A project data mining-based construction engineering company project matching degree analysis method

Abstract

The application discloses a kind of based on project data mining's construction engineering company project matching degree analysis method, the method includes collecting multiple construction engineering company corresponding multiple engineering project information as metadata set, classification is carried out to metadata set, and metadata set is preprocessed;Establish project matching prediction model CDF, and input model after pre-processing metadata and output prediction result;The prediction result of CDF model is compared with real label, and the parameters of model are iteratively updated until model converges;Real engineering project data is input into model, the matching degree value corresponding to each construction engineering company of the engineering project is predicted, and the top-n construction engineering company with the highest matching degree value is provided, and finally the relatively matched construction engineering company can be accurately and objectively selected, and the selection efficiency is improved.The application solves the problem of low accuracy and high labor cost caused by manual screening in traditional methods.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of construction engineering and data mining, and particularly relates to a construction engineering company project matching degree analysis method. BACKGROUND

[0002] In the field of construction engineering, it is crucial to establish a reasonable match between construction engineering companies and construction engineering projects. Different engineering projects have different characteristics and different construction difficulties, and construction engineering companies are the direct executors of engineering projects, playing a crucial role in the smooth implementation of engineering projects.

[0003] Arranging an unreasonable construction engineering company for an engineering project may lead to a series of problems such as slow construction progress, unreasonable allocation of engineering resources, and unsafe construction process. Matching the best construction engineering company for a construction engineering project is beneficial for the rational use of resources, controlling the cost of project development, and shortening the development cycle of construction engineering projects.

[0004] Currently, the informatization level of the construction engineering field is slightly insufficient, and there are fewer mature technologies related to it. Most of the solutions to the matching problem between construction engineering companies and engineering projects still rely on traditional methods. Traditional construction engineering project matching analysis methods are highly dependent on expert experience, have limited information, and most of the work involves manual work. Therefore, they are low in efficiency, prone to errors, and consume a lot of manpower and resources, which may not be able to propose a reasonable matching solution, leading to difficulties in the smooth completion of construction engineering projects. In order to make the construction process more smooth and achieve better construction results, a scientific method is needed to reasonably analyze the matching degree between construction engineering companies and construction engineering projects, and to efficiently match the best construction engineering company for a large number of construction engineering projects in a short period of time. SUMMARY

[0005] The present application provides a construction engineering company project matching degree analysis method based on project data mining, which is used to overcome the above-mentioned defects in the prior art.

[0006] In order to achieve the above-mentioned application purposes, the technical solutions adopted by the present application are as follows:

[0007] A construction engineering company project matching degree analysis method based on project data mining, comprising the following steps:

[0008] S1: Collecting a plurality of engineering project information corresponding to a plurality of construction engineering companies as a metadata set, classifying the metadata set, and preprocessing the metadata set;

[0009] S2: Establishing a project matching prediction model CDF, inputting the preprocessed metadata into the model and outputting a prediction result;

[0010] S3: Compare the prediction results of the CDF model with the true labels, and iteratively update the parameters of the model until the model converges;

[0011] S4: Input the real engineering project data into the model, predict the matching degree values of the engineering project and each construction engineering company, and provide the top-n construction engineering companies with the highest matching degree values.

[0012] Further, the step S1 specifically comprises:

[0013] S101: Obtain a plurality of historical engineering project information of a plurality of construction engineering companies by using a data collector;

[0014] S102: Identify the engineering project information and extract the engineering project feature items of each data label;

[0015] S103: Vectorize and encode the engineering project feature items;

[0016] S104: According to the vectorization coding result, filter out the feature vectors with dimension greater than n d β times sampling to obtain a sample matrix X s ;

[0017] S105: According to the sample matrix X s , perform dimensionality reduction processing on the data;

[0018] S106: According to the data after dimensionality reduction processing, construct a feature vector set (x0, x1, …, x m ).

[0019] Further, the dimensionality reduction calculation method in the step S105 is as follows:

[0020]

[0021]

[0022]

[0023]

[0024] Wherein, X s is a sample matrix, α is the dimension of the feature vector, β is the sampling times, S is a score matrix, the size of the matrix S is α*β, L is a load matrix, the size of the matrix L is β*β, ∑ is a summation function, δ is a count variable, δ = 1, 2, …, α, s δ is a score principal component, l δ is a load vector, is the transpose of the load vector, λ δ is an eigenvalue of the covariance matrix, and μ isδ corresponding feature vector, is the transpose of the sample matrix, C δ is the cumulative contribution rate, α1 is the number of principal components, α1≤α, if C δ >η, then the principal component construction can be performed, η is a constant threshold, X o is the principal component construction result.

[0025] Further, the step S2 specifically comprises:

[0026] S201: constructing a project matching prediction model CDF according to the feature vector set obtained after preprocessing;

[0027] S202: inputting the feature vector into the project matching prediction model CDF to generate a predictive matching degree value of the construction engineering project and the construction engineering company.

[0028] Further, the project matching prediction model CDF in the step S201 comprises: a feature vector embedding model, a convolutional neural network model, two feedforward neural network models, and a matching degree prediction model.

[0029] Further, the input end of the project matching prediction model CDF serves as the input end of the feature vector embedding model; the output end of the feature vector embedding model is connected with the input end of the convolutional neural network model; the output end of the convolutional neural network model is connected with the input end of the first feedforward neural network model; the output end of the first feedforward neural network model is connected with the input end of the second feedforward neural network model; the output end of the second feedforward neural network model is connected with the input end of the matching degree prediction model; the output end of the matching degree prediction model serves as the output end of the project matching prediction model CDF;

[0030] The feature vector embedding model comprises: a Relu activation function layer, a batch processing layer, and a full connection layer; the input end of the feature vector embedding model serves as the input end of the Relu activation function layer; the output end of the Relu activation function layer is connected with the input end of the batch processing layer; the output end of the batch processing layer is connected with the input end of the full connection layer; the output end of the full connection layer serves as the output end of the feature vector embedding model, and is connected with the convolutional layer of the convolutional neural network model;

[0031] The convolutional neural network model comprises: a convolutional layer, a pooling layer, and an output layer; the input end of the convolutional neural network model serves as the input end of the convolutional layer; the output end of the convolutional layer is connected with the input end of the pooling layer; the output end of the pooling layer is connected with the input end of the output layer; the output end of the output layer serves as the output end of the convolutional neural network model, and is connected with the full connection layer of the first feedforward neural network model;

[0032] The first feedforward neural network model comprises a full connection layer, a feedforward neural network layer and a Dropout layer; the input end of the first feedforward neural network model serves as the input end of the full connection layer; the output end of the full connection layer is connected with the input end of the feedforward neural network layer; the output end of the feedforward neural network layer is connected with the input end of the Dropout layer; the output end of the Dropout layer serves as the output end of the first feedforward neural network model and is connected with the full connection layer of the second feedforward neural network model;

[0033] The second feedforward neural network model comprises a full connection layer, a feedforward neural network layer and a Dropout layer; the input end of the second feedforward neural network model serves as the input end of the full connection layer, and the output end of the full connection layer is connected with the input end of the feedforward neural network layer; the output end of the feedforward neural network layer is connected with the input end of the Dropout layer; the output end of the Dropout layer serves as the output end of the second feedforward neural network model and is connected with the matching degree prediction model.

[0034] Further, the calculation formula of the building engineering company project matching degree generated in the step S202 is as follows:

[0035]

[0036] V2=Tanh(Q2*V1+k1)

[0037] W2=Tanh(θ2*Random(p)*V2+c2)

[0038] V3=Relu(Q3*W2+k2)

[0039] W3=Relu(θ3*Random(p)*V3+c3)

[0040] F sc =Sigmod(Q4*W3+k3)

[0041] wherein V1 is a convolution layer output feature vector, LeakyRelu(·) is an activation function, ∑ is a summation function, n is 1 / 2 times of the total number m of feature vectors, i is the serial number of the feature vector (i=1, 3, 5, …, 2n+1), Q v is a weight vector of a convolution layer filter, x i is the concatenation of the i-th embedded feature vector in the window, j is the serial number of the feature vector (j=0, 2, 4, …, 2n), P v is a weight vector of a convolution layer filter, x jLet V2 be the concatenation of the j-th embedded feature vector within the window, V3 be the output feature vector of the first feedforward neural network layer, Tanh(·) be the activation function, Q2 be the weight matrix of the first feedforward neural network layer, k1 be the bias term of the first feedforward neural network layer, W2 be the output feature vector of the first Dropout layer, θ2 be the adjustment parameter of the first Dropout layer, Random(·) be the probability vector generation function, c2 be the bias term of the first Dropout layer, V3 be the output feature vector of the second feedforward neural network layer, ReLU(·) be the activation function, Q3 be the weight matrix of the second feedforward neural network layer, k2 be the bias term of the second feedforward neural network layer, W3 be the output feature vector of the second Dropout layer, θ3 be the adjustment parameter of the second Dropout layer, c3 be the bias term of the second Dropout layer, and F be the output feature vector of the second Dropout layer. sc is the model's predicted matching degree value, Sigmod(·) is the activation function, Q4 is the weight matrix of the prediction layer, and k3 is the bias term of the prediction layer.

[0042] Furthermore, step S3 specifically includes:

[0043] S301: Construct a loss function based on the meta-dataset, feature vectors, and prediction results of the item matching prediction model;

[0044] S302: Calculate the corresponding parameter values ​​of the item matching prediction model based on the loss function;

[0045] S303: Backpropagate the gradients of the parameters in the item matching prediction model to update the parameter values ​​in the item matching prediction model; further, the loss function in step S301 is as follows:

[0046]

[0047] Where Loss is the loss function, k is the current item matching prediction task number (k = 1, 2, ..., m), m is the total number of feature vectors, ∑ is the summation function, and F sc F is used to predict the matching degree value for the model. re This represents the actual matching degree of the engineering project. are variable coefficients, and When the loss function value is less than ε (ε is a constant, a very small value) and the loss function value remains stable and does not decrease, the model converges and the model parameters have been iteratively updated.

[0048] Furthermore, step S4 specifically includes:

[0049] S401: input different engineering project data into the model, calculate the matching degree values of the engineering projects and respective construction companies, and arrange the construction companies in descending order according to the matching degree values;

[0050] S402: output the top n construction companies according to actual engineering needs.

[0051] The beneficial effects of the present application include:

[0052] (1) The present application is based on the idea of supervised learning, and first processes the obtained engineering project information to provide comprehensive, accurate and standardized engineering project feature metadata for the project matching prediction model CDF. In addition, the processed engineering project information is encoded, which can accurately represent the characteristics of the engineering project information, and further provide identifiable and usable input data for the project matching prediction model.

[0053] (2) The project matching prediction model proposed by the present application can automatically obtain information in the engineering project metadata and analyze and process it to mine the effective information hidden therein, classify the features according to the dimension of the extracted feature vector and effectively reduce the dimension of the input data, thereby reducing the model operation time, while preserving the prediction accuracy of the model, improving the processing efficiency of the model.

[0054] (3) The present application can effectively avoid the sparsity of the feature vector by converting the encoded engineering project information through the feature vector embedding model, prevent overfitting by using the Dropout layer to improve the generalization ability of the model, and further reasonably predict the project matching degree of the construction company.

[0055] (4) Compared with the traditional method of manually screening matching construction companies and projects, the present application automatically trains model parameters using deep learning methods without human intervention, and is completely based on a large amount of existing real construction engineering data, so the reliability of the prediction result is higher, and the efficiency of the construction work is greatly improved.

[0056] Other advantages, objects and features of the present application will be set forth in part in the following specification, and in part will become apparent to those skilled in the art from the examination of the following specification, or can be learned from practice of the present application. The objects and other advantages of the present application can be realized and obtained by the following description and the preceding claims. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings, in which:

[0058] Figure 1A project data mining based construction engineering company project matching degree analysis method flowchart of the present application is shown in the figure.

[0059] Figure 2 A real engineering project training set data schematic diagram is shown in the figure.

[0060] Figure 3 A sample matrix X of the embodiment of the present application is shown in the figure. s A schematic diagram is shown in the figure.

[0061] Figure 4 A project matching degree prediction model CDF working principle diagram is shown in the figure.

[0062] Figure 5 A real engineering project test set data schematic diagram is shown in the figure. DETAILED DESCRIPTION

[0063] The preferred embodiment of the present application will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiment is only for the purpose of illustrating the present application, and is not intended to limit the protection scope of the present application.

[0064] A project data mining based construction engineering company project matching degree analysis method of the present application is shown in the figure. Figure 1 The method comprises the following steps:

[0065] Step S1: Collect a plurality of engineering project information corresponding to a plurality of construction engineering companies as a metadata set, classify the metadata set, and pre-process the metadata set;

[0066] Step S2: Establish a project matching prediction model CDF, input the pre-processed metadata into the model and output the prediction result;

[0067] Step S3: Compare the prediction result of the CDF model with the real label, and iteratively update the parameters of the model until the model converges;

[0068] Step S4: Input the real engineering project data into the model, predict the matching degree value of the engineering project and each construction engineering company, and provide the top-n construction engineering companies with the highest matching degree value.

[0069] It should be noted that any process or method described in the flowchart of the present application or otherwise described herein can be understood as representing a module, a segment or a portion of code including one or more executable instructions for implementing a specific logical function or process, and the scope of the preferred embodiment of the present application includes additional implementations in which the functions can be performed in an order other than that shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0070] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical apparatus), a portable computer diskette (magnetic apparatus), a random-access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical apparatus), and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via the optical scanner of a device or other suitable device and then compiled, interpreted, or processed in a suitable manner, if necessary, and stored in a computer memory.

[0071] The specific steps of the above method will be further described below through a specific embodiment. In this embodiment, the data collector is a sub-function module of a real construction project query system, and the project data is as shown in the table. Figure 2

[0072] Step S1: Collect a plurality of project information corresponding to a plurality of construction companies as a metadata set, classify the metadata set, and preprocess the metadata set;

[0073] In this embodiment, step S1 specifically includes the following steps,

[0074] Step S101: Obtain a plurality of historical project information of a plurality of construction companies by using a data collector;

[0075] The project information in step S101 is as shown in the table. Figure 2 The format of the project information is: {company: A comp , construction building purpose: A purp , construction building property: A natu , building scale: A scal , building layers: A floo , building structure material: A mate , outdoor plastering area ratio: A​prpa , the proportion of outdoor door and window area: A prwa , the proportion of template area: A prfa , construction period: A dura , construction investment cost: A cost , average temperature during construction: A temp}, wherein A comp represents the construction company of the corresponding building project, A purp represents the use of the completed engineering building, A natu represents whether the building project is new, A scal represents the building scale of the engineering building, A floo represents the number of floors of the engineering building, A mate represents the material used in the building structure, A prpa represents the proportion of outdoor plastering area and building area of the engineering building, A prwa represents the proportion of outdoor door and window area and building area of the engineering building, A prfa represents the proportion of template area and building area of the engineering building, A dura represents the total time consumed from the beginning to the end of the project, A cost represents the total cost consumed by the construction, A temp represents the average temperature during construction.

[0076] Step S102: identify the engineering project information and extract the engineering project feature items of each data label;

[0077] The engineering project feature items in step S102 are: {company: A comp , construction building purpose: A purp , construction building property: A natu , building scale: A scal , building floor number: A floo , building structure material: A mate , the proportion of outdoor plastering area: A prpa , the proportion of outdoor door and window area: A prwa , the proportion of template area: A prfa , construction period: A dura , construction investment cost: A cost , average temperature during construction: A temp}.

[0078] Step S103: vectorization coding of engineering project feature items;

[0079] The vectorization coding method of engineering project information in step S103 is:

[0080] Apurp The value set of A is: {factory, greenhouse, farm, repair station, residence, apartment, villa, dormitory, office building, school, library, cultural palace, kindergarten, scientific research laboratory, hospital, commercial center, shop, theater, sports venue, hotel, station, airport, port, communication building, park, memorial hall, prison, police station, fire station, others}, A is encoded using a one-hot vector in the feature metadata, where the dimension of the one-hot vector is the length of the value set, and the value of a certain position in the one-hot vector is 1, then A is the value of the corresponding position in the value set; purp purp

[0081] A is defined in the feature metadata natu The value set of A is: {new, expansion, reconstruction}, A is encoded using a one-hot vector in the feature metadata, where the dimension of the one-hot vector is the length of the value set, and the value of a certain position in the one-hot vector is 1, then A is the value of the corresponding position in the value set; natu natu

[0082] A is defined in the feature metadata scal The value set of A is: {large-scale building, large-scale building}, A is encoded using a one-hot vector in the feature metadata, where the dimension of the one-hot vector is the length of the value set, and the value of a certain position in the one-hot vector is 1, then A is the value of the corresponding position in the value set; scal scal

[0083] A is defined in the feature metadata floo The value set of A is: {low-rise building, multi-story building, mid-rise building, high-rise building, super high-rise building}, A is encoded using a one-hot vector in the feature metadata, where the dimension of the one-hot vector is the length of the value set, and the value of a certain position in the one-hot vector is 1, then A is the value of the corresponding position in the value set; floo floo

[0084] A is defined in the feature metadata mate The value set of A is: {wood structure building, brick-wood structure building, brick-concrete structure building, reinforced concrete structure building, steel structure building, other structure building}, A is encoded using a one-hot vector in the feature metadata, where the dimension of the one-hot vector is the length of the value set, and the value of a certain position in the one-hot vector is 1, then A is the value of the corresponding position in the value set; mate mate

[0085] A is defined in the feature metadata prpa ​​​​​​​​​​The set of values ​​for A is: {proportion less than 0.4, proportion between 0.4 and 0.5, proportion between 0.5 and 0.6, proportion between 0.6 and 0.7, proportion between 0.7 and 0.8, proportion greater than 0.8}. One-hot vectors are used to analyze A in the feature metadata. prpa Encoding is performed where the dimension of the one-hot vector is the length of the set of values, and a value of 1 at a certain position in the one-hot vector represents A. prpa The value at the corresponding position in the set of values;

[0086] Define A in feature metadata prwa The set of values ​​for A is: {proportion less than 0.1, proportion between 0.1 and 0.15, proportion between 0.15 and 0.2, proportion between 0.2 and 0.25, proportion between 0.25 and 0.3, proportion greater than 0.3}. A one-hot vector is used to analyze the A value in the feature metadata. prwa Encoding is performed where the dimension of the one-hot vector is the length of the set of values, and a value of 1 at a certain position in the one-hot vector represents A. prwa The value at the corresponding position in the set of values;

[0087] Define A in feature metadata prfa The set of values ​​for A is: {proportion less than 2, proportion 2–2.5, proportion 2.5–3, proportion 3–3.5, proportion 3.5–4, proportion greater than 4}. One-hot vectors are used to analyze A in the feature metadata. prfa Encoding is performed where the dimension of the one-hot vector is the length of the set of values, and a value of 1 at a certain position in the one-hot vector represents A. prfa The value at the corresponding position in the set of values;

[0088] Define A in feature metadata dura The value set is: {less than 12 months, 12-18 months, 18-24 months, 24-30 months, 30-36 months, more than 36 months}. One-hot vectors are used to analyze the A value in the feature metadata. dura Encoding is performed where the dimension of the one-hot vector is the length of the set of values, and a value of 1 at a certain position in the one-hot vector represents A. dura The value at the corresponding position in the set of values;

[0089] Define A in feature metadata cost The set of values ​​is: {less than 2000 yuan / m 2 2000 yuan / m 2 ~3000 yuan / m 2 3000 yuan / m 2 ~4000 yuan / m 2 4000 yuan / m 2 ~5000 yuan / m 2 5000 yuan / m 2~6000 yuan / m 2 , 6000 yuan / m 2 ~7000 yuan / m 2 , 7000 yuan / m 2 The above}, the A cost in the feature metadata is encoded using a one-hot vector, where the dimension of the one-hot vector is the length of the value set, and the value of a certain position in the one-hot vector is 1, then the A cost is the value of the corresponding position in the value set;

[0090] The value set of the A temp in the feature metadata is defined as: {less than 8℃, 8℃-12℃, 12℃-15℃, 15℃-18℃, 18℃-21℃, greater than 21℃}, the A temp in the feature metadata is encoded using a one-hot vector, where the dimension of the one-hot vector is the length of the value set, and the value of a certain position in the one-hot vector is 1, then the A temp is the value of the corresponding position in the value set;

[0091] Step S104: According to the vectorization encoding result, the feature vectors with dimension greater than n d are screened out to obtain the sample matrix X s by β times sampling;

[0092] The value of n d in step S104 is set to 5, and the screened features are A purp , A mate , A prpa , A prwa , A prfa , A dura , A cost , A temp ; The sample matrix X s is shown in the schematic diagram as Figure 3 ;

[0093] Step S105: According to the sample matrix X s , the data is processed by dimension reduction;

[0094] The dimension reduction calculation method in step S105 is as follows:

[0095]

[0096]

[0097]

[0098]

[0099] Where X sis a sample matrix, a is a dimension of a feature vector, β is a sampling number, S is a score matrix, the size of the matrix S is a*β, L is a loading matrix, the size of the matrix L is β*β, ∑ is a summation function, δ is a counting variable, δ = 1, 2, …, a, s δ is a score principal component, l δ is a loading vector, is a transpose of a loading vector, λ δ is an eigenvalue of a covariance matrix, μ is λ δ corresponding to a feature vector, is a transpose of a sample matrix, C δ is a cumulative contribution rate, a1 is a number of principal components, a1 ≤ a, if C δ > η, then principal component construction can be performed, η is a constant threshold value, the value of η is set to 0.85, X o is a principal component construction result.

[0100] Step S106: constructing a feature vector set (x0, x1, …, x m ) according to the data processed by dimension reduction.

[0101] Step S2: establishing a project matching prediction model CDF, inputting the preprocessed metadata into the model and outputting a prediction result;

[0102] Step S2 specifically includes the following steps:

[0103] Step S201: constructing a project matching prediction model CDF according to the feature vector set (x0, x1, …, x m ) obtained after preprocessing;

[0104] The project matching prediction model CDF includes: a feature vector embedding model, a convolutional neural network model, two feedforward neural network models, and a matching degree prediction model, as shown in Figure 4 .

[0105] In this embodiment, the input end of the project matching prediction model CDF serves as the input end of the feature vector embedding model; the output end of the feature vector embedding model is connected with the input end of the convolutional neural network model; the output end of the convolutional neural network model is connected with the input end of the first feedforward neural network model; the output end of the first feedforward neural network model is connected with the input end of the second feedforward neural network model; the output end of the second feedforward neural network model is connected with the input end of the matching degree prediction model; and the output end of the matching degree prediction model serves as the output end of the project matching prediction model CDF.

[0106] In the embodiment, the feature vector embedding model comprises a Relu activation function layer, a batch layer and a full connection layer; an input end of the feature vector embedding model is an input end of the Relu activation function layer; an output end of the Relu activation function layer is connected with an input end of the batch layer; an output end of the batch layer is connected with an input end of the full connection layer; and an output end of the full connection layer is an output end of the feature vector embedding model and is connected with the convolution layer of the convolutional neural network model.

[0107] In the embodiment, the convolutional neural network model comprises a convolution layer, a pooling layer and an output layer; an input end of the convolutional neural network model is an input end of the convolution layer; an output end of the convolution layer is connected with an input end of the pooling layer; an output end of the pooling layer is connected with an input end of the output layer; and an output end of the output layer is an output end of the convolutional neural network model and is connected with the full connection layer of the first feedforward neural network model.

[0108] In the embodiment, the first feedforward neural network model comprises a full connection layer, a feedforward neural network layer and a Dropout layer; an input end of the first feedforward neural network model is an input end of the full connection layer; an output end of the full connection layer is connected with an input end of the feedforward neural network layer; an output end of the feedforward neural network layer is connected with an input end of the Dropout layer; and an output end of the Dropout layer is an output end of the first feedforward neural network model and is connected with the full connection layer of the second feedforward neural network model.

[0109] The second feedforward neural network model comprises a full connection layer, a feedforward neural network layer and a Dropout layer; an input end of the second feedforward neural network model is an input end of the full connection layer; an output end of the full connection layer is connected with an input end of the feedforward neural network layer; an output end of the feedforward neural network layer is connected with an input end of the Dropout layer; and an output end of the Dropout layer is an output end of the second feedforward neural network model and is connected with the matching degree prediction model.

[0110] Step S202: inputting the feature vector into the project matching prediction model CDF to generate a predicted matching degree value of the construction engineering project and the construction engineering company.

[0111] In the embodiment, the input engineering project information feature vector set (x0, x1, …, x m ), and the output is a predicted matching degree value, and the calculation method is as follows:

[0112]

[0113] V2=Tanh(Q2*V1+k1)

[0114] W2=Tanh(θ2*Random(p)*V2+c2)

[0115] V3 = Relu(Q3 * W2 + k2)

[0116] W3 = Relu(θ3 * Random(p) * V3 + c3)

[0117] F sc = Sigmod(Q4 * W3 + k3)

[0118] wherein, V1 is the output feature vector of the convolution layer, LeakyRelu(·) is an activation function, ∑ is a summation function, n is 1 / 2 of the total number of feature vectors m, i is the serial number of the feature vector (i = 1, 3, 5, …, 2n + 1), Q v is the weight vector of the convolution layer filter, x i is the concatenation of the i-th embedded feature vector in the window, j is the serial number of the feature vector (j = 0, 2, 4, …, 2n), P v is the weight vector of the convolution layer filter, x j is the concatenation of the j-th embedded feature vector in the window, V2 is the output feature vector of the first feedforward neural network layer, Tanh(·) is an activation function, Q2 is the weight matrix of the first feedforward neural network layer, k1 is the bias term of the first feedforward neural network layer, W2 is the output feature vector of the first Dropout layer, θ2 is the adjustment parameter of the first Dropout layer, Random(·) is a probability vector generation function, c2 is the bias term of the first Dropout layer, V3 is the output feature vector of the second feedforward neural network layer, Relu(·) is an activation function, Q3 is the weight matrix of the second feedforward neural network layer, k2 is the bias term of the second feedforward neural network layer, W3 is the output feature vector of the second Dropout layer, θ3 is the adjustment parameter of the second Dropout layer, c3 is the bias term of the second Dropout layer, F sc is the model prediction matching degree value, Sigmod(·) is an activation function, Q4 is the weight matrix of the prediction layer, and k3 is the bias term of the prediction layer.

[0119] Step S3: comparing the prediction result of the CDF model with the true label to update the parameters of the model;

[0120] Step S3 specifically includes the following steps:

[0121] Step S301: constructing a loss function according to the metadata set, the feature vector, and the prediction result of the item matching prediction model; the loss function in the item matching prediction model is as follows:

[0122]

[0123] Wherein, Loss is a loss function, k is a current project matching prediction task sequence number (k = 1, 2, …, m), m is a total number of feature vectors, ∑ is a summation function, F sc is a model predicted matching degree value, F re is a real project matching degree value, is a variable coefficient, and When the Loss loss function value is less than ε (the value of ε is set to 0.001) and the Loss loss function value is stable and does not decrease, the model converges, and the model parameter iteration update is completed.

[0124] Step S302: According to the loss function, the corresponding parameters of the project matching prediction model are calculated.

[0125] Step S303: The parameter gradient in the project matching prediction model is back propagated to update the parameters in the project matching prediction model.

[0126] Step S4: The real engineering project data is input into the model, the matching degree values of the engineering project and each construction engineering company are predicted, and the top 5 construction engineering companies with the highest matching degree values are provided.

[0127] In this embodiment, the real engineering project data is as shown in Figure 5 The real engineering project data is input into the trained project matching prediction model, the matching degree values of the engineering project and each engineering project company are output, and the top 5 construction engineering companies with the highest matching degree values are provided for selection, wherein the top 5 construction engineering companies with the highest matching degree values and the corresponding matching degree values F sc are: {Sichuan Group Co., Ltd.: 93.26, Chengdu Construction Engineering Co., Ltd.: 89.43, Sichuan Construction Engineering Co., Ltd.: 82.34, Sichuan Construction Engineering Co., Ltd. 2: 81.35, Group Co., Ltd.: 80.96}.

[0128] In this embodiment, all the engineering project information collected by the data collector is divided into a training set and a test set, Figure 2 The engineering project information shown in the figure is derived from the training set, the project matching prediction model is trained multiple times based on the training set, the trained project matching prediction model can calculate the matching degree of each construction engineering company according to the input construction engineering project data in the test set, automatically complete project recommendation, thereby improving the efficiency of recommendation and reducing the prediction inaccuracy caused by the incompleteness and incompleteness of the project information considered by manual recommendation.

[0129] It should be understood that each part of the present application can be realized by hardware, software, firmware or their combination. In the above-mentioned embodiments, a plurality of steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized by hardware, and as in another embodiment, it can be realized by any one or their combination of the following technologies known in the art: discrete logic circuit with logic gate circuit for implementing logic function on data signal, application specific integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) and the like.

[0130] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment method can be completed by a program instructing the relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0131] In addition, each functional unit in each embodiment of the present application can be integrated in one processing module, or each unit can exist physically alone, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0132] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0133] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the purpose and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A project data mining-based construction engineering company project matching degree analysis method, characterized by: The method comprises the following steps: S1: Collecting a plurality of project information corresponding to a plurality of construction companies as a metadata set, classifying the metadata set, and preprocessing the metadata set; S2: Establishing a project matching prediction model CDF, inputting the preprocessed metadata into the model, and outputting a prediction result; the project matching prediction model CDF comprises a feature vector embedding model, a convolutional neural network model, two feedforward neural network models, and a matching degree prediction model; the input end of the project matching prediction model CDF serves as the input end of the feature vector embedding model; the output end of the feature vector embedding model is connected with the input end of the convolutional neural network model; the output end of the convolutional neural network model is connected with the input end of the first feedforward neural network model; the output end of the first feedforward neural network model is connected with the input end of the second feedforward neural network model; the output end of the second feedforward neural network model is connected with the input end of the matching degree prediction model; and the output end of the matching degree prediction model serves as the output end of the project matching prediction model CDF; The feature vector embedding model comprises a Relu activation function layer, a batch processing layer, and a full connection layer; the input end of the feature vector embedding model serves as the input end of the Relu activation function layer; the output end of the Relu activation function layer is connected with the input end of the batch processing layer; the output end of the batch processing layer is connected with the input end of the full connection layer; and the output end of the full connection layer serves as the output end of the feature vector embedding model and is connected with the convolutional layer of the convolutional neural network model; The convolutional neural network model comprises a convolutional layer, a pooling layer, and an output layer; the input end of the convolutional neural network model serves as the input end of the convolutional layer; the output end of the convolutional layer is connected with the input end of the pooling layer; the output end of the pooling layer is connected with the input end of the output layer; and the output end of the output layer serves as the output end of the convolutional neural network model and is connected with the full connection layer of the first feedforward neural network model; The first feedforward neural network model comprises a full connection layer, a feedforward neural network layer, and a Dropout layer; the input end of the first feedforward neural network model serves as the input end of the full connection layer; the output end of the full connection layer is connected with the input end of the feedforward neural network layer; the output end of the feedforward neural network layer is connected with the input end of the Dropout layer; and the output end of the Dropout layer serves as the output end of the first feedforward neural network model and is connected with the full connection layer of the second feedforward neural network model; The second feedforward neural network model comprises a full connection layer, a feedforward neural network layer, and a Dropout layer; the input end of the second feedforward neural network model serves as the input end of the full connection layer; the output end of the full connection layer is connected with the input end of the feedforward neural network layer; the output end of the feedforward neural network layer is connected with the input end of the Dropout layer; and the output end of the Dropout layer serves as the output end of the second feedforward neural network model and is connected with the matching degree prediction model; The calculation formula of the project matching degree of the construction company is as follows: ​ wherein, is an output feature vector of the convolutional layer, is an activation function, , , , , is a corresponding convolutional layer filter weight vector, , , , is a corresponding convolutional layer filter weight vector, concatenation of the embedding feature vectors, is an output feature vector of the first feedforward neural network layer, is an activation function, is a weight matrix of the first feedforward neural network layer, is a bias term of the first feedforward neural network layer, is an output feature vector of the first Dropout layer, is a tuning parameter of the first Dropout layer, is a probability vector generating function, is a bias term of the first Dropout layer, is an output feature vector of the second feedforward neural network layer, is an activation function, is a weight matrix of the second feedforward neural network layer, is a bias term of the second feedforward neural network layer, is an output feature vector of the second Dropout layer, is a tuning parameter of the second Dropout layer, is a bias term of the second Dropout layer, is a model predicted matching degree value, is an activation function, is a weight matrix of the prediction layer, is a bias term of the prediction layer. S3: Compare the prediction results of the CDF model with the true labels, and iteratively update the parameters of the model until the model converges; S4: input the real engineering project data into the model, predict the matching degree value of the engineering project and each construction engineering company, and provide the construction engineering company with the highest matching degree value. one construction engineering company with the highest matching degree value.

2. The project data mining based construction company project matching degree analysis method according to claim 1, characterized in that: The step S1 specifically comprises: S101: Obtain a plurality of historical engineering project information of a plurality of construction engineering companies by using a data collector; S102: Identify the engineering project information, and extract the engineering project feature items of each data label; S103: Vectorize and encode the engineering project feature items; S104: according to the vector quantization coding result, screening out the feature vectors with dimensions greater than to perform sub-sampling to obtain a sample matrix ; S105: according to the sample matrix dimension reduction is performed on the data; S106: Construct a feature vector set according to the data after dimension reduction processing .

3. The project data mining based construction company project matching degree analysis method according to claim 2, characterized in that: The dimension reduction calculation method in the step S105 is as follows: wherein, is a sample matrix, , , , , is a score principal component, is a loading vector, is a transpose of the loading vector, is an eigenvalue of the covariance matrix, , is a transpose of the sample matrix, is a cumulative contribution rate, then principal component construction can be performed, is a constant threshold value, is a principal component construction result.

4. The project data mining based construction company project matching degree analysis method according to claim 1, characterized in that: The step S2 specifically comprises: S201: Construct a project matching prediction model CDF according to the obtained feature vector set after preprocessing; S202: Input the feature vector into the project matching prediction model CDF to generate the predicted matching degree values of the construction engineering projects and the construction engineering companies.

5. The project data mining based construction company project matching degree analysis method according to claim 1, characterized in that: The step S3 specifically comprises: S301: Construct a loss function according to the metadata set, the feature vector, and the prediction results of the project matching prediction model; S302: Calculate the corresponding parameter values of the project matching prediction model according to the loss function; S303: Update the parameter values in the project matching prediction model by reversely propagating the parameter gradient in the project matching prediction model.

6. The project data mining based construction company project matching degree analysis method according to claim 5, characterized in that: The loss function in the step S301 is as follows: wherein, is a loss function, , , is a model predicted matching degree value, is an engineering project real matching degree value, ; The loss function value is stable and does not decrease, the model converges, and the model parameter iteration update is completed.

7. The project data mining based construction company project matching degree analysis method according to claim 1, characterized in that: The step S4 specifically comprises: S401: Input different engineering project data into the model, calculate the matching degree values of the engineering projects and the respective construction engineering companies, and arrange the matching degree values in descending order according to the size of the matching degree values; S402: According to the real engineering needs, output ranking a construction company.

Images