Project Risk Level Assessment Method Based on Approximate Personalized Transmission

Abstract

The present application discloses a method for evaluating the risk level of a project based on approximate personalized transmission, which relates to the technical field of engineering project risk assessment. The method includes: obtaining a project data set of a power engineering construction project; determining data division indicators corresponding to the risk level assessment structure of each project data in the project data set, and determining the data division indicators as the data characteristics of the project data; inputting each project data and the data characteristics into a pre-trained risk level assessment model to obtain a risk score corresponding to the power engineering construction project output by the risk level assessment model; and determining the project risk level corresponding to the risk score based on the pre-established correspondence between the risk score and the project risk level. The present application is used to solve the problems of poor accuracy and low evaluation efficiency in the existing technology during the evaluation of the project risk level, and to improve the accuracy and evaluation efficiency of the project risk level evaluation.

Description

Technical Field

[0001] The present application relates to the technical field of engineering project risk assessment, and in particular to a project risk level assessment method based on approximate personalized propagation. Background Art

[0002] The risk management of power engineering projects is very important in project management. At present, most of the risk level assessments of power engineering projects are subjectively judged by infrastructure engineering personnel based on past experience. Since there are problems such as difficult identification of key information, few factors considered, and insufficient reflection of details during the project construction process, the accuracy of the risk level evaluated manually is relatively low, and the efficiency of the manual evaluation method is also low.

[0003] At present, some risk level assessments of power engineering projects also adopt machine learning algorithms, use convolutional neural networks to iteratively train models, and then evaluate the risk level through the obtained models. However, this method has a strong dependence on model hyperparameters, which may lead to poor accuracy of the predicted risk level, and the calculation process is relatively complex, with high requirements for computing resources, resulting in low evaluation efficiency. Summary of the Invention

[0004] In view of the above problems and technical requirements, the applicant has proposed a project risk level assessment method based on approximate personalized propagation to solve the problems of poor accuracy and low evaluation efficiency in the existing technology for project risk level assessment, and to improve the accuracy and evaluation efficiency of project risk level assessment.

[0005] An embodiment of the present application provides a project risk level assessment method based on approximate personalized propagation.

[0006] Obtain a project data set of a power engineering construction project;

[0007] Determine the data division indicators corresponding to the risk level assessment structures of each project data in the project data set, and determine the data division indicators as the data features of the project data, where the risk level assessment structure is created based on the project risk requirements of the power engineering construction project;

[0008] Input each project data and the data features into a pre-trained risk level assessment model to obtain the risk score corresponding to the power engineering construction project output by the risk level assessment model, where the risk level assessment model is a graph neural network model, and approximate personalized propagation values are transmitted among the nodes of the graph neural network model through an approximate personalized propagation mechanism, and the risk level assessment model is trained based on project data samples, data feature samples, risk score samples, and approximate personalized propagation values.

[0009] Determine the project risk level corresponding to the risk score based on the pre-established correspondence between the risk score and the project risk level.

[0010] According to the project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application, the risk level assessment model includes: at least two nodes;

[0011] The training process of training the risk level assessment model based on the approximate personalized propagation mechanism includes:

[0012] Obtain at least two training data sample sets, where each training data sample set includes: a project data sample corresponding to a power engineering construction project, a data feature sample corresponding to the project data sample, and a risk score sample corresponding to the power engineering construction project, and one node corresponds to one training data sample set;

[0013] Input the training data samples into the risk level assessment model, and perform the following data processing operations on each node through the risk level assessment model:

[0014] Based on the project data sample corresponding to the current node, the data feature sample, and the previous approximate personalized propagation value sample passed from the previous node, obtain the current predicted risk score; based on the project data sample, the data feature sample, and the current predicted risk score, obtain the current approximate personalized propagation value sample, and input the current approximate personalized propagation value sample into the next node;

[0015] Optimize the model parameters of the risk level assessment model based on the predicted risk scores corresponding to each node and the risk score samples corresponding to each node, until the number of node jumps reaches the preset number of times, and determine that the training of the risk level assessment model is completed.

[0016] According to the project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application, the data processing operations performed on each node further include:

[0017] When the current node is the first node, based on the project data sample corresponding to the first node and the data feature sample, obtain the current predicted risk score; based on the project data sample, the data feature sample, and the current predicted risk score, obtain the current approximate personalized propagation value sample, and input the current approximate personalized propagation value sample into the next node.

[0018] According to the project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application, the data processing operations performed on each node further include:

[0019] When the current node is the last node, based on the project data sample corresponding to the last node, the data feature sample, and the previous approximate personalized transfer value sample passed from the previous node, the current predicted risk score is obtained.

[0020] According to the project risk level assessment method based on approximate personalized transfer provided by the embodiments of the present application, the determination process of the node jump count π(i x ) includes:

[0021] Perform vectorization processing on the project data sample and the data feature sample to obtain the node feature matrix ix;

[0022] Input the node feature matrix into the jump count calculation formula to obtain the jump count output by the jump count calculation formula;

[0023] Among them, the jump count calculation formula includes:

[0024]

[0025] Among them, π(i x ) represents the jump count, α represents the jump probability of moving from the current node to the next node, represents a preset matrix, and i x represents the node feature matrix.

[0026] According to the project risk level assessment method based on approximate personalized transfer provided by the embodiments of the present application, obtaining the current approximate personalized transfer value sample based on the project data sample, the data feature sample, and the current predicted risk score includes:

[0027] Input the project data sample, the data feature sample, and the current predicted risk score into a pre-created personalized transfer value calculation formula to obtain the personalized transfer value sample output by the personalized transfer value calculation formula;

[0028] Perform approximation processing on the personalized transfer value sample to obtain the approximate personalized transfer value sample;

[0029] Among them, the personalized transfer value calculation formula includes:

[0030]

[0031] Among them, Z represents the personalized transfer value sample, softmax represents the normalization exponential function, α represents the jump probability of moving from the current node to the next node, represents a preset matrix, H represents the current predicted risk score, and i x represents the node feature matrix.

[0032] According to the project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application, the process of approximating the personalized propagation value samples to obtain approximate personalized propagation value samples includes:

[0033] Obtain the initial value of the approximate personalized propagation value samples;

[0034] Based on the node jumps, optimize the model parameters of the graph neural network model until the number of node jumps reaches a preset number;

[0035] Based on the approximate personalized propagation value sample calculation formula obtained for the current node in each iteration, obtain the approximate personalized propagation value samples through the approximate personalized propagation value sample calculation formula;

[0036] Among them, the approximate personalized propagation value sample calculation formula includes:

[0037]

[0038] Among them, Z ′ (k) represents the approximate personalized propagation value sample corresponding to the current node, and Z ′ (k-1) represents the approximate personalized propagation value sample corresponding to the previous node.

[0039] According to the project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application, optimizing the model parameters of the risk level assessment model includes:

[0040] Optimizing the parameter weights corresponding to the model parameters includes:

[0041] Using a preset weight calculation formula, calculate the parameter weights corresponding to the model parameters in the current iteration process;

[0042] Among them, the weight calculation formula includes:

[0043]

[0044] Among them, Q t represents the current parameter weight corresponding to the current iteration process, Q t-1 represents the previous parameter weight corresponding to the previous iteration process, γ represents the parameter multiplier, β represents the learning rate, represents the first-order momentum of the t gradient, represents the second-order momentum of the t gradient, ∈ represents a constant, and λ represents the regularization coefficient.

[0045] According to the project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application, the process of determining the risk score samples includes:

[0046] Calculate the risk score for the project data sample based on a preset risk score level standard to obtain the calculation result;

[0047] Convert the calculation result into a corresponding risk coding sample, and use the risk coding sample as the risk score sample.

[0048] According to the project risk level assessment method based on approximate personalized transmission provided by the embodiments of the present application, determining that the risk level assessment model is trained and completed includes:

[0049] Obtain at least two test data sample sets, where each test data sample set includes: test project data, test data features corresponding to the test data sample;

[0050] Perform the following test process on each test data sample set:

[0051] Input the test project data and the test data sample into the risk level assessment model, and the test risk score output by the risk level assessment model;

[0052] Determine the accuracy rate of the test risk scores corresponding to all the test data sample sets. When it is determined that the accuracy rate is greater than the preset accuracy rate, determine that the risk level assessment model is trained and completed.

[0053] The project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application obtains a project data set of a power engineering construction project; determines data division indicators corresponding to each project data in the project data set for a risk level assessment structure, and determines the data division indicators as data characteristics of the project data. The risk level assessment structure is created based on the project requirements of the power engineering construction project. It can be seen that the present application provides an effective data basis for subsequent project risk level assessment through project data and data characteristics related to the project risk requirements of the power engineering construction project. Furthermore, each project data and data characteristics are input into a pre-trained risk level assessment model to obtain a risk score corresponding to the power engineering construction project output by the risk level assessment model. The risk level assessment model is a graph neural network model, and approximate personalized propagation values are transmitted among the nodes of the graph neural network model through an approximate personalized propagation mechanism. The risk level assessment model is trained based on project data samples, data characteristic samples, risk score samples, and approximate personalized propagation values. Based on the pre-established correspondence between risk scores and project risk levels, the project risk level corresponding to the risk score is determined. The present application uses a graph neural network model trained with a personalized propagation mechanism, which can not only mine the key information carried by the project data itself but also mine the association information between nodes, ensuring the accuracy of the output result of the trained graph neural network model. Moreover, compared with the convolutional neural network in the prior art that can only process structured data, the graph neural network model can process unstructured data, reducing the workload of data processing and improving the training speed of the graph neural network model. The trained graph neural network is used to predict the risk score, and finally the project risk level of the power engineering construction project is obtained, achieving the purpose of improving the accuracy and evaluation efficiency of project risk level assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for description in the embodiments or the prior art. Obviously, the following described drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is a schematic flowchart of the project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application;

[0056] Figure 2 It is a schematic structural diagram of the project risk level assessment device based on approximate personalized propagation provided by the embodiments of the present application;

[0057] Figure 3It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present invention.

[0059] An embodiment of the present application provides a method for evaluating the project risk level based on approximate personalized propagation. This method can be applied to intelligent terminals and can also be applied to servers. Some other descriptions in the embodiments of the present application are for illustrative purposes and are not used to limit the protection scope of the present application, and will not be described one by one hereinafter. The specific implementation of this method is as Figure 1 shown below:

[0060] Step 101: Obtain the project data set of the power engineering construction project.

[0061] Step 102: Determine the data division indicators corresponding to the risk level evaluation structure of each project data in the project data set, and determine the data division indicators as the data characteristics of the project data.

[0062] Among them, the risk level evaluation structure is created based on the project risk requirements of the power engineering construction project.

[0063] Step 103: Input each project data and data characteristics into a pre-trained risk level evaluation model to obtain the risk score corresponding to the power engineering construction project output by the risk level evaluation model.

[0064] Among them, the risk level evaluation model is a graph neural network model, and the approximate personalized propagation value is transmitted among the nodes of the graph neural network model through the approximate personalized propagation mechanism. The risk level evaluation model is trained based on project data samples, data characteristic samples, risk score samples, and approximate personalized propagation values.

[0065] Step 104: Determine the project risk level corresponding to the risk score based on the pre-created correspondence between the risk score and the project risk level.

[0066] Among them, the risk score is transformed to obtain the corresponding risk code, and based on the pre-created correspondence between the risk code and the project risk level, the project risk level is obtained.

[0067] The project risk level assessment method based on approximate personalized propagation provided by the embodiments of the present application obtains a project data set of a power engineering construction project; determines data division indicators corresponding to each project data in the project data set for a risk level assessment structure, and determines the data division indicators as data characteristics of the project data. The risk level assessment structure is created based on the project requirements of the power engineering construction project. It can be seen that the present application provides an effective data basis for subsequent project risk level assessment through project data and data characteristics related to the project risk requirements of the power engineering construction project. Furthermore, each project data and data characteristics are input into a pre-trained risk level assessment model to obtain the risk score corresponding to the power engineering construction project output by the risk level assessment model. The risk level assessment model is a graph neural network model, and approximate personalized propagation values are transmitted among the nodes of the graph neural network model through an approximate personalized propagation mechanism. The risk level assessment model is trained based on project data samples, data characteristic samples, risk score samples, and approximate personalized propagation values. Based on the pre-established correspondence between risk scores and project risk levels, the project risk level corresponding to the risk score is determined. The present application uses a graph neural network model trained by a personalized propagation mechanism, which can not only mine the key information carried by the project data itself, but also mine the association information between nodes, ensuring the accuracy of the output result of the trained graph neural network model. Moreover, compared with the convolutional neural network in the prior art that can only process structured data, the graph neural network model can process unstructured data, reducing the workload of data processing and improving the training speed of the graph neural network model. The trained graph neural network is used to predict the risk score, and finally the project risk level of the power engineering construction project is obtained, achieving the purpose of improving the accuracy and efficiency of project risk level assessment.

[0068] Specifically, the project data is deconstructed through the risk level assessment structure. The risk level assessment structure is obtained through discretization processing and weight addition processing of the level assessment structure.

[0069] Among them, the risk level assessment structure is set as a three-level assessment structure, corresponding to three-level division indicators, including: primary indicators, secondary indicators, and tertiary indicators. Among them, the three-level assessment structure is a subordinate structure.

[0070] Among them, the primary indicators include: five indicators of personnel factors, mechanical equipment factors, material factors, operation method factors, and environmental factors corresponding to the power engineering construction project.

[0071] The secondary indicators include: 12 indicators of the supervision unit, construction unit, construction unit, construction machinery and equipment, safety system and tools, construction materials, document materials, operation / technical methods, management methods, operation environment, natural environment, and social environment corresponding to the power engineering construction project.

[0072] The third-level indicators include: safety inspection level, quality inspection level, review execution intensity, drawings, timeliness of fund delivery, requirement changes during construction, basic quality status of operating personnel, number of operating personnel, age status of operating personnel, training time of operating personnel, basic quality status of team leaders, qualifications of team leaders, equipment maintenance situation, equipment load operation situation (equipment performance), inspection of equipment safety protection devices, service life of equipment, equipment transportation, site and time involved in assembly, system stability, system accuracy, spot-check situation of tools and protective equipment, spot-check quality of materials, material replenishment and transportation time, material storage status and applicability, material hazard level, project-related documents, formalities information, time for qualification approval, document sorting and backup situation, construction technology level, number of operating personnel (single / multiple, etc.), nature of operation (general / temporary / emergency repair, etc.), overtime construction situation, frequency of accident hazard rectification and investigation, project quality, emergency plan and drill preparation, improvement of rules and regulations and safety inspection, salary reward and punishment system, water and electricity supply situation, construction site layout (building location, material storage location, etc.), operation section (high altitude, crossing railway, etc.), operation period (day / night), geological and hydrological conditions, climate, epidemic and natural disasters, temperature, laws, policies, opinions of residents and relevant departments, a total of 44 indicators.

[0073] Specifically, the weights of each data division indicator are obtained through a judgment matrix, as follows:

[0074] For example, the judgment matrix is

[0075] Among them, ABC represents the data division indicators, and the values of x, y, and z range from (1, 3, 5, 7, 9), corresponding to the importance values of each data division indicator. For example, 1 means less important, 3 means slightly important, 5 means significantly important, 7 means strongly important, and 9 means extremely important.

[0076] The subscripts of x, y, and z represent each data division indicator. The above judgment matrix is for illustrative purposes only, that is, it is illustrated with three data division indicators ABC. The specific data division indicators are the specific indicators of the above-listed first-level indicators, second-level indicators, and third-level indicators.

[0077] Of course, users can also set the weights of each data division indicator manually based on the actual situation.

[0078] In a specific embodiment, the risk level assessment model includes: at least two nodes. The training process of training the risk level assessment model based on the approximate personalized transfer mechanism includes:

[0079] Obtain at least two sets of training data samples; input the training data samples into the risk level evaluation model, and perform the following data processing operations on each node through the risk level evaluation model: obtain the current predicted risk score based on the project data sample, data feature sample corresponding to the current node, and the previous approximate personalized transfer value sample passed from the previous node; obtain the current approximate personalized transfer value sample based on the project data sample, data feature sample, and the current predicted risk score, and input the current approximate personalized transfer value sample into the next node; optimize the model parameters of the risk level evaluation model based on the predicted risk scores corresponding to each node and the risk score samples corresponding to each node, until the number of node jumps reaches the preset number, and then determine that the training of the risk level evaluation model is completed.

[0080] Among them, each set of training data samples includes: the project data sample corresponding to a power engineering construction project, the data feature sample corresponding to the project data sample, and the risk score sample corresponding to the power engineering construction project. One node corresponds to one set of training data samples.

[0081] Specifically, establish the association relationship between nodes based on the project categories and affiliated units of each power engineering construction project.

[0082] Among them, the project categories include: thermal power engineering projects, hydropower engineering projects, wind power engineering projects, power supply line engineering projects, photovoltaic projects, nuclear power projects, and power supply bureau facility projects.

[0083] Input the samples in multiple sets of training data samples configured with the association relationship between nodes into the risk level evaluation model, perform data processing operations on the samples in each node through the risk level evaluation model, and compare the predicted risk scores and risk score samples of each node. Optimize the model parameters of the risk level evaluation model based on the comparison results until the number of node jumps reaches the preset number, and then determine that the training of the risk level evaluation model is completed.

[0084] Among them, the association relationship between nodes is also used as the data feature sample of the samples in the associated nodes.

[0085] Among them, the samples in multiple sets of training data samples configured with the association relationship between nodes are input into the risk level evaluation model in the form of a graph structure.

[0086] In a specific embodiment, the data processing operations performed on each node further include:

[0087] When the current node is the first node, based on the project data sample and data feature sample corresponding to the first node, obtain the current predicted risk score; based on the project data sample, data feature sample and current predicted risk score, obtain the current approximate personalized transfer value sample, and input the current approximate personalized transfer value sample into the next node.

[0088] Specifically, after inputting the samples in multiple training data sample sets configured with the association relationships between nodes into the risk level assessment model, randomly determine a node and use this node as the first node. When predicting the risk score of the sample corresponding to the first node, since there is no previous approximate personalized transfer value sample passed from the previous node, obtain the current predicted risk score based on the project data sample and data feature sample.

[0089] Among them, the data feature sample includes: each data division index, the node markers of each node having an association relationship with the first node, and each association relationship. The data feature samples corresponding to all nodes include the above content.

[0090] In a specific embodiment, the data processing operations performed on each node further include:

[0091] When the current node is the last node, based on the project data sample, data feature sample corresponding to the last node, and the previous approximate personalized transfer value sample passed from the previous node, obtain the current predicted risk score.

[0092] Specifically, for the last node, since there is no further node transfer, therefore, the last node does not need to calculate the approximate personalized transfer value sample anymore, and only needs to obtain the corresponding predicted risk score.

[0093] In a specific embodiment, the determination process of the node jump count includes:

[0094] Perform vectorization processing on the project data sample and data feature sample to obtain a node feature matrix; input the node feature matrix into the jump count calculation formula to obtain the jump count output by the jump count calculation formula.

[0095] Among them, the jump count calculation formula is shown in formula (1):

[0096]

[0097] Among them, π(i x ) represents the node jump count, α represents the jump probability of moving from the current node to the next node, represents a preset matrix, and i x represents the node feature matrix.

[0098] Among them, the preset matrix is obtained by performing an inverse operation on the matrix value obtained after normalizing the preset self-loop edge adjacency matrix and then performing an inverse operation with the preset degree matrix.

[0099] In a specific embodiment, risk score samples are determined in advance, and the process of determining the risk score samples includes:

[0100] The risk score samples are divided into five risk score levels, for example, as shown in Table 1:

[0101]

[0102] Table 1 Risk Score Level Division Table

[0103] Specifically, based on the risk score level standard, the risk scores of the project data of the power engineering construction project are calculated, and the risk scores are converted into corresponding risk coding samples to obtain risk score samples.

[0104] Specifically, the risk score of the power engineering construction project is obtained by multiplying the weights corresponding to the data division indicators of each project data by the scores corresponding to the data division indicators and then performing a summation process.

[0105] In a specific embodiment, the specific implementation of obtaining the current approximate personalized transfer value sample includes:

[0106] The project data sample, the data feature sample, and the current predicted risk score are input into the personalized transfer value calculation formula to obtain the personalized transfer value sample output by the personalized transfer value calculation formula; the personalized transfer value sample is approximated to obtain the approximate personalized transfer value sample.

[0107] Among them, the personalized transfer value calculation formula is shown in Formula (2):

[0108]

[0109] Among them, Z represents the personalized transfer value sample, softmax represents the normalized exponential function, α represents the jump probability of walking from the current node to the next node, represents the preset matrix, H represents the current predicted risk score, i x represents the node feature matrix.

[0110] Among them, the node feature matrix is obtained by vectorizing the project data sample and the data feature sample.

[0111] Specifically, the specific implementation of approximating the personalized transfer value sample includes:

[0112] The personalized transfer value is initialized, specifically referring to Formula (3):

[0113] Z ′ (0) = H = f θ (X)……………………………(3)

[0114] Among them, Z ′ (0) represents the initial value of an approximate personalized transfer value sample, and f θ (X) represents a graph neural network model with model parameters θ.

[0115] Through node jumps, the graph neural network model is optimized. After k iterations, the (k + 1)-th approximate personalized transfer value sample is obtained through formula (4):

[0116]

[0117] Among them, Z' (k+1) represents the (k + 1)-th approximate personalized transfer value sample.

[0118] Normalize the last two iterations to obtain the calculation formula for the approximate personalized transfer value sample of the current node. See formula (5):

[0119]

[0120] Among them, Z ′ (k) represents the approximate personalized transfer value sample corresponding to the current node, and Z ′ (k-1) represents the approximate personalized transfer value sample corresponding to the previous node.

[0121] Specifically, the approximate personalized transfer value sample corresponding to the previous node, the jump probability, and the prediction score sample can also be directly input into formula (5) to obtain the approximate personalized transfer value sample corresponding to the current node.

[0122] Specifically, for the first node, the approximate personalized transfer value sample corresponding to the previous node is set to zero, and for the last node, there is no need to calculate the approximate personalized transfer value sample.

[0123] In a specific embodiment, the model parameters of the optimization risk level assessment model include: optimizing the parameter weights corresponding to the model parameters.

[0124] The specific optimization method of the parameter weights is to calculate the parameter weights corresponding to the model parameters in the current iteration process using a preset weight calculation formula.

[0125] Among them, the weight calculation formula is shown in formula (6):

[0126]

[0127] Among them, Q t represents the current parameter weight corresponding to the current iteration process, and Q t-1 represents the previous parameter weight corresponding to the previous iteration process, γ represents the parameter multiplier, β represents the learning rate, represents the first-order momentum of the t-gradient, represents the second-order momentum of the t-gradient, ∈ represents a constant, and λ represents the regularization coefficient.

[0128] In a specific embodiment, the specific implementation of determining that the risk level evaluation model training is completed includes:

[0129] Obtain at least two sets of test data samples; perform the following test process on each set of test data samples: input the test item data and the test data samples into the risk level evaluation model, and the risk level evaluation model outputs the test risk scores; determine the accuracy rate of the test risk scores corresponding to all sets of test data samples, and when it is determined that the accuracy rate is greater than the preset accuracy rate, determine that the risk level evaluation model training is completed.

[0130] Among them, each set of test data samples includes: test item data, test data features corresponding to the test data samples.

[0131] This application decouples the process of feature transformation and data transmission, making the depth of the graph neural network model completely independent of the propagation algorithm, ensuring that nodes can obtain distance information farther away without over-parameterization. Moreover, in the process of obtaining graph structure data, the relationship between nodes themselves and between nodes can be considered, overcoming the problem that in the calculation process of convolutional neural networks, due to a large amount of node information loss, the model has too many parameters and low prediction efficiency. And because approximate personalized propagation uses less parameter information in the calculation, the result can more truly reflect the original data information, reducing the probability of overfitting and making the result more valuable for practical applications.

[0132] The embodiment of this application also provides a device for the project risk level evaluation method based on approximate personalized propagation. The specific implementation of this device can refer to the description in the method for the project risk level evaluation method based on approximate personalized propagation, and the repeated parts will not be elaborated. As Figure 2 shown, this device includes:

[0133] An acquisition module 201, configured to acquire a set of project data of a power engineering construction project;

[0134] A first determination module 202, configured to determine the data division index corresponding to each project data in the set of project data, and determine the data division index as the data feature of the project data, where the risk level evaluation structure is created based on the project risk requirements of the power engineering construction project;

[0135] A prediction module 203 is configured to input each item of data and data features into a pre-trained risk level assessment model, and obtain a risk score corresponding to the power engineering construction project output by the risk level assessment model. The risk level assessment model is a graph neural network model, and approximate personalized propagation values are transmitted among the nodes of the graph neural network model through an approximate personalized propagation mechanism. The risk level assessment model is trained based on project data samples, data feature samples, risk score samples, and approximate personalized propagation values.

[0136] A second determination module 204 is configured to determine the project risk level corresponding to the risk score based on the pre-created correspondence between the risk score and the project risk level.

[0137] Figure 3 An example of the physical structure diagram of an electronic device is shown as Figure 3 shown. The electronic device may include: a processor 301, a communication interface 302, a memory 303, and a communication bus 304. Among them, the processor 301, the communication interface 302, and the memory 303 complete communication with each other through the communication bus 304. The processor 301 can call the logical instructions in the memory 303 to execute the project risk level assessment method based on approximate personalized propagation. The method includes: obtaining a project data set of a power engineering construction project; determining data division indicators corresponding to each item of data in the project data set for a risk level assessment structure, and determining the data division indicators as the data features of the project data. The risk level assessment structure is created based on the project risk requirements of the power engineering construction project; inputting each item of data and data features into a pre-trained risk level assessment model, and obtaining a risk score corresponding to the power engineering construction project output by the risk level assessment model. The risk level assessment model is a graph neural network model, and approximate personalized propagation values are transmitted among the nodes of the graph neural network model through an approximate personalized propagation mechanism. The risk level assessment model is trained based on project data samples, data feature samples, risk score samples, and approximate personalized propagation values; determining the project risk level corresponding to the risk score based on the pre-created correspondence between the risk score and the project risk level.

[0138] In addition, when the logical instructions in the above-mentioned memory 303 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0139] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the project risk level assessment method based on approximate personalized transmission provided by the above-mentioned various methods. The method includes: obtaining a project data set of a power engineering construction project; determining data division indicators corresponding to each project data in the project data set for a risk level assessment structure, and determining the data division indicators as data characteristics of the project data, where the risk level assessment structure is created based on the project risk requirements of the power engineering construction project; inputting each project data and data characteristics into a pre-trained risk level assessment model to obtain a risk score corresponding to the power engineering construction project output by the risk level assessment model. The risk level assessment model is a graph neural network model, and approximate personalized transmission values are transmitted among the nodes of the graph neural network model through an approximate personalized transmission mechanism. The risk level assessment model is trained based on project data samples, data characteristic samples, risk score samples, and approximate personalized transmission values; determining the project risk level corresponding to the risk score based on a pre-created correspondence between the risk score and the project risk level.

[0140] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the project risk level assessment method based on approximate personalized transmission provided in the above-mentioned embodiments, the method comprising: obtaining a project data set of a power engineering construction project; determining a data division index of a risk level assessment structure corresponding to each project data in the project data set, and determining the data division index as a data feature of the project data, wherein the risk level assessment structure is created based on the project risk requirements of the power engineering construction project; inputting each project data and data feature into a pre-trained risk level assessment model to obtain a risk score corresponding to the power engineering construction project output by the risk level assessment model, wherein the risk level assessment model is a graph neural network model, and an approximate personalized transmission value is transmitted in each node of the graph neural network model through an approximate personalized transmission mechanism, and the risk level assessment model is trained based on project data samples, data feature samples, risk score samples and approximate personalized transmission values; based on the correspondence between the pre-created risk score and the project risk level, determining the project risk level corresponding to the risk score.

[0141] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0142] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0143] Finally, it should be noted that the above is only the preferred implementation of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the scope of protection of the present application.

Claims

1. A project risk level assessment method based on approximate personalized transmission, characterized in that The method includes: Obtaining a project data set of a power engineering construction project; Determining data division indicators corresponding to each project data in the project data set for a risk level assessment structure, and determining the data division indicators as data characteristics of the project data, where the risk level assessment structure is created based on the project risk requirements of a power engineering construction project; Inputting each of the project data and the data characteristics into a pre-trained risk level assessment model to obtain a risk score corresponding to the power engineering construction project output by the risk level assessment model, where the risk level assessment model is a graph neural network model, and approximate personalized propagation values are transmitted among the nodes of the graph neural network model through an approximate personalized propagation mechanism, and the risk level assessment model is trained based on project data samples, data characteristic samples, risk score samples, and approximate personalized propagation values; Determining the project risk level corresponding to the risk score based on a pre-created correspondence between the risk score and the project risk level; Wherein, the risk level assessment model includes: at least two nodes; The training process of training the risk level assessment model based on the approximate personalized propagation mechanism includes: Obtaining at least two training data sample sets, where each training data sample set includes: a project data sample corresponding to a power engineering construction project, a data characteristic sample corresponding to the project data sample, and a risk score sample corresponding to the power engineering construction project, and one node corresponds to one training data sample set; Inputting the training data samples into the risk level assessment model, and performing the following data processing operations on each node through the risk level assessment model: Based on the project data sample corresponding to the current node, the data characteristic sample, and the previous approximate personalized propagation value sample transmitted by the previous node, obtaining a current predicted risk score; based on the project data sample, the data characteristic sample, and the current predicted risk score, obtaining a current approximate personalized propagation value sample, and inputting the current approximate personalized propagation value sample into the next node; Based on the predicted risk scores corresponding to each node and the risk score samples corresponding to each node, optimizing the model parameters of the risk level assessment model until, when the number of node jumps reaches a preset number of times, it is determined that the training of the risk level assessment model is completed.

2. The item risk level assessment method based on approximate personalized transmission according to claim 1, wherein, The data processing operations performed on each node further include: When the current node is the first node, obtaining a current predicted risk score based on the project data sample corresponding to the first node and the data characteristic sample; based on the project data sample, the data characteristic sample, and the current predicted risk score, obtaining a current approximate personalized propagation value sample, and inputting the current approximate personalized propagation value sample into the next node.

3. The project risk level assessment method based on approximate personalized transmission according to claim 2, wherein The data processing operations performed on each node further include: When the current node is the last node, obtaining a current predicted risk score based on the project data sample corresponding to the last node, the data characteristic sample, and the previous approximate personalized propagation value sample transmitted by the previous node.

4. The project risk level assessment method based on approximate personalized transmission according to claim 1, wherein The determination process of the number of node jumps π(i x ) includes: Perform vectorization processing on the project data sample and the data feature sample to obtain the node feature matrix i x ; Input the node feature matrix into the jump count calculation formula to obtain the jump count output by the jump count calculation formula; Among them, the jump count calculation formula includes: Among them, π(i x ) represents the number of jumps, α represents the jump probability of moving from the current node to the next node, represents a preset matrix, and i x represents the node feature matrix.

5. The item risk level assessment method based on approximate personalized transmission according to claim 1, characterized in that The obtaining of the current approximate personalized transfer value sample based on the project data sample, the data feature sample, and the current predicted risk score includes: Input the project data sample, the data feature sample, and the current predicted risk score into a pre-created personalized transfer value calculation formula to obtain the personalized transfer value sample output by the personalized transfer value calculation formula; Perform an approximation process on the personalized transfer value sample to obtain an approximate personalized transfer value sample; Among them, the personalized transfer value calculation formula includes: Among them, Z represents the personalized transfer value sample, softmax represents the normalized exponential function, α represents the jump probability of moving from the current node to the next node, represents the preset matrix, H represents the current predicted risk score, and i x represents the node feature matrix.

6. The item risk level assessment method based on approximate personalized transmission according to claim 5, characterized in that The performing of the approximation process on the personalized transfer value sample to obtain an approximate personalized transfer value sample includes: Obtain the initial value of the approximate personalized transfer value sample; Based on the jump of the node, optimize the model parameters of the graph neural network model until the node jump count reaches a preset number of times; Based on the approximate personalized transfer value sample calculation formula obtained for the current node in each iteration, obtain the approximate personalized transfer value sample through the approximate personalized transfer value sample calculation formula; Among them, the approximate personalized transfer value sample calculation formula includes: Among them, Z ′ (k) represents the approximate personalized transfer value sample corresponding to the current node, and Z ′ (k-1) represents the approximate personalized transfer value sample corresponding to the previous node.

7. The project risk level assessment method based on approximate personalized transmission according to claim 1, characterized in that Optimizing the model parameters of the risk level assessment model includes: Optimizing the parameter weights corresponding to the model parameters includes: Use a preset weight calculation formula to calculate the parameter weights corresponding to the model parameters in the current iteration process; Among them, the weight calculation formula includes: Among them, Q t represents the current parameter weight corresponding to the current iteration process, Q t-1 represents the previous parameter weight corresponding to the previous iteration process, γ represents the parameter multiplier, β represents the learning rate, represents the first-order momentum of the t-gradient, represents the second-order momentum of the t-gradient, ∈ represents a constant, and λ represents the regularization coefficient.

8. The project risk level assessment method based on approximate personalized transmission according to claim 1, characterized in that The determination process of the risk score sample includes: Calculate the risk score for the project data sample based on a preset risk score level standard to obtain the calculation result; Convert the calculation result into a corresponding risk coding sample, and use the risk coding sample as the risk score sample.

9. The item risk level assessment method based on approximate personalized transmission according to claim 1, wherein Determining that the training of the risk level assessment model is completed includes: Obtain at least two test data sample sets, where each test data sample set includes: Test project data, test data features corresponding to the test data sample; Perform the following test process on each test data sample set: Input the test project data and the test data sample into the risk level assessment model, and the test risk score output by the risk level assessment model; Determine the accuracy rate of the test risk scores corresponding to all the test data sample sets, and determine that the training of the risk level assessment model is completed when it is determined that the accuracy rate is greater than the preset accuracy rate.

Images