CN101561904A - Process data-based method and system for determining cost of software project - Google Patents

Abstract

The invention discloses a process data-based method and a process data-based system for determining the cost of a software project and belongs to the field of software engineering. The method comprises the following steps: taking various cost driving factors and uncertain information thereof in the process of a software project into overall consideration, acquiring the probability distribution of the values of the various cost-driving factors of the project; establishing a determination model base, selecting parameters of a determination model according to the data calibration of the cost-driving factors of a historical project, and obtaining the probability distribution of the parameters; generating a determination model instance and cost-driving factor instance by sampling; simulating the cost of the software project, setting the times of sampling experiments, and carrying out repeated software project cost simulation experiments; and calculating the results of the software project cost simulation experiments and determining cost information. The system of the invention realizes automatic processing; and ordinary software development personnel can use the system to determine software project cost and to recognize the uncertainty of the cost of the software project and corresponding risks of various cost budgeting levels conveniently.

Description

A kind of software project cost assay method and system based on process data

Technical field

The invention belongs to the computer software engineering field, relate to cost determination techniques in the software project process, particularly a kind of assay method of computer software project cost and system.

Background technology

Since computing machine was born, it used each field such as space flight, finance, manufacturing, communication, the energy, medical treatment, education that have been deep into, and people's work relies on computer software more and more with life.People not only need computer software to realize more function, are also had higher requirement in aspects such as the ease for use of software, reliability, security, user experience.For satisfying people computer software is got more and more and more and more higher demand, soft project is arisen at the historic moment.

Software process is defined by a series of partial order process steps, each step generally comprises product, personnel, computer resource, institutional framework, constraint etc. (sees also document Carlo Montangero, Jean-Claude Derniame, Badara AliKaba, and Brian Warboys.The software process:Modelling and technology.In Derniame et al.[126] .Pages 1-14. and CMMI Product Team.Cmmi for development, version 1.2-improvingprocesses for better products.Technical Report CMU/SEI-2006-TR-008, SEI, CMU, 2006.).One software process can be explained by various data, and the data of these reflection software process can be stored in the software process management database, forms the history item data and (sees also document Jones 2000, ISBSG; Capers Jones, Software Assessments, Benchmarks, and Best Practices, Addison-Wesley Professional, 2000, ISBSG, www.isbsg.org), these data have comprised multiple cost and have driven the factor.

Because the target of computer software is to be user's creation of value, and people must make suitable decision-making under resource (as: human and material resources, time) condition of limited in the development and maintenance process of software, and economics (promptly studying the science how people make a policy under the situation of resource-constrained) becomes an important science of the required utilization of soft project.And in soft project, meeting economic decision-making for making, a most crucial activity is that the software project cost is measured.

Software project cost mensuration is the prediction to cost attribute in the software process, because the overwhelming majority is a human cost in the software project cost, the software project cost is measured the mensuration that is often referred to workload (human cost).The importance that the software project cost is measured software project is embodied in: be that analysis software project feasibility, formulation software project budget and software project relate to the basis that the crowd holds consultation, being balance software development strategy important evidence, also is to improve the software process and the important evidence of boosting productivity.

The complicacy that the software project cost is measured, at first being embodied in has the multiple cost driving factor (as software size, software complexity, developer's experience, developing instrument etc.) that the influence that the software project cost produces complexity (is seen also document Boehm 1981 in the software development process, Johns 2000, ISBSG).Therefore, existing a lot of cost assay methods are by setting up the mode of cost rating model, describe clearly the software project cost drive the factor and and the software project cost between get in touch, measure to help people.These existing rating models mainly comprise: analog model (referring to document Shepperd andSchofield 1997), regression model are (referring to document Boehm 1981; Draper and Smith 1981), classification regression tree (referring to document Briand and Wust 2001), artificial neural network (referring to document Shukla 2000) etc.

But the software project cost is measured the difficulty that also faces a core and is not well solved, i.e. the uncertainty of software project cost mensuration.Each stage at software project exists various uncertainties, as: fuzzy and omission, changes in demand, commercial member applicability, product complexity, architectural approach, system's external interface, the flow of personnel etc. that relate to crowd, software requirement of project.The software project cost is measured and is made under these probabilistic environment, and the deviation amplitude of measurement result or uncertain can the reduction gradually along with the progress of project.

Uncertainty is the essential attribute that the software project cost is measured, and the project manager do not know how to carry out correct mensuration usually, particularly the uncertainty measured of processing cost correctly.Early stage at software project, the cost of software project and progress have very high uncertainty, and why a lot of projects fail is exactly not exist because in project management these uncertainties are used as.The uncertainty that the software project cost is measured causes or has increased the weight of a series of concrete problems usually, as: be difficult to measure in early days, be difficult in project assessed cost risk, measurement result be difficult to by the client accept, the rating model range of application is narrow etc.

At present, in the soft project particularly the software project cost measure the field, following problem demanding prompt solution is arranged: 1) how in the software evolution process, particularly early stage software requirement of software process and the cost driving factor all exists under higher probabilistic situation, handles these probabilistic information better in cost is measured; 2) how help project relates to the uncertainties of many correct understanding software project costs, the risk of assessment software project cost over-expense, thus provide better support for software project cost budgeting, plan and cost control.Also there are not a kind of method or system at present, can expand common rating model, as one man modeling of process software project cost and cost drive the uncertainty of factor value, produce the possible probability distribution of software project cost, help the user correctly to be familiar with software project cost and risk thereof.

Summary of the invention

Measure the probabilistic difficulty of processing that faces at the software project cost, and the deficiency of existing software project cost assay method, the present invention has considered that comprehensively existing various costs drive the factor in the software process, proposes a kind of cost assay method and system of software project.Its purpose is that cost in process software project cost rating model clearly and the software process drives the uncertain information of the factor, determine the probability distribution of certain software project cost, and produce the corresponding tables of software project cost budgeting and corresponding cost overrun risk.

Technical scheme of the present invention is summarized as follows:

A kind of software project cost assay method based on process data may further comprise the steps:

1) loads cost and drive factor attribute library, this cost drives in the factor attribute library storing software process various to the influential cost driving of the cost factor, described each cost is driven the factor be described with the attribute of setting, the attribute of described setting is to describe with form of probability;

2) load cost rating model storehouse, the various cost rating models of this cost rating model library storage, the definition cost drives the relation between the factor and attribute and software project cost;

3) item attribute in the history item and cost drive factor data and set up the history item database in the acquisition software process management database, and, obtain the probability distribution of parameter according to the parameter that the cost driving factor data of history item is proofreaied and correct the selected model from cost rating model storehouse;

4) set the probability distribution that each cost of current project drives factor value;

5) cost is driven factor value and sample, manufacturing cost drives factor example, and the attribute that cost in model parameter and the attribute library is driven the factor is sampled, and generates the example of rating model;

6) drive factor example according to rating model example that generates and cost, simulate the software project cost under this sampling experiment, and set sampling experiment number of times, the software project cost emulation experiment of carrying out repetition;

7) result who produces according to software project cost emulation experiment, the cost of mensuration software project, output cost information.

Described step 1) each cost in this attribute library drives the factor and has set 6 attributes, comprises: 1) title; 2) describe, promptly to driving the explanation of factor implication; 3) possible grade promptly drives various grade or state that the factor may value.Cost drives the grade point that the factor may be chosen, and is described by minimum a, most probable and Senior Three value.As: the reliability requirement of software systems can be basic, normal, high; 4) selection rule of grade promptly under which type of situation, is chosen any grade.As: when the operation of software systems may bring huge financial losses, should be " height " to the reliability requirement of these software systems; 5) the possible value of throughput rate coverage, promptly this factor is under maximum and minimum value condition, to the difference of throughput rate influence.As: the software that the low-down software of reliability is very high with respect to reliability, the former software development productivity are 1.6 times of the latter, or the latter requires 60% extra work amount.6) the possible value of throughput rate influence value, promptly this factor is under various value grades, to the influence of throughput rate and cost of development.As: when reliability when being common, do not require the increase cost of development; When value when being high, need to increase by 20% cost of development.(have only in these 6 attributes the 2nd attribute " description " can be default, other all are essential, and attribute can be expanded as required)

Described step 2) the cost rating model is to select from cost rating model storehouse and read, and each rating model in this storehouse has all been decided cost by relational expression and driven getting in touch between the factor and software project cost, and provides the possible value of parameter in the relational expression.

Described step 3) history item attribute comprises following information: project name, item description, application, operation platform, type of architecture, the actual cost of development of software project.

Described step 4) is obtained the probability distribution that each cost drives the value of the factor by dual mode: direct given cost drives the probability distribution of factor value; Perhaps count in the history item database software size and cost in the similar software project and drive the probability distribution of factor value, and as estimated value.

Further, use triangle distribution to describe the probability distribution of software size and cost driving factor value:

Use minimum a, most probable c and three values of the highest b that triangle distribution is described, then probability density function is:

$\left\{\begin{array}{c}\frac{2(x-a)}{(b-a)(c-a)}\mathrm{fora}\≤x\≤c\\ \frac{2(b-x)}{(b-a)(b-c)}\mathrm{forc}\≤x\≤b\end{array}\right.,$ X is an independent variable, shows certain value.

The emulation sampling test that described step 6) uses the emulation of Monte Carlo type to carry out repetition.

The information that described step 7) software project cost is measured comprises: the probability distribution graph of software project cost, descriptive statistic information, " budget/risk " corresponding tables.

A kind of software project cost assay method cost based on process data is measured system, it is characterized in that, comprising: software project cost simulation kernel module is used for the emulation experiment of software project cost, and produces software project cost simulation result;

Cost drives factor attribute library, comprises various the influential cost of cost to be driven the factor, each cost is driven factor be described with the attribute of setting;

The history item database drives the cost that defines in the factor attribute library according to cost and drives the factor, and the corresponding information in the history item is collected and stored;

Cost rating model storehouse, storage is described cost and is driven the rating model that concerns between the factor and software project cost;

Cost drives factor value device, drives the value of the factor by user's Input Software project cost, or produces the value of the cost driving factor according to the history item data analysis;

Software project cost statistics and venture analysis device are used to gather the software project cost measured value that has produced, produce the software project cost and measure statistical information.

Described software project cost simulation kernel module comprises:

Randomizer is sampled and is produced random number;

The software project cost drives factor example maker, calls randomizer and produces the sample instances that cost drives the factor;

Rating model example maker drives the cost driving factor attribute that comprises in the factor attribute library according to cost rating model storehouse and cost, calls the sample instances that randomizer generates software project cost rating model;

Software project cost maker uses the rating model sample instances and the cost that have generated to drive factor sample instances, generates software project cost measured value.

Technique effect of the present invention is: 1) described the cost driving factor and uncertainty thereof in the software process in the mode of standard, thereby can carry out modeling and quantitative analysis to the software project cost; 2) accept software project cost rating model and the cost driving uncertain information that the factor comprised clearly as input, and have probabilistic information with the emulation mode processing; 3) emulation produces possible probability distribution and " budget/risk " corresponding tables of software project cost, has reflected the possible value and the pairing risk of individual budget level of software project cost under uncertain environment; 4) realized the processing of robotization, common software developer can measure software project cost, the uncertainty of understanding software project cost, the pairing risk of the various cost budgeting levels of identification easily by this system.

Description of drawings

Fig. 1 is the framework synoptic diagram that the software project cost is measured system;

Fig. 2 software project cost is measured system's execution in step synoptic diagram;

The implementing procedure figure of Fig. 3 software project cost assay method

Embodiment

Below in conjunction with accompanying drawing, the specific implementation process of measuring by the software project cost is described further invention, but is not construed as limiting the invention.The implementation process that the software project cost is measured comprises following three phases and corresponding work step as shown in Figure 3, wherein the software project cost measure system framework as shown in Figure 1, the software project cost is measured system's execution in step as shown in Figure 2:

(1) initial phase

To finish in the software project cost mensuration system initialization of " cost drives factor attribute library ", " history item database " and " cost rating model storehouse " in this step.

1) cost drives the initialization of factor attribute library

Cost driving factor attribute library has comprised a collection of specification description and (has referred to by 6 attributes that comprised in the factor attribute library each factor is described.These 6 attributes also can according to circumstances be expanded.) cost drive the factor, cost drive the factor be according to sum up in the existing software project process of industry come out to the influential data of cost.Can on the basis of these existing cost driving factors, carry out certain reduction or expansion in the initialization procedure according to features on project that current enterprise develops software.Each cost that cost drives in the factor attribute library drives the factor, has comprised six attributes: title, description, possible grade, the selection rule of grade, the possible value of throughput rate coverage, the possible value of throughput rate influence value.The difference of setting each influence value such as throughput rate such as inter-stage such as grade that drives the factor during the software project cost is measured is equidistant, based on the throughput rate influence value that this setting can produce each grade automatically by the optional grade and the throughput rate coverage (PR) of the factor, its computing formula is as shown in the table:

(represented cost to drive the optional rate range of the factor, as: on behalf of five grades from very low to very high, " very low-very high " can select.)

2) initialization of history item database

After the initialization cost drives factor attribute library, can collect that enterprise has finished the relevant information of history item in the software process management database, and set up the history item database.Each bar history item data in this database can comprise following attribute: project name, and item description, application, operation platform, type of architecture, software size, each cost drives the value of the factor, the actual cost of development of software project.

Data in this history item database will be used for training software project cost rating model.

(as shown in Figure 1 the history item data can: 1) influence cost and drive factor attribute library by statistical study to data; 2) influence model parameter by calibration, thereby influence the rating model storehouse rating model.)

3) initialization in cost rating model storehouse

Cost rating model storehouse has comprised defined cost rating model equation, can be analog model, multivariate regression model, substep regression model etc.In the initialization procedure, can according to circumstances select required rating model, as selecting following regression model:

$\mathrm{PM}=A*{\mathrm{Size}}^B*\underset{i=1}{\overset{k}{\mathrm{\Π}}}{\mathrm{EM}}_i$

Wherein: PM is a workload, is typically expressed as the man month; A is the constant term calibration parameter; Size is the tolerance of the functional dimension of the software module that workload is additive property influence; B is for weighing software development scale economics or uneconomic exponential term calibration parameter; EM is for influence the workload multiplier of workload, is to be determined by the throughput rate influence value that cost drives the selected grade of the factor and this grade.

Use the data in the history item database, can calibrate factors A and B.Can and obtain A, the trained values of B in the enterprising line retrace analysis of historical data by following formula:

$\ln \mathrm{PM}-\ln \left(\underset{i=1}{\overset{k}{\mathrm{\Π}}}{\mathrm{EM}}_i\right)=\ln A+B*\ln \mathrm{Size},$

(2) the emulation execute phase

As shown in Figure 1, at certain mensuration project, can import the cost relevant information of this project, software development cost simulation kernel module will be carried out the automatic simulation of Monte Carlo type to the cost of development of this project afterwards.

1) input of project information

Be the mensuration software development cost, and the project of being measured is carried out record, need give the currentitem purpose with reference to the item attribute in the history item database: project name, item description, application, operation platform, type of architecture, software size, each cost drives the value of the factor.

Wherein, project name, item description, application, operation platform and type of architecture all are the item description information that can directly obtain.

Software size and each cost drive the value of the factor, can use " cost drives factor value device ", and obtain by dual mode: the one, and measure personnel by cost and directly provide probability distribution; The 2nd, drive the possible probability distribution of factor value by software size and cost in the similar software project in the systematic analysis history item database, and as estimated value.

Default use (but being not limited to) triangle distribution is described software size and the possible probability distribution of cost driving factor value in native system.Use minimum a, most probable c and three values of the highest b that triangle distribution is described, then probability density function is:

$\left\{\begin{array}{c}\frac{2(x-a)}{(b-a)(c-a)}\mathrm{fora}\≤x\≤c\\ \frac{2(b-x)}{(b-a)(b-c)}\mathrm{forc}\≤x\≤b\end{array}\right.$

(a, b, three values of c are the descriptions to probability distribution, can obtain by dual mode: 1) can come out by the history item data statistics, be certain numerical value probability distribution in history; 2) can directly be specified by the user, be the result that the expert rule of thumb judges.X is an independent variable, show certain value, and equation shows the pairing probability density of this value.)

Randomizer can be sampled to value according to this density function.

2) emulation automatically performs

Emulation mode of the present invention has been used the emulation of Monte Carlo type, and its ultimate principle is according to the probability distribution of describing in the input information, carries out the sampling experiment of repetition.In system of the present invention, emulation has comprised " generating the rating model example ", " manufacturing cost drives factor example " and " measuring the software project cost " three parts, below describes in detail:

A) generate the rating model example

According in the cost rating model storehouse to the description of cost rating model, and the software project cost drives in the factor attribute library the description of software project cost attribute, emulation produces the example of rating model.At first calling random number generator samples to model parameter according to the possible probability distribution of rating model parameter; The probability distribution that drives possible the value (PR) of factor throughput rate coverage according to cost is sampled to PR afterwards, and derives the cost driving factor at the pairing throughput rate influence value of each value grade; Can set up the example of rating model at last according to sampling results.The input information of rating model example maker comprises: the 1) model equation described in the cost rating model storehouse; 2) cost drives PR value and the throughput rate influence value in the factor attribute library.Wherein, PR value and throughput rate influence value that the parameter of rating model, software project cost drive the factor are comprising uncertainty usually, are the formal descriptions with probability distribution.Rating model example maker will call randomizer, and produce corresponding sample instances according to probability distribution.At last, rating model example maker will be according to model equation, factor attribute sampling example, model parameter sample instances, and the example that produces rating model is as output.

Probability distribution is still obtained by aforementioned dual mode, and the probability density function of rating model parameter is that a default value is arranged, and calibrates according to the history item data; The cost factor value is that equity level is carried out value, and people just choose grade when measuring, and and do not know to choose of the influence of this grade to throughput rate.Cost factor is by the rating model decision to the influence of throughput rate, and finishes corresponding calculating in the mensuration process automatically; The history item data are to determine the rating model example indirectly, and the history item data influence the parameter of rating model or influence cost impact factor attribute library by statistical study by model calibration.

B) manufacturing cost drives factor example

According to the probability distribution of software size given in the input information and cost driving factor value, call random number generator, the value of the software size and the cost driving factor is sampled, thereby generation software size and cost drive the example of the factor.

C) measure the software project cost

Software project cost maker is according to the sampling experiment number of times of setting in the system, repeatedly call " cost drives factor example maker " and " cost rating model example maker ", and, be determined at the software project cost under this sampling experiment according to rating model example that is produced and cost driving factor example.

(3) stage of statistical study as a result

Software project cost statistics and venture analysis device are mainly used in result's statistical study, gather the software project cost measured value that has produced, produce the software project cost and measure statistical information.Statistical information has comprised: provide the probability distribution graph of software project cost in the software project cost simulation result, provide the descriptive statistic information of software project cost probability distribution, generate " budget/risk " corresponding tables.Comprise each software project cost budgeting level and corresponding cost overrun risk.

1) probability distribution graph of software project cost

Software project cost instance number according to each cost interval in the software project cost simulation result is occurred generates the histogram of describing software project cost probability distribution.

2) descriptive statistic information

Including but not limited to the mean value of most probable software project cost, software project cost, the variance of software project cost variation degree has been described.

3) " budget/risk " corresponding tables

Comprise each software project cost budgeting level and corresponding cost overrun risk.Use this corresponding tables, can judge the risk of software project cost overrun under this budget level, also can judge corresponding software project cost budgeting by receptible cost overrun risk level by the software project cost budgeting.

Claims (10)

1, a kind of software project cost assay method based on process data may further comprise the steps:

1) loads cost and drive factor attribute library, this cost drives in the factor attribute library storing software process various to the influential cost driving of the cost factor, described each cost is driven the factor be described with the attribute of setting, the attribute of described setting is to describe with form of probability;

2) load cost rating model storehouse, the various cost rating models of this cost rating model library storage, the definition cost drives the relation between the factor and attribute and software project cost;

3) item attribute in the history item and cost drive factor data and set up the history item database in the acquisition software process management database, and, obtain the probability distribution of parameter according to the parameter that the cost driving factor data of history item is proofreaied and correct the selected model from cost rating model storehouse;

4) set the probability distribution that each cost of current project drives factor value;

5) cost is driven factor value and sample, manufacturing cost drives factor example, and the attribute that cost in model parameter and the attribute library is driven the factor is sampled, and generates the example of rating model;

6) drive factor example according to rating model example that generates and cost, simulate the software project cost under this sampling experiment, and set sampling experiment number of times, the software project cost emulation experiment of carrying out repetition;

7) result who produces according to software project cost emulation experiment, the cost of mensuration software project, output cost information.

2, the method for claim 1, it is characterized in that the cost that described step 1) is set drives factor attribute and comprises: but but title, description, optionally grade, the selection rule of grade, the selected value of throughput rate coverage, the selected value of throughput rate influence value.

3, the method for claim 1 is characterized in that, described step 2) model in the cost rating model storehouse comprises analog model, multivariate regression model, substep regression model.

4, the method for claim 1 is characterized in that, described step 3) history item attribute comprises following information: project name, item description, application, operation platform, type of architecture, the actual cost of development of software project.

5, the method for claim 1 is characterized in that, described step 4) is set the probability distribution that each cost drives the value of the factor by dual mode: direct given cost drives the probability distribution of factor value; Perhaps count in the history item database software size and cost in the similar software project and drive the probability distribution of factor value, and as estimated value.

6, method as claimed in claim 5 is characterized in that, uses triangle distribution to describe the probability distribution of software size and cost driving factor value,

Use minimum a, most probable c and three values of the highest b that triangle distribution is described, then probability density function is:

$\left\{\begin{array}{c}\frac{2(x-a)}{(b-a)(c-a)}\mathrm{fora}\≤x\≤c\\ \frac{2(b-x)}{(b-a)(b-c)}\mathrm{forc}\≤x\≤b\end{array}\right.,$ X is an independent variable, shows certain value.

7, the method for claim 1 is characterized in that, the emulation sampling test that described step 6) uses the emulation of Monte Carlo type to carry out repetition.

8, the method for claim 1 is characterized in that, the cost information of described step 7) output comprises: the probability distribution graph of software project cost, descriptive statistic information, " budget/risk " corresponding tables.

9, a kind of software project cost based on process data is measured system, it is characterized in that, comprising: software project cost simulation kernel module is used for the emulation experiment of software project cost, and produces software project cost simulation result;

Cost drives factor attribute library, comprises various the influential cost of cost to be driven the factor, each cost is driven factor be described with the attribute of setting;

The history item database drives the cost that defines in the factor attribute library according to cost and drives the factor, and the corresponding information in the history item is collected and stored;

Cost rating model storehouse, storage is described cost and is driven the rating model that concerns between the factor and software project cost;

Cost drives factor value device, drives the value of the factor by user's Input Software project cost, or produces the value of the cost driving factor according to the history item data analysis;

Software project cost statistics and venture analysis device are used to gather the software project cost measured value that has produced, produce the software project cost and measure statistical information.

10, system as claimed in claim 9 is characterized in that, described software project cost simulation kernel module comprises:

Randomizer is sampled and is produced random number;

The software project cost drives factor example maker, calls randomizer and produces the sample instances that cost drives the factor;

Rating model example maker drives the cost driving factor attribute that comprises in the factor attribute library according to cost rating model storehouse and cost, calls the sample instances that randomizer generates software project cost rating model;

Software project cost maker uses the rating model sample instances and the cost that have generated to drive factor sample instances, generates software project cost measured value.

Images