Project progress prediction method and system based on machine learning

Abstract

The invention discloses a project progress prediction method and system based on machine learning, and the method comprises the steps: collecting and integrating project data, and carrying out the preprocessing and feature engineering of the project data; determining a project life cycle, and performing project risk prediction on the project; and performing project progress prediction on the project based on the feature engineering and the project risk prediction result. According to the method, the progress prediction accuracy is improved, dynamic risk assessment and model adaptation are realized, and the project management effect is enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a project progress prediction method and system based on machine learning. Background Art

[0002] In project management, accurately predicting the progress of a project is crucial to ensuring that the project is completed on time, on quality, and on budget. Traditional project progress prediction methods mostly rely on the experience of project managers or are based on simple statistical models, and are difficult to adapt to the complex and changeable risk factors in the project process, such as uneven resource allocation, team performance fluctuations, cost overruns, etc. For example, patent CN114528285A discloses a project prediction method and device, including: obtaining historical project data of a target project; inputting the historical project data into a pre-trained prediction model to obtain the prediction result of the target project, wherein the prediction model is trained using a sample sequence, and the sample sequence is based on an accumulation cycle, and the sample data in the initial data sequence collected according to a preset sampling cycle is accumulated. With the increase in project scale and complexity, traditional methods often show problems of insufficient prediction accuracy and poor robustness when dealing with high uncertainty and multi-dimensional features. Therefore, there is an urgent need for a method and system that can comprehensively consider project data and dynamically evaluate project risk levels for progress prediction. Summary of the invention

[0003] In view of the defects existing in the above-mentioned prior art, the present invention provides a project progress prediction method based on machine learning, and the project progress prediction method comprises:

[0004] Collect and integrate project data, and perform preprocessing and feature engineering on project data;

[0005] Determine the project life cycle and conduct risk forecasting for the project;

[0006] Project progress is predicted based on the results of feature engineering and project risk prediction.

[0007] Specifically, the project schedule forecasting method includes the preparation stage and the forecasting stage;

[0008] The preparation stage requires data collection and integration, as well as preprocessing and feature engineering;

[0009] The prediction stage includes a first stage and a second stage. The first stage determines the project life cycle and performs project risk prediction on the project. In the second stage, the project progress is predicted based on feature engineering and project risk prediction results.

[0010] Wherein, in the preparation stage, the feature engineering includes:

[0011] Calculation of task progress deviation TSD, human resource utilization rate HRUR, capital utilization rate FUR and team efficiency TE.

[0012] Among them, the first stage is to conduct project risk prediction, including dividing the project into multiple project stages according to the project life cycle, and the project stages include the start-up stage, planning stage, execution stage, monitoring and control stage and closing stage;

[0013] And, each stage in the project life cycle corresponds to a different risk assessment function.

[0014] Among them, the risk assessment function satisfies the following relationship:

[0015] R=w1×TSD+w2×HRUR+w3×FUR+w4×TE

[0016] Where R is the risk score, that is, the result value of the risk assessment function, TSD is the task progress deviation, HRUR is the human resource utilization rate, FUR is the capital utilization rate, TE is the team efficiency, w1, w2, w3, and w4 are the weight coefficients of task progress deviation, human resource utilization rate, capital utilization rate, and team efficiency, respectively.

[0017] Among them, the risk items in all risk assessment functions are the same, and the risk items include task progress deviation, human resource utilization rate, capital utilization rate and team efficiency;

[0018] Also, the weights of the risk terms in all risk functions are different.

[0019] Among them, the task progress deviation calculation satisfies the following relationship:

[0020]

[0021] The calculation of human resource utilization satisfies the following relationship:

[0022]

[0023] The calculation of fund utilization rate satisfies the following relationship:

[0024]

[0025] Team efficiency calculation satisfies the following relationship:

[0026]

[0027] In the formula, TSD is the task progress deviation, HRUR is the human resource utilization rate, FUR is the capital utilization rate, and TE is the team efficiency.

[0028] The correspondence between each risk item and its weight can be stored in the form of a list, and the system obtains the weight distribution of different risk items corresponding to the project stage by querying the weight table.

[0029] Among them, in the first stage, four risk levels of low risk, medium risk, high risk and extremely high risk are determined based on the output results of the risk assessment function, including determining the membership of the output result R (i.e., RiskScore) of the risk assessment function under the four risk levels based on the fuzzy membership function, and determining the risk level of the current stage through the maximum membership principle, and selecting the risk level with the highest membership as the risk level of R.

[0030] Among them, the membership of the output result R under the four risk levels satisfies the following relationship:

[0031]

[0032] In the formula, μ s (R) is the membership degree of the output result R at the corresponding risk level, μ L (R) is the membership degree at low risk level, k L >0, which controls the steepness (slope) of the membership function at low risk levels, c L is the center (turning point) of the membership function at low risk level; μ M (R) is the membership degree under the medium risk level, c M is the center position of the membership function under medium risk level, σ M >0, which controls the width (standard deviation) of the membership function under the medium risk level; μ H (R) is the membership degree under high risk level, c H is the central position of the membership function under high risk level, σ H Controls the width of the membership function at high risk levels; μ VH (R) is the membership degree under extremely high risk level, k VH >0 controls the steepness of the membership function at extremely high risk levels, c VH is the center of the membership function at the extremely high risk level.

[0033] Among them, the project progress forecast is carried out in the second stage, including selecting different machine learning models according to the risk level obtained by project risk forecasting, and forecasting the project progress in combination with the results of feature engineering.

[0034] Among them, the selection of machine learning models based on risk level includes:

[0035] At low and medium risk, the multiple linear regression model was selected;

[0036] Under high risk, choose the decision tree regression model;

[0037] and, at extremely high risk, choosing a deep neural network model.

[0038] Among them, when the risk is low and medium, the selected multiple linear regression model satisfies the following relationship:

[0039]

[0040] In the formula, is the predicted value of the project progress, β0 is the intercept term of the model, β k is the regression coefficient of the kth feature, X k is the kth feature, including the risk term, and p is the total number of features;

[0041] The corresponding loss function satisfies the following relationship:

[0042]

[0043] In the formula, L is the total loss, m is the number of samples, and y is j is the true value of the jth sample, is the predicted value of the jth sample, w j is the weight of the jth sample, which is calculated based on the risk item of the sample and the corresponding weight

[0044] Among them, according to the project stage of the project, select the machine learning model corresponding to the project stage.

[0045] The present invention discloses a project progress prediction system based on machine learning, which adopts the project progress prediction method based on machine learning. The system includes:

[0046] The acquisition and processing module is used to collect and integrate project data and perform preprocessing and feature engineering on the project data;

[0047] The risk prediction module is used to determine the project life cycle and predict the project risks, including:

[0048] Divide the project into multiple project phases according to the project life cycle, including the initiation phase, planning phase, execution phase, monitoring and control phase, and closing phase;

[0049] Based on different risk assessment functions corresponding to each project stage in the project life cycle, project risk prediction is performed on the project;

[0050] The progress prediction module is used to predict the project progress based on the results of feature engineering and project risk prediction.

[0051] The present invention also discloses a project progress prediction system based on machine learning, the system comprising a processor and a memory, the memory storing a computer program, and the processor being used to execute the computer program stored in the memory to implement the aforementioned method.

[0052] The present invention solves the problems of insufficient risk consideration and low prediction accuracy at different stages in the existing methods by introducing a progress prediction method based on machine learning. Four risk factors, namely task progress deviation, human resource utilization, capital utilization and team efficiency, are defined, and weights are assigned according to each stage. The risk level is divided using a fuzzy membership function, and a linear regression, decision tree regression or deep neural network model is dynamically selected. The weighted mean square error is introduced in model training to increase sensitivity to key risks. This method improves the accuracy of progress prediction, realizes dynamic risk assessment and model adaptation, and enhances project management and resource allocation effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0054] Figure 1 is a flow chart showing a project progress prediction method based on machine learning according to an embodiment of the present invention;

[0055] Figure 2 is a schematic diagram showing a project progress prediction system based on machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0058] It should be understood that although the terms first, second, third, etc. may be used to describe ... in the embodiments of the present invention, these ... should not be limited to these terms. These terms are only used to distinguish .... For example, without departing from the scope of the embodiments of the present invention, the first ... may also be referred to as the second ..., and similarly, the second ... may also be referred to as the first ....

[0059] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0060] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0061] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0062] The invention discloses a project progress prediction method based on machine learning. The project progress prediction method comprises a preparation stage and a prediction stage. The prediction stage comprises a first stage and a second stage.

[0063] Among them, the preparation stage requires data collection and integration, preprocessing, and feature engineering.

[0064] Data collection and integration, including the collection of project management data, team performance data, human resources and other data. The above data are integrated into a unified data set as the basis for subsequent analysis and modeling. Among them, project management data includes the number of planned completed tasks, the actual number of completed tasks, actual expenditures and budgeted expenditures of the project at the corresponding stage. Team performance data includes the number of each team member and the work efficiency score of each member at the corresponding stage of the project. Human resources include the actual manpower used and the allocated manpower of the project at the corresponding stage. The work efficiency score of each member can be determined by the progress of the tasks assigned to each member in the corresponding stage and the progress of the tasks actually completed.

[0065] Data preprocessing includes cleaning the collected data, handling missing values ​​and outliers, and ensuring the integrity and accuracy of the data.

[0066] Feature engineering, which involves selecting important features as input to the model.

[0067] The prediction stage includes a first stage and a second stage. The first stage performs project risk prediction, and the second stage performs project schedule prediction based on the project risk prediction results.

[0068] In one embodiment, in the preparation stage, the feature engineering includes four important features, which can fully reflect the current status and potential risks of the project and ensure the efficiency of the model.

[0069] Specifically, the risk item definition includes:

[0070] Task Schedule Deviation (TSD):

[0071]

[0072] Human Resource Utilization Rate (HRUR):

[0073]

[0074] Funds Utilization Rate (FUR):

[0075]

[0076] Team Efficiency (TE):

[0077]

[0078] Among them, TSD (task progress deviation): reflects the difference between the actual progress of the project and the planned progress. The larger the deviation, the higher the risk of task delay. HRUR (human resource utilization rate): evaluates the efficiency of human resource utilization. Too high utilization may lead to excessive fatigue of personnel, while too low utilization indicates waste of resources. FUR (fund utilization rate): shows the ratio between actual fund consumption and budget. Overspending will increase the financial risk of the project. TE (team efficiency): measures the work efficiency of team members. Inefficiency may lead to project delays or reduced quality.

[0079] In one embodiment, project risk prediction is performed in the first stage, including dividing the project into a start-up stage, a planning stage, an execution stage, a monitoring and control stage, and a closing stage according to the project life cycle.

[0080] Optionally, each stage in the project life cycle corresponds to a different risk assessment function, and the risk items in the different risk assessment functions are the same, including task progress deviation, human resource utilization, capital utilization, and team efficiency. And, according to the characteristics of different stages, the weights of the risk items in the risk functions of different stages are different. For example, in the execution stage, more attention will be paid to human resource utilization and team efficiency, while in the planning stage, more attention will be paid to task progress deviation and capital utilization.

[0081] Optionally, dynamic risk assessment is performed through project risk prediction, including using fuzzy logic to calculate the membership of the output results of the risk assessment function, accurately reflecting the main risks faced by the project at different stages, and providing a scientific basis for the selection of models for project schedule prediction.

[0082] In one embodiment, the following are the risk assessment functions, corresponding formulas and parameter definitions for each stage.

[0083] 1. Start-up phase

[0084] The start-up phase focuses on determining project goals, feasibility analysis, and initial resource allocation. This phase pays close attention to the initial investment of funds and human resources.

[0085] Risk assessment function: RiskScore = w3×FUR+w2×HRUR,

[0086] Weight coefficient:

[0087] w3=0.6

[0088] w2=0.4

[0089] 2. Planning Stage

[0090] The planning stage focuses on developing a detailed project plan, including task breakdown, schedule scheduling, and budget preparation. This stage requires a high level of accuracy in task schedules and funding allocation.

[0091] Risk assessment function: RiskScore = w1×TSD+w3×FUR

[0092] Weight coefficient:

[0093] w1=0.5

[0094] w3=0.5

[0095] 3. Implementation Phase

[0096] The execution phase is the phase where project activities are actually carried out, with the emphasis on effective use of resources and team collaboration, so there is a strong focus on human resource utilization and team efficiency.

[0097] Risk assessment function: RiskScore = w2×HRUR+w4×TE,

[0098] Weight coefficient:

[0099] w2=0.5

[0100] w4=0.5

[0101] 4. Monitoring and Control Phase

[0102] The monitoring and control phase continuously focuses on the execution of the project, monitors progress, cost, and quality, and ensures that the project proceeds as planned.

[0103] Risk assessment function: RiskScore = w1×TSD+w2×HRUR+w3×FUR+w4×TE,

[0104] Weight coefficient:

[0105] w1=0.25

[0106] w2=0.25

[0107] w3=0.25

[0108] w4=0.25

[0109] 5. Closing stage

[0110] The closing phase mainly involves the final delivery, acceptance and release of resources of the project, focusing on the on-time completion of tasks and the final use of budget.

[0111] Risk assessment function: RiskScore = w1×TSD+w3×FUR,

[0112] Weight coefficient:

[0113] w1=0.6

[0114] w3=0.4

[0115] In one embodiment, as shown in the following example table, at different stages of project management, the corresponding relationship between each risk item and weight can be stored in the form of a list. In the risk assessment at different stages, the system can obtain the weight distribution of different risk items corresponding to the project stage by querying the weight table.

[0116] Project Phases TSD HRUR FUR TE Start-up phase 0 0.4 0.6 0 Planning Phase 0.5 0 0.5 0 Execution Phase 0 0.5 0 0.5 Monitoring and control phase 0.25 0.25 0.25 0.25 Closing Stage 0.6 0 0.4 0

[0117] Optionally, in the first stage, four risk levels of low risk, medium risk, high risk and extremely high risk are determined based on the output result of the risk assessment function, including determining the membership μ of the output result R of the risk assessment function at the four risk levels based on the fuzzy membership function. L (R), and determine the risk level of the current stage through the maximum membership principle, and select the risk level with the highest membership as the risk level of R. That is, for the output result R of the given risk assessment function, calculate its membership μ at each risk level S (R), select the risk level with the highest membership as the risk level of R, that is, the risk level of the current stage. If R belongs to multiple risk levels at the same time in a certain interval, its final level is determined by which level has a higher membership.

[0118] In one embodiment, a smooth continuous function is used to define the fuzzy membership function.

[0119] The value range of risk score R is set to: R∈[0,100].

[0120] Risk level definitions, including:

[0121] Low risk (Low, denoted as L)

[0122] Medium risk (Medium, denoted as M)

[0123] High risk (High, denoted as H)

[0124] Very High (VH)

[0125] In order to achieve a smooth transition between risk levels, a smoothing function is used to define the membership function, and Gaussian function and Sigmoid function (S-type function) are used as smoothing functions.

[0126] In one embodiment, the membership function based on a smoothing function may include:

[0127] 1. Low risk membership function μ L (R):

[0128] Take the left sigmoid function:

[0129] in,

[0130] k L >0: controls the steepness (slope) of the function.

[0131] c L : The center (turning point) of the function.

[0132] Optionally, k L =0.1, c L =50.

[0133] in,

[0134] When R=0, μ L (0)≈1, indicating a completely low risk.

[0135] When R = c L =50, μ L (50)=0.5.

[0136] When R→100, μ L (R)→0.

[0137] 2. Medium risk membership function μ M (R):

[0138] Using Gaussian function:

[0139] in,

[0140] c M : The center position of the function.

[0141] σ M >0: controls the width (standard deviation) of the function.

[0142] Optionally: c M =50,σ M =15

[0143] When R = c M =50, μ M (50) = 1, indicating that the medium risk reaches the maximum membership.

[0144] Function at R = c M ±σ M The membership degree at φ drops to about 0.6065.

[0145] The function is symmetric about the center point c M.

[0146] 3. High-risk membership function μ H (R):

[0147] Using Gaussian function:

[0148] in:

[0149] c H =75: High-risk central location.

[0150] σ H =15: Control function width.

[0151] Optionally: c H =75,σ H =15

[0152] When R = c H =75, μ H (75)=1.

[0153] Function at R = c H ±σ H The membership degree at φ drops to about 0.6065.

[0154] 4. Extremely high risk membership function μ VH (R):

[0155] Use the right-hand side sigmoid function:

[0156] in:

[0157] k VH >0: controls the steepness (slope) of the function.

[0158] c VH : The center (turning point) of the function.

[0159] Optionally: k VH =0.1, c VH =75

[0160] When R = c VH =75, μ VH (75)=0.5.

[0161] When R→100, μ VH (R)→1, indicating that the risk is extremely high.

[0162] In one embodiment, the risk level is ultimately determined as:

[0163]

[0164] Parameter definition:

[0165] Risk Level: The risk level corresponding to the comprehensive risk score R.

[0166] arg max: Take the risk level that maximizes the degree of membership among the parameters.

[0167] S: Risk level, the values ​​are low risk, medium risk, high risk, and extremely high risk.

[0168] μ S (R): The membership degree of risk level S corresponding to R.

[0169] For example, assuming that the risk score R=65, calculate its membership at each risk level:

[0170] 1. Low risk membership μ L (65):

[0171] 2. Medium risk membership μ M (65):

[0172] 3. High risk membership μ H (65):

[0173] 4. Extremely high risk membership μ VH (65):

[0174] According to the maximum membership principle, select the risk level with the largest membership:

[0175] μ L (65)≈0.1824

[0176] μ M (65)≈0.3679

[0177] μ H (65)≈0.6412(maximum)

[0178] μ VH (65)≈0.2689

[0179] Therefore, the risk level at this stage is high.

[0180] In one embodiment, the second stage performs a project schedule forecast, including selecting different machine learning models to forecast the project schedule according to the risk level obtained by the project risk forecast. Selecting different machine learning models based on the risk level improves the accuracy and robustness of the forecast, and can adapt to the needs of different risk changes during the project process.

[0181] In one embodiment, the model selection based on risk level includes:

[0182] For low risk and medium risk, select the multiple linear regression model.

[0183] Under High Risk, select Decision Tree Regression Model.

[0184] And under Very High Risk, select the Deep Neural Network model.

[0185] In a certain embodiment, the project schedule is predicted in the second stage, including selecting different machine learning models for project schedule prediction according to the risk level obtained by the project risk prediction, and selecting the machine learning model corresponding to the project stage according to the project stage in which the project is located. In a certain embodiment, the project schedule is predicted in the second stage, including selecting different machine learning models for project schedule prediction according to the risk level obtained by the project risk prediction, and selecting the machine learning model of the corresponding stage according to the stage in which the project is located, and the machine learning model of the corresponding stage is trained based on different loss functions. The risk items of the different loss functions are the same, but the weights corresponding to the same risk item in different stages are different. Optionally, the weight can be obtained by querying the weight table. Wherein, when the project schedule is predicted according to the selected machine learning model, the features or parameters of the input machine learning model may include the completion time and historical human and material input of historical similar projects, or external environmental factors and external environmental factors of historical similar projects in addition to the risk items. External environmental factors may include weather conditions, natural disasters, etc. In addition, when implementing project schedule forecasting, the forecast results may include at least one of the following: estimated completion time, schedule deviation, critical path update, resource demand forecast, risk assessment, project milestone, task priority, project status report, adjustment suggestion, performance indicator, and visualization chart. Estimated completion time: predict the estimated completion date of a project or a specific task in the project. Schedule deviation: the difference between the actual progress and the planned progress, usually expressed in time (days) or percentage. Critical path update: the critical path of the project may change due to changes in the actual progress, and the forecast results will update the critical path to reflect the latest project status; the critical path is the longest logical task chain in the project, which determines the total duration of the project, and any delay on the critical path will cause the entire project to be delayed. Resource demand forecast: based on the predicted progress, the amount of resources required for each future stage can be predicted, including manpower, materials, and funds. Risk assessment: the possible impact of potential risk items on the project progress. Project milestone: predict the completion time of each key milestone in the project. Task priority: based on the forecast results, it can be determined which tasks should be executed first to avoid project delays. Project Status Report: Provides a detailed report on the current status of the project, including completed work, work in progress, and planned work. Adjustment Suggestions: If the forecast results show that the project may be delayed, the forecast results may include suggestions for adjusting the project plan, such as adding resources, optimizing workflows, or reallocating tasks. Performance Indicators: The forecast results may include indicators of project performance, such as the Cost Performance Index (CPI), Schedule Performance Index (SPI), etc. Visual Charts: The results of the schedule forecast are often presented in the form of a Gantt chart, S-curve chart, or other project management chart to visually display the project progress.

[0186] In one embodiment, when the loss function is adjusted based on the project stage, including for low risk (Low Risk) and medium risk (Medium Risk), the project schedule prediction model is a multiple linear regression model, and the model form is:

[0187]

[0188] in:

[0189] Project schedule forecast

[0190] β0: intercept term of the model

[0191] β k : The regression coefficient of the kth feature

[0192] X k : The kth feature, including the risk term

[0193] p: total number of features, p=4.

[0194] The corresponding loss function is:

[0195]

[0196] in,

[0197] L: Total loss

[0198] m: sample size

[0199] y j : The true value of the jth sample

[0200] The predicted value of the jth sample

[0201] w j : The weight of the jth sample, calculated based on the risk item of the sample and the corresponding weight.

[0202] Among them, the sample weight w j The calculation includes obtaining the risk item weight ω from the weight table according to the current project stage. i .

[0203] For each sample j, calculate:

[0204] in,

[0205] ω i: The weight of the i-th risk item in the current project stage (obtained from the weight table)

[0206] X ji : The eigenvalue of the i-th risk item of the j-th sample

[0207] n: number of risk items (here n=4).

[0208] In one embodiment, the loss function is adjusted based on the project stage, including for high risk, the project progress prediction model is a decision tree regression model, the data set is recursively divided according to the features to form a tree structure, and the leaf nodes contain the predicted values.

[0209] When training a decision tree regression model, the key to node splitting is to select the best features and thresholds to split the data set into left and right child nodes so that the model can best fit the data. j The weighted mean square error (WMSE) is used as the criterion for node splitting to ensure that the model pays more attention to the key risk items at this stage during training.

[0210] The goal of node splitting in the decision tree regression model includes finding an optimal splitting solution (i.e., selecting the best features and splitting points) at each non-leaf node, splitting the sample set S of the current node into left child nodes S left and right child node S right , so that the sum of weighted mean square errors after splitting is minimized.

[0211] The calculation of weighted mean square error includes:

[0212] For node S, its weighted mean square error L(S) is defined as:

[0213]

[0214] Among them, the weighted average of the target variable in node S Defined as: in:

[0215] S: The sample set of the current node, including all samples on this node.

[0216] j∈S: the jth sample belongs to node S.

[0217] w j : The weight of the jth sample, calculated as: in,

[0218] ω i : The weight of the i-th risk item in the current project stage (obtained from the weight table). X ji : The eigenvalue of the i-th risk item of the j-th sample. n: The number of risk items (here 4). j : The true target value (project progress) of the jth sample.

[0219] The weighted average of the target variable of the samples in node S.

[0220] The specific steps of the node splitting process are:

[0221] 1. At the beginning of training, the root node contains all training samples.

[0222] 2. For the current node S, perform the following steps:

[0223] (1) Traverse all possible split features and split points:

[0224] Feature selection: For each available feature k.

[0225] Split point selection: For feature k, consider all possible split points (usually the middle value of the feature value).

[0226] (2) For each possible split, do the following:

[0227] Split node S into its left child node S left and right child node S right :Left child node: S left ={j∈S|X jk ≤t}, right child node: S right ={j∈S|X jk >t}. Among them, X jk : The value of the jth sample on the kth feature; t: The split point currently considered (a certain value of feature k).

[0228] Calculate the weighted mean square error sum after splitting:

[0229] in,

[0230] L total : The sum of weighted mean square errors after splitting. It is used to measure the weighted mean square errors of the left and right child nodes under the current splitting scheme.

[0231] S: The sample set of the current node, including all samples on this node.

[0232] S left : The sample set of the left child node contains samples that meet the splitting conditions (such as a certain eigenvalue is less than or equal to the splitting point).

[0233] S right : The sample set of the right child node contains samples that do not meet the splitting conditions (such as a certain eigenvalue is greater than the splitting point).

[0234] w j: The weight of the jth sample, reflecting the importance of the sample in model training.

[0235] L(S left ): Weighted mean square error of the left child node.

[0236] L(S right ): Weighted mean square error of the right child node.

[0237] w j : The weight of the jth sample.

[0238] Where L(S left ): Weighted mean square error of the left child node.

[0239]

[0240] Parameter definition:

[0241] L(S left ): The weighted mean square error of the left child node, which represents the weighted average square error between the predicted value and the true value in the left child node.

[0242] The sum of weights of all samples in the left child node.

[0243] y j : The true target value of the jth sample (such as the actual progress of the project).

[0244] The weighted average of the target variable in the left child.

[0245] w j : The weight of the jth sample.

[0246] in, Weighted average of the target variable in the left child:

[0247]

[0248] in,

[0249] The weighted average of the target variable (such as project progress) in the left child node.

[0250] The sum of the product of the target value of each sample in the left child node and its weight.

[0251] w j : The weight of the jth sample.

[0252] Where: L(S right ): Weighted mean square error of the right child node:

[0253]

[0254] in,

[0255] L(S right ): The weighted mean square error of the right child node, which represents the weighted average square error between the predicted value and the true value in the right child node.

[0256] The sum of weights of all samples in the right child node.

[0257] y j : The true target value of the jth sample.

[0258] The weighted average of the target variable in the right child node.

[0259] w j : The weight of the jth sample.

[0260] Weighted average of the target variable in the right child node:

[0261]

[0262] in:

[0263] The weighted average of the target variable in the right child node.

[0264] The sum of the product of the target value of each sample in the right child node and its weight.

[0265] w j : The weight of the jth sample.

[0266] Among them, the sample weight w j :

[0267] in,

[0268] w j : The weight of the jth sample.

[0269] n: number of risk items (here 4 risk items).

[0270] ω i : The weight of the ith risk item in the current project stage (obtained from the risk item weight table).

[0271] X ji : The characteristic value of the i-th risk item of the j-th sample.

[0272] That is w j =ω1Xj1 +ω2X j2 +ω3X j3 +ω4X j4 , where X j1 ,X j2 ,X j3 ,X j4 They represent the eigenvalues ​​of TSD, HRUR, FUR, and TE of the j-th sample respectively.

[0273] (3) Choose the weighted mean square error L total Minimal splitting solution:

[0274] Among all possible splits, find the one that makes L total The smallest feature k * and split point t * .

[0275] 3. Update the tree structure:

[0276] According to the selected optimal splitting scheme, node S is split into S left and S right .

[0277] 4. Recursively perform the above process on the left and right child nodes:

[0278] For each child node, repeat the above steps until the stopping condition is met.

[0279] The training stop condition is:

[0280] Preset maximum depth: When the depth of the tree reaches the preset maximum depth, stop splitting. Minimum number of samples: When the number of samples in a node is less than the preset minimum number of samples, stop splitting.

[0281] Insufficient decrease in weighted mean square error: When the decrease in weighted mean square error after splitting is not enough to improve model performance, stop splitting.

[0282] In one embodiment, during the node splitting process of the decision tree regression model, for each possible splitting scheme, the weighted mean square error of the left and right child nodes is calculated, and the following steps are performed:

[0283] 1. Calculate sample weight w j :

[0284] ·According to the risk item weights of the current project stageω i and the risk item eigenvalue X of the sample ji , calculate the weight of each sample.

[0285] 2. Calculate the weighted mean square error L(S left ) and L(Sright ):

[0286] Using the sample weights, we can perform a weighted average of the sample target values ​​in the left and right child nodes to obtain and Calculate the weighted mean square error of the left and right child nodes to measure the degree of deviation between the predicted value and the true value.

[0287] 3. Calculate the weighted mean square error sum L after splitting total :

[0288] The weighted mean square errors of the left and right child nodes are weighted summed according to the proportion of their respective sample weights to the total weight to obtain the total error after splitting.

[0289] 4. Choose the best splitting scheme:

[0290] Traverse all possible features and split points and repeat the above calculations.

[0291] Select L total The smallest split solution is taken as the optimal split for the current node.

[0292] In one embodiment, the loss function is adjusted based on the project stage, including for extremely high risk, the project progress prediction model is a deep neural network model, the model structure includes an input layer, multiple hidden layers and an output layer, and the nodes of each layer are fully connected to the nodes of the next layer. In the forward propagation, the activation value of the lth layer: a (l) =f (l) (W (l) a (l-1) +b (l) ), the output layer is the predicted value:

[0293] in:

[0294] a (l) : The activation value vector of the lth layer

[0295] W (l) : The weight matrix of the lth layer

[0296] b (l) : Bias vector of the lth layer

[0297] f (l) : The activation function of the lth layer

[0298] L: total number of layers in the network

[0299] The predicted value of the jth sample

[0300] n (l) : The number of neurons in the lth layer

[0301] The activation value of the jth sample in the output layer

[0302] The loss function in the deep neural network model is:

[0303]

[0304] in,

[0305] L: Total loss

[0306] m: sample size

[0307] y j : The true value of the jth sample

[0308] The predicted value of the jth sample

[0309] w j : The weight of the jth sample, calculated based on the risk item of the sample and the corresponding weight.

[0310] Among them, the sample weight w j The calculation includes obtaining the risk item weight ω from the weight table according to the current project stage. i .

[0311] For each sample j, calculate:

[0312] in,

[0313] ω i : The weight of the i-th risk item in the current project stage (obtained from the weight table)

[0314] X ji : The eigenvalue of the i-th risk item of the j-th sample

[0315] n: number of risk items (here n=4).

[0316] The present invention selects three machine learning models corresponding to different risk levels, and within each machine learning model, five different models of different loss functions correspond to different project stages, so 3*5=15 trained machine learning models can be provided. Different risk levels correspond to different machine learning models, high risk corresponds to high computational model, and low risk corresponds to low computational model. Different loss functions are introduced in the machine learning model corresponding to different current project stages. By querying the weight table, the weights of different features at different stages are different to reflect the difference in loss functions at different project stages. In the implementation stage, the risk level is obtained by inputting the fuzzy membership function according to the risk score, and then the corresponding trained machine learning model is selected according to the obtained risk level and the current project stage, and the output can obtain the prediction of the project progress. The present invention is to refine the machine learning model training, train different machine learning models for project progress prediction according to the different features concerned in the project stage, and achieve more accurate project progress prediction results. At the same time, selecting different machine learning models can also reduce the use of computing resources in training models and implementation.

[0317] The present invention solves the problem of insufficient consideration of risk factors at different stages and low prediction accuracy in the existing project progress prediction method. By introducing a project progress prediction method based on machine learning, data collection and preprocessing, feature engineering, project risk prediction and progress prediction are integrated. In particular, four risk items, namely task progress deviation, human resource utilization, capital utilization and team efficiency, are defined for different stages of the project life cycle, and the weight distribution of risk items at each stage is determined by a weight table. The risk assessment results are divided into risk levels using a fuzzy membership function, and a machine learning model (such as linear regression, decision tree regression or deep neural network) suitable for the current risk level is dynamically selected. In the model training process, a weighted mean square error loss function is introduced, and the sensitivity of the model to key risk factors is improved in combination with the weight of the risk item. Through the above method, the accuracy of project progress prediction is improved, the dynamic evaluation of project risks and the adaptive selection of models are realized, and the effectiveness of project management and resource allocation is enhanced.

[0318] like Figure 2 As shown, the present invention discloses a project progress prediction system based on machine learning, which adopts the above-mentioned project progress prediction method based on machine learning. The system includes:

[0319] The acquisition and processing module is used to collect and integrate project data and perform preprocessing and feature engineering on the project data;

[0320] Risk prediction module, which is used to determine the project life cycle and predict the project risks;

[0321] The progress prediction module is used to predict the project progress based on the results of feature engineering and project risk prediction.

[0322] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0323] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0324] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0325] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0326] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0327] The above introduces the preferred embodiments of the present invention, which is intended to make the spirit of the present invention clearer and easier to understand, but is not intended to limit the present invention. All modifications, substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection outlined by the claims attached to the present invention.

Claims

1. A project progress prediction method based on machine learning, the project progress prediction method comprising: Collect and integrate project data, and perform preprocessing and feature engineering on project data; Determine the project life cycle and conduct risk forecasting for the project; Forecast the project progress based on the results of feature engineering and project risk prediction; Among them, project risk forecasting includes: Divide the project into multiple project phases according to the project life cycle, including the initiation phase, planning phase, execution phase, monitoring and control phase, and closing phase; Based on the different risk assessment functions corresponding to each project stage in the project life cycle, project risk prediction is performed on the project.

2. A project progress prediction method based on machine learning as claimed in claim 1, characterized in that: The feature engineering includes: Calculation of task progress deviation, human resource utilization, capital utilization and team efficiency.

3. A project progress prediction method based on machine learning as claimed in claim 1, characterized in that: The risk items in all risk assessment functions are the same, and the risk items include task progress deviation, human resource utilization rate, capital utilization rate and team efficiency; Also, the weights of the risk terms in all risk functions are different.

4. A project progress prediction method based on machine learning as claimed in claim 3, characterized in that: The correspondence between each risk item and its weight is stored in the form of a list, and the weight distribution of different risk items corresponding to the project stage is obtained by querying the weight table.

5. A project progress prediction method based on machine learning as claimed in claim 1, characterized in that: Based on the output of the risk assessment function, four risk levels of low risk, medium risk, high risk and extremely high risk are determined, including the membership of the output of the risk assessment function under the four risk levels μ respectively determined based on the fuzzy membership function. S (R), and determine the risk level of the project stage corresponding to the risk assessment function through the maximum membership principle.

6. A project progress prediction method based on machine learning as claimed in claim 5, characterized in that: The risk assessment function satisfies the following relationship: R=w1×TSD+w2×HRUR+w3×FUR+w4×TE Where R is the risk score, TSD is the task progress deviation, HRUR is the human resource utilization rate, FUR is the capital utilization rate, TE is the team efficiency, w1, w2, w3, and w4 are the weight coefficients of task progress deviation, human resource utilization rate, capital utilization rate, and team efficiency, respectively.

7. A project progress prediction method based on machine learning as claimed in claim 5, characterized in that: Perform project progress forecasting, including selecting different machine learning models based on the risk level obtained from project risk forecasting, and forecasting the project progress based on the results of feature engineering.

8. A project progress prediction method based on machine learning as claimed in claim 7, characterized in that: Machine learning model selection based on risk level, including: At low and medium risk, the multiple linear regression model was selected; Under high risk, choose the decision tree regression model; and, at extremely high risk, choosing a deep neural network model.

9. A project progress prediction method based on machine learning as claimed in claim 5, characterized in that: The output of the risk assessment function has membership μ under four risk levels S (R), and determine the risk level of the project stage corresponding to the risk assessment function through the maximum membership principle, satisfying the following relationship: In the formula, μ s (R) is the membership degree of the output result R at the corresponding risk level, μ L (R) is the membership degree at low risk level, k L is the steepness of the membership function at low risk level, c L is the center of the membership function at low risk level; μ M (R) is the membership degree under the medium risk level, c M is the center position of the membership function under medium risk level, σ M is the width of the membership function at the medium risk level; μ H (R) is the membership degree under high risk level, c H is the central position of the membership function under high risk level, σ H is the width of the membership function at high risk level; μ VH (R) is the membership degree under extremely high risk level, k VH is the steepness of the membership function at the extremely high risk level, c VH is the center of the membership function at the extremely high risk level.

10. A project progress prediction system based on machine learning, characterized in that: Using the project progress prediction method based on machine learning as described in any one of claims 1 to 9, the system includes: The acquisition and processing module is used to collect and integrate project data and perform preprocessing and feature engineering on the project data; The risk prediction module is used to determine the project life cycle and predict the project risks, including: Divide the project into multiple project phases according to the project life cycle, including the initiation phase, planning phase, execution phase, monitoring and control phase, and closing phase; Based on different risk assessment functions corresponding to each project stage in the project life cycle, project risk prediction is performed on the project; The progress prediction module is used to predict the project progress based on the results of feature engineering and project risk prediction.

Images