Principal Component Analysis) for dimensionality reduction

[0131] 7. Perform model training

[0132] 8. Increase the number of n_estimators (number of trees)

[0133] Detailed process of step eight:

[0134] 1. Preliminary determination of parameter range through learning curve:

[0135] n_estimators (number of trees): You can start with a smaller range, such as 50-200, with an interval of 10. This parameter mainly affects the stability and generalization ability of the model. A larger number of trees generally makes the model more stable, but increases the computational cost.

[0136] max_depth (maximum depth of the tree): Try starting from around 3-8. It controls the growth depth of the decision tree. A shallow depth may lead to underfitting, while a deep depth may lead to overfitting.

[0137] min_samples_split (minimum number of samples required for internal node re-division): The range can be set to 2-10. This parameter determines the minimum number of samples a node must contain before it is split.

[0138] min_samples_leaf (minimum number of samples for leaf nodes): Try to set it from 1 to 5. It ensures that each leaf node contains enough samples to prevent overfitting due to too few samples in the leaf node.

[0139] max_features (Maximum number of features considered): You can try 'auto' (automatically select, usually the square root of the number of features),'sqrt' (square root of the number of features), 'log2' (logarithm to the base 2), or a specific integer (such as 1 / 3 of the number of features, etc.). It controls the maximum number of features considered when splitting nodes each time, and is used to balance the randomness of the model and prevent overfitting.

[0140] 2. Select evaluation metrics:

[0141] Coefficient of determination (R 2 ): The calculation formula is where is the predicted value of the model, is the mean of the true values, and n is the number of samples. The value range of R 2 is (-∞, 1), and the closer it is to 1, the better the model fitting effect.

[0142] 3. Parameter search method:

[0143] Grid search: Grid search will exhaustively enumerate all possible parameter combinations. This method can find the optimal combination within the given parameter range.

[0144] 4. Cross-validation to evaluate model performance:

[0145] Adopt k-fold cross-validation, set k = 10. Divide the dataset into k parts, and each time use k - 1 of them as the training set and the remaining 1 part as the validation set, so that k model performance evaluation results can be obtained.

[0146] 5. Select the best parameter combination:

[0147] Compare the evaluation metrics of different parameter combinations after cross-validation, and select the parameter combination with the best performance as the final model parameters. For example, if the average MSE of a parameter combination is the smallest after 10-fold cross-validation, then select this parameter combination to build the final random forest regression model.

[0148] Preferably, the method for removing outliers in Step 3 of the present application can also adopt methods such as standard deviation. Step 4 can perform feature screening through analysis of variance and chi-square test. Step 5 can be replaced by other encoding methods, such as Label Encoding. In Step 7, there will be other combinations for feature engineering training. The random forest model can perform incremental learning, but there are also other models that can perform incremental learning, such as K-nearest neighbor and LightGBM models, which can be replaced by other machine learning or deep learning models. In Step 8, there can be other methods for tuning parameters. For different models, the parameters are different, and the impact on the model is different. There are also various methods for parameter search: such as random grid search and half-and-half grid search.

[0149] Compared with the prior art, the present invention overcomes the following technical difficulties existing in the prior art:

[0150] Difficulty 1: The background is difficult to replicate, it is difficult to establish a standard bill of quantities, and it is difficult to generate enough data.

[0151] Difficulty 2: The methods for statistical data, including data cleaning, are summarized from practice.

[0152] Difficulty 3: Step 5 is an encoding method designed specifically for a large amount of data.

[0153] Difficulty 4: Step 4 is a training method designed for training a huge amount of data.

[0154] Difficulty 5: Model tuning is adjusted according to the characteristics of model parameters.

[0155] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent prediction method for construction unit prices based on a standard bill of quantities, characterized in that It includes the following steps: S1. Investigate the influencing factors of the construction unit price of the bill of quantities. S2. Query the target data in the database and clean the data according to the rules. S3. Data exploration: Draw the frequency distribution histogram and box plot of each feature, and delete the outliers according to the box plot. S4. Feature screening: Screen the features according to the KDE distribution plots on the training set and the test set, and delete the features with large differences. S5. Count the data containing all categories of each discrete feature, encode each discrete feature using the One-Hot method, and save the encoder. S6. Use the method of random sampling to divide the training set and the test set, with a ratio of 7:

3. S7. Transform the original data through feature engineering, construct a random forest model, and use the incremental learning method for training. S8. Model tuning: First, use the learning curve to determine the value range of the main parameters of the random forest, and then use the grid search method to determine the optimal parameters. S9. For the bill of quantities whose construction unit price is to be predicted, input the corresponding influencing factors into the trained random forest model to predict the predicted value of the cost.

2. The intelligent prediction method for construction unit price based on standard bill of quantities according to claim 1, characterized in that The specific implementation method of step S2 is as follows: S2.

1. Count all the data associated with the standard bill of quantities. S2.

2. Screen the data. The bill of quantities data associated with each standard bill of quantities should meet the following conditions: the number of bills > 400 and the number of covered projects > 50. S2.

3. Clean the bill of quantities associated with each standard bill of quantities. The cleaning rules are as follows: 1) Clean the unreasonable price data; 2) Data with units not conforming to the units of the standard bill of quantities; 3) Data with empty quantity or empty unit price; 4) Data with incorrect association with the standard list; 5) The number of bills > 400; 6) The number of covered projects > 50.

3. A construction unit price intelligent prediction method based on a standard bill of quantities according to claim 1, characterized in that The specific implementation method of step S3 is as follows: Draw the histogram and box plot, and observe whether the distribution of the frequency histogram conforms to the business situation; for the box plot, if there are values in the data that are less than Q1 - 1.5×IQR or greater than Q3 + 1.5×IQR, they are considered outliers and need to be removed; Where: the lower boundary of the box is Q1, the upper boundary is Q3, the line in the middle of the box represents the median Q2, and IQR is the length of the box: IQR = Q3 - Q1.

4. The intelligent prediction method for construction unit price based on standard bill of quantities according to claim 1, characterized in that, The specific implementation method of step S4 is as follows: S4.

1. Screen the features according to the KDE distribution plots on the training set and the test set, and delete the features with large differences. S4.

2. Find k features most relevant to the target, and feature variables with a Pearson correlation coefficient with the target greater than 0.

5. Calculation formula: Suppose there are two variables X = (x1, x2,... x n ) and Y = (y1, y2,... y n ), and their Pearson correlation coefficient ρ X , Y The calculation formula is: wherein is the mean value of variable X, is the mean value of variable Y.

5. A construction unit price intelligent prediction method based on a standard bill of quantities according to claim 1, characterized in that The specific implementation method of step S5 is as follows: Perform One-Hot encoding for each discrete feature to obtain the encoder, and save the encoder. The One-Hot encoding method is as follows: For each observed value x in the categorical variable X " (i = 1, 2,..., n, where n is the number of observed values), construct a vector y of length k " =(y "1 , y "2 ,..., y "k ) for One - Hot encoding; If x " belongs to category C ％ , then y "％ = 1, and for all l ≠ j, y "％ = 0; where j = 1, 2,... k.

6. The intelligent prediction method for construction unit price based on standard bill of quantities according to claim 1, characterized in that The specific implementation method of step S7 is as follows: S7.

1. Read 100,000 pieces of data. S7.

2. Separate the features and labels. S7.

3. Perform min-max normalization on continuous features. For a given set of data x = {x1, x2,..., x n}, where x 2"n is the minimum value of this set of data, and x 234 is the maximum value of this set of data, the data after min-max normalization is x n562 ; The mathematical formula for min-max normalization is: S7.

4. Perform Yeo-Johnson transformation on the continuous features, aiming to make the data closer to the normal distribution. The calculation formula is as follows: When x >= 0: When x < 0: λ is the parameter of the transformation, and its optimal value is determined by a certain method to make the transformed data closest to the normal distribution; S7.

5. Use the encoder in step S5 to transform the discrete features in the data; S7.

6. Use PCA (Principal Component Analysis) for dimensionality reduction; S7.

7. Conduct model training; S7.

8. Increase the number of n_estimators.

7. A method for intelligent prediction of construction unit prices based on a standard bill of quantities according to claim 1, characterized in that The specific implementation method of step S8 is as follows: S8.

1. Initially determine the parameter range through the learning curve: n_estimators - the number of trees, max_depth - the maximum depth of the tree, min_samples_split - the minimum number of samples required for internal node re - division, min_samples_leaf - the minimum number of samples in the leaf node, max_features - the maximum number of features considered; S8.

2. Select the evaluation metric; Coefficient of determination (R 2 ): The calculation formula is where is the predicted value of the model, is the mean of the true values, and n is the number of samples; The value range of R 2 is (-∞, 1), and the closer it is to 1, the better the model fitting effect; S8.

3. Parameter search method: Grid search: Grid search will exhaust all possible parameter combinations; S8.

4. Use cross - validation to evaluate the model performance: Adopt k - fold cross - validation, set k = 10; divide the dataset into k parts, each time use k - 1 parts as the training set, and the remaining 1 part as the validation set to obtain k model performance evaluation results; S8.

5. Select the best parameter combination: Compare the evaluation metrics of different parameter combinations after cross - validation, and select the parameter combination with the best performance as the final model parameters.

Images