CN105469145A - Intelligent test paper method based on genetic particle swarm optimization algorithm - Google Patents

Abstract

The invention relates to an intelligent test paper method based on a genetic particle swarm optimization algorithm, comprising: generating the objective function corresponding to each constraint condition according to the constraint condition corresponding to test paper attribute information, and calculating the fitness function of test paper according to the objective function corresponding to each constraint condition; obtaining test questions from an item bank to form a plurality of pieces of test paper, and performing chromosome coding on each piece of test paper, wherein each piece of test paper corresponds to a chromosome, the chromosome includes a plurality of segments, each segment of chromosome corresponds to a type of test questions, and includes a plurality of genes, and each gene corresponds to a test question; obtaining an initial population through a particle swarm algorithm; and processing the initial population through a genetic algorithm to obtain a new population to output test paper therein. According to the technical scheme, the method employs test paper attribute information as constraint conditions to generate a fitness function, and performs particle swarm algorithm and genetic algorithm treatment on test paper according to the fitness function, thereby obtaining test paper meeting user needs.

Description

A kind of Intelligent Auto-generating Test Paper method based on Genetic Particle Swarm Algorithm

Technical field

The present invention relates to Intelligent Auto-generating Test Paper technical field, in particular to a kind of Intelligent Auto-generating Test Paper method based on Genetic Particle Swarm Algorithm.

Background technology

Traditional group volume operation almost relies on completely and manually completes, or needs semi-artificial carrying out.Artificial group of volume is with high costs and error rate is high, easily occurs careless omission and inevitable human factor, often causes paper group volume scientific and reasonable not.Although semi-artificial group volume part alleviates the labor capacity of staff, the gordian technique work such as selection, layout of examination question still must by manually completing, so still exist, efficiency is lower waits limitation and deficiency.

In recent years, some scientific research institutions of home and abroad have developed some intelligent automatic Test Paper Generation Systems, and employing method mainly contains priority method, random choice method, backtracking trial method, error compensation method etc.Although use these methods can realize automatic volume group to a certain extent, group volume is a multi-objective optimization question, and it is comparatively simple that these methods have; Some group volume methods have larger randomness and uncertainty, are difficult to the actual demand meeting group volume; Some test papers algorithm are large to the occupancy of internal memory, program structure more complicated, and the group volume time is long.The limitation of these methods may cause its paper generated cannot practical requirement.

Summary of the invention

Technical matters to be solved by this invention is, how to improve the intellectuality of group volume, makes to organize the requirement of rolling up and more can meet user.

For this purpose, the present invention proposes a kind of Intelligent Auto-generating Test Paper method based on Genetic Particle Swarm Algorithm, comprising:

S1, the constraint condition according to corresponding to paper attribute information generates objective function corresponding to each constraint condition, calculates the fitness function of paper according to the objective function of each constraint condition;

S2, obtains examination question and forms many parts of initial papers, carry out chromosome coding to every part of paper from exam pool, every part of corresponding chromosome of paper, chromosome comprises multistage, every section of corresponding class examination question of chromosome, every section of chromosome comprises multiple gene, each gene corresponding one examination question;

S3, adopts particle cluster algorithm, using every part of paper as a particle, calculates many parts of papers, to obtain many parts of new papers as initial population;

S4, the fitness value of every part of paper in initial population is calculated according to described fitness function, sort according to fitness value, from multiple fitness value, select fitness value to be greater than the paper heredity of default fitness value to the next generation, to generate first generation population according to default select probability;

S5, in first generation population, by the chromosome in the population selected, match between two at random, crossover probability and mutation probability is obtained respectively according to each chromosomal fitness value in two chromosomes of pairing, maximum adaptation angle value and average fitness value calculation, by described crossover probability and mutation probability, two chromosomes to pairing carry out interlace operation

Described interlace operation comprises:

Using pairing two chromosomal multiple corresponding section at least one section as transposition section, exchange the gene of the corresponding transposition section of two chromosomes, to generate two new chromosome, will be new individual as second generation population to the multiple chromosomes generated after chromosome pairings all in first generation population;

S6, in second generation population, arranges at least one change point in each chromosomal every section of chromosome, obtains the examination question that the examination question identical with this change point type replaces this change point, to generate third generation population as new population from exam pool;

S7, calculates the fitness value of multiple individuality in new population, judge whether new population meets pre-conditioned, if meet, then exports corresponding paper, otherwise, new population is returned step S4 as initial population.

Preferably, described paper attribute information comprises: the score value ratio of paper total score, topic type, Reaction time, degree-of-difficulty factor, discrimination, the distribution proportion comprising the A to Z of point, knowledge point score value ratio, cognitive level, topic type, the score value ratio of each difficulty examination question.

Preferably, the constraint of Distribution of knowledge gists wherein, t is the number ratio that knowledge point that paper relates to accounts for the A to Z of point, and T is the ratio that knowledge point that paper that user sets relates to accounts for the A to Z of point;

Degree-of-difficulty factor retrains wherein, h is the actual degree-of-difficulty factor of paper, and H is the degree-of-difficulty factor that user sets;

Discrimination retrains wherein, d is paper actual zone calibration, and D is the discrimination that user sets;

Reaction time retrains wherein, t is the actual Reaction time of paper, and T is the Reaction time that user sets, T _i≤ T;

Cognitive level constraint $f_5=\frac{\underset{i=1}{\overset{n}{\Σ}}\left|\frac{R_i-L_i}{L_i}\right|}{n},$ Wherein, $R_i=\underset{j=1}{\overset{m}{\Σ}}{r}_{j,i},\left|\frac{R_i-L_i}{L_i}\right|$ And f ₆be less than or equal to 1, R _irepresent the actual total score of examination question of the cognitive level of the i-th class in paper, L _irepresent the examination question total score of the cognitive level of the i-th class of user's setting, n represents the number of cognitive level, the number of M to be cognitive level be whole examination questions of i, r _j,ibe the examination question mark of the jth of i for cognitive level, represent that cognitive level is the total score of whole examination questions of i;

The score value ratio constraint of topic type $f_6=\frac{\underset{i=1}{\overset{n}{\Σ}}\left|\frac{K_i-L_i}{L_i}\right|}{n},$ Wherein, $K_i=\underset{j=1}{\overset{m}{\Σ}}{k}_{j,i},\left|\frac{K_i-L_i}{L_i}\right|$ And f ₇be less than or equal to 1, K _irepresent the actual total score of examination question of the i-th class examination question type in paper, L _irepresent the examination question total score of the i-th class examination question type of user's setting, n represents the number of examination question type.The number of M to be examination question type be whole examination questions of i, k _j,ibe the examination question mark of the jth of i for examination question type, represent that examination question type is the total score of whole examination questions of i;

The score value ratio constraint of each difficulty topic type wherein, and f ₈be less than or equal to 1, H _irepresent the actual total score of examination question of the i-th class examination question type in paper, L _irepresent the examination question total score of the i-th class examination question type of user's setting, n represents the number of examination question type, the number of M to be examination question type be whole examination questions of i, h _j,ibe the examination question mark of the jth of i for examination question type, represent that examination question type is the total score of whole examination questions of i.

Preferably, described step S1 comprises:

To institute's Constrained weighted sum, $f=\underset{i=1}{\overset{n}{\Σ}}{w}_i{f}_i,i=1,\mathrm{...},7,$

The target generic function of f expression group volume, w _irepresent the weights of i-th index, w _i> 0, w _if _irepresent the objective function of i-th paper attribute information constraint.

Preferably, described step S1 comprises:

Generate fitness function F=1/ (1+f).

Preferably, described step S3 comprises:

S31: the speed v of each particle in initialization population m ₁, position x ₁, population size, be vector x to the position of arbitrary particle i and dimension s, particle i _i=(x _i1, x _i2..., x _is), pace of change is v _i=(v _i1, v _i2..., v _is).Particle i is at position range [-x _max, x _max] in obey be uniformly distributed produce x _is, to arbitrary particle i and dimension s, at velocity range [-v _max, v _max] in obey be uniformly distributed produce v _is, wherein, m is the quantity of paper in population, and s equals the quantity of constraint condition, x _maxand v _maxfor the position that presets and velocity amplitude;

S32: each particle according to initialized position and velocity variations, by the x of particle i _isubstitute into fitness function and calculate corresponding fitness value, at f ₁middle x _ifor t, at f ₂middle x _ifor h, at f ₃middle x _ifor d, at f ₄middle x _ifor t, at f ₅middle x _ifor R _i, at f ₆middle x _ifor K _i, at f ₇middle x _ifor H _i, calculate the fitness value of each particle in change procedure, the optimal location P of each particle in storage change process _bestand the fitness value of correspondence, using the optimal location G of the position of particle maximum for fitness value in population in change procedure as population _bestif, P _best=(p _i1, p _i2..., p _is), G _best=(g _i1, g _i2..., g _is), p _isfor the value of s constraint in particle i change procedure, g _isfor the value of s constraint in particle i change procedure;

S33: the optimal location P that the fitness value of each particle is lived through with it _bestcorresponding fitness value compares, if comparatively large, then using position corresponding for this fitness value as current optimal location;

S34: by the optimal location G of population in change procedure before the fitness value of each particle and its _bestcorresponding fitness value compares, if comparatively large, then using position corresponding for this fitness value as current population optimal location;

S35: judge whether change frequency is greater than preset times, or whether meet: the fitness value of current each particle equals the P in previous change _bestcorresponding fitness value, and the maximum adaptation angle value in current population equals the G in previous change _bestcorresponding fitness value, if change frequency reaches preset times or meets above-mentioned condition, then will obtain many parts of new papers as initial population, otherwise returns step S33.

Preferably, described step S32 comprises:

In the t time change of i-th particle, the position of the n-th dimension is speed is distance between i-th particle current location and its optimal location lived through is i-th particle current location and before it in change procedure population optimum position between distance be the position of i-th particle, the t+1 time change speed as follows:

$v_{in}^{t+1}={\mathrm{wv}}_{in}^t+c_1{r}_1(p_{in}^t-x_{in}^t)+c_2{r}_2(p_{gn}^t-x_{in}^t);$

$x_{in}^{t+1}=x_{in}^t+v_{in}^{t+1};$

Wherein 1≤i≤m, 1≤n≤s, c1, c2 are the nonnegative constant that user sets, r ₁and r ₂for the random number between [0,1] of user's setting, w is inertia weight coefficient, represent the speed of i-th particle, n-th dimension in the t time iteration, as the acceleration of i-th particle in the t+1 time change procedure, as the acceleration of colony in the t+1 time change procedure.

Preferably, described step S5 comprises:

Each chromosomal crossover probability P is obtained respectively according to each chromosomal fitness value in two chromosomes of pairing, maximum adaptation angle value and average fitness value calculation _cwith mutation probability P _m, wherein,

$P_c=\left\{\begin{array}{cc}{c}_1\frac{(f_{\max }-f_c)}{f_{\max }-f_{avg}}& f_c\&GreaterEqual;f_{avg}\\ c_2& f_c<f_{avg}\end{array}\right.,P_m=\left\{\begin{array}{cc}{k}_1\frac{(f_{\max }-f_m)}{f_{\max }-f_{avg}}& f_m\&GreaterEqual;f_{avg}\\ k_2& f_m<f_{avg}\end{array}\right.,$

F _cfor participating in the larger fitness in the fitness of the two parts of papers intersected, f _mfor the individual fitness value that makes a variation, f _maxfor the maximum adaptation angle value in population, f _avgfor the average fitness value of population, c ₁, c ₂, k ₁, k ₂be user set between (0,1] constant.

Preferably, described pre-conditioned comprise following at least one or its combination:

User's initial conditions terminates heredity;

In maximum adaptation degree in current population and former generation population, fitness difference is less than preset value;

Reach default evolutionary generation.

According to technique scheme, fitness function can be generated using the attribute information of paper as constraint condition, and according to fitness function, particle cluster algorithm and genetic algorithm process be carried out to paper, thus be met the paper of user's needs.

Accompanying drawing explanation

Can understanding the features and advantages of the present invention clearly by reference to accompanying drawing, accompanying drawing is schematic and should not be construed as and carry out any restriction to the present invention, in the accompanying drawings:

Fig. 1 shows according to an embodiment of the invention based on the schematic flow diagram of the Intelligent Auto-generating Test Paper method of Genetic Particle Swarm Algorithm;

Fig. 2 shows the schematic flow diagram of particle cluster algorithm according to an embodiment of the invention.

Embodiment

Can more clearly understand above-mentioned purpose of the present invention, feature and advantage, below in conjunction with the drawings and specific embodiments, the present invention is further described in detail.It should be noted that, when not conflicting, the feature in the embodiment of the application and embodiment can combine mutually.

Set forth a lot of detail in the following description so that fully understand the present invention; but; the present invention can also adopt other to be different from other modes described here and implement, and therefore, protection scope of the present invention is not by the restriction of following public specific embodiment.

As shown in Figure 1, according to an embodiment of the invention based on the Intelligent Auto-generating Test Paper method of Genetic Particle Swarm Algorithm, comprising:

S1, the constraint condition according to corresponding to paper attribute information generates objective function corresponding to each constraint condition, calculates the fitness function of paper according to the objective function of each constraint condition;

S2, obtains examination question and forms many parts of initial papers, carry out chromosome coding to every part of paper from exam pool, every part of corresponding chromosome of paper, chromosome comprises multistage, every section of corresponding class examination question of chromosome, every section of chromosome comprises multiple gene, each gene corresponding one examination question;

S3, adopts particle cluster algorithm, using every part of paper as a particle, calculates many parts of papers, to obtain many parts of new papers as initial population;

S4, the fitness value of every part of paper in initial population is calculated according to fitness function, sort according to fitness value, from multiple fitness value, select fitness value to be greater than the paper heredity of default fitness value to the next generation, to generate first generation population according to default select probability;

S5, in first generation population, by the chromosome in the population selected, match between two at random, crossover probability and mutation probability is obtained respectively according to each chromosomal fitness value in two chromosomes of pairing, maximum adaptation angle value and average fitness value calculation, by crossover probability and mutation probability, two chromosomes to pairing carry out interlace operation

Interlace operation comprises:

Using pairing two chromosomal multiple corresponding section at least one section as transposition section, exchange the gene of the corresponding transposition section of two chromosomes, to generate two new chromosome, will be new individual as second generation population to the multiple chromosomes generated after chromosome pairings all in first generation population;

S6, in second generation population, arranges at least one change point in each chromosomal every section of chromosome, obtains the examination question that the examination question identical with this change point type replaces this change point, to generate third generation population as new population from exam pool;

S7, calculates the fitness value of multiple individuality in new population, judge whether new population meets pre-conditioned, if meet, then exports corresponding paper, otherwise, new population is returned step S4 as initial population.

The present embodiment is before carrying out genetic algorithm, first by particle cluster algorithm, iterative processing is carried out to paper group, the quality drawing paper can be evaluated by fitness value, it is more simple relative to genetic algorithm rule, " intersection " and " variation " in genetic algorithm is not had to operate, by follow current search to optimal value find global optimum, initial population can be drawn rapidly.Further genetic algorithm process is carried out to initial population, can realize restraining quickly.

The present embodiment neatly using every part of paper as a chromosome, every problem is participated in the computing of genetic algorithm as the gene of on chromosome, thus can be calculated the many parts of papers as a colony by genetic algorithm, according to the principle of the survival of the fittest and the survival of the fittest, develop by generation and produce the approximate solution of becoming better and better, in every generation, select individual according to fitness size individual in Problem Areas, and carry out the operation of combination crossover and mutation by means of the genetic operator of natural genetics, produce the population representing new disaggregation, the same rear life of the worm images of a group of characters natural evolution of many parts of paper compositions is made more to be adapted to environment (more meeting user to require) for population than former generation population, optimum individual in last reign of a dynasty population is through decoding, namely can as problem approximate optimal solution, also the paper namely needed for user exports.

Preferably, paper attribute information comprises: paper total score (such as 100 points), topic type (such as comprises topic of filling a vacancy, multiple-choice question, letter answer, True-False, calculation question, proof question, synthesis problem), Reaction time (such as 90 minutes), degree-of-difficulty factor (the complexity index of paper, generally can be arranged on (0-1] between), (index of candidate ability level height distinguished by paper to discrimination, generally at (0-1] between), comprise the ratio (such as covering 80% of required investigation knowledge point) of the A to Z of point, (total score of the corresponding examination question in such as single knowledge point is no more than at most 15% of whole paper total score to knowledge point score value distribution proportion, ), cognitive level (such as requires that the topic remembered accounts for 20%, require that the topic understood accounts for 30%, require that the topic of application accounts for 40%, require that the topic of integrated application accounts for 10%), (such as multiple-choice question accounts for 20% to the score value ratio of topic type, topic of filling a vacancy accounts for 20%, calculation question accounts for 20%, proof question accounts for 30%, synthesis problem accounts for 10%), (such as easily topic accounts for 40% to the score value ratio of each difficulty topic type, medium topic accounts for 40%, a difficult problem accounts for 20%).

The multiple attribute information as constraint condition that the present embodiment provides can meet the group volume needs of user to the full extent.In addition, in step s 2, from exam pool, obtain examination question when forming many parts of papers, need to judge that paper is the need of satisfied 2 conditions: one, whether examination question total score meets preset requirement, and such as paper total score is 100 points; Its two, the maximal value of the single knowledge point mark that examination question is corresponding meets preset requirement, and the such as score value of single knowledge point is no more than 15% of the total score value of paper.

Preferably, the constraint of Distribution of knowledge gists wherein, t is the number ratio that knowledge point that paper relates to accounts for the A to Z of point, and T is the ratio that knowledge point that paper that user sets relates to accounts for the A to Z of point;

Degree-of-difficulty factor retrains wherein, h is the actual degree-of-difficulty factor of paper, and H is the degree-of-difficulty factor that user sets;

Discrimination retrains wherein, d is paper actual zone calibration, and D is the discrimination that user sets;

Reaction time retrains wherein, t is the actual Reaction time of paper, and T is the Reaction time that user sets, T _i≤ T;

Cognitive level constraint $f_5=\frac{\underset{i=1}{\overset{n}{\&Sigma;}}\left|\frac{R_i-L_i}{L_i}\right|}{n},$ Wherein, $R_i=\underset{j=1}{\overset{m}{\&Sigma;}}{r}_{j,i},\left|\frac{R_i-L_i}{L_i}\right|$ And f ₆be less than or equal to 1, R _irepresent the actual total score of examination question of the cognitive level of the i-th class in paper, L _irepresent the examination question total score of the cognitive level of the i-th class of user's setting, n represents the number of cognitive level, the number of M to be cognitive level be whole examination questions of i, r _j,ibe the examination question mark of the jth of i for cognitive level, represent that cognitive level is the total score of whole examination questions of i;

The score value ratio constraint of topic type $f_6=\frac{\underset{i=1}{\overset{n}{\&Sigma;}}\left|\frac{K_i-L_i}{L_i}\right|}{n},$ Wherein, $K_i=\underset{j=1}{\overset{m}{\&Sigma;}}{k}_{j,i},\left|\frac{K_i-L_i}{L_i}\right|$ And f ₇be less than or equal to 1, K _irepresent the actual total score of examination question of the i-th class examination question type in paper, L _irepresent the examination question total score of the i-th class examination question type of user's setting, n represents the number of examination question type.The number of M to be examination question type be whole examination questions of i, k _j,ibe the examination question mark of the jth of i for examination question type, represent that examination question type is the total score of whole examination questions of i;

The score value ratio constraint of each difficulty topic type wherein, and f ₈be less than or equal to 1, H _irepresent the actual total score of examination question of the i-th class examination question type in paper, L _irepresent the examination question total score of the i-th class examination question type of user's setting, n represents the number of examination question type, the number of M to be examination question type be whole examination questions of i, h _j,ibe the examination question mark of the jth of i for examination question type, represent that examination question type is the total score of whole examination questions of i.

Preferably, step S1 comprises:

To institute's Constrained weighted sum, $f=\underset{i=1}{\overset{n}{\&Sigma;}}{w}_i{f}_i,i=1,\mathrm{...},7,$

The target generic function of f expression group volume, w _irepresent the weights of i-th index, w _i> 0, w _if _irepresent the objective function of i-th paper attribute information constraint.

Preferably, step S1 comprises:

Generate fitness function F=1/ (1+f).

As shown in Figure 2, preferably, step S3 comprises:

S31: the speed v of each particle in initialization population m ₁, position x ₁, population size, be x to the position of arbitrary particle i and dimension s, particle i _i=(x _i1, x _i2..., x _is), pace of change is v _i=(v _i1, v _i2..., v _is).Particle is at position range [-x _max, x _max] in obey be uniformly distributed produce x _is, to arbitrary particle i and dimension s, at velocity range [-v _max, v _max] in obey be uniformly distributed produce v _is, wherein, m is the quantity of paper in population, and s equals the quantity of constraint condition, x _maxand v _maxfor the position that presets and velocity amplitude;

S32: each particle according to initialized position and velocity variations, by the x of particle i _isubstitute into fitness function and calculate corresponding fitness value, at f ₁middle x _ifor t, at f ₂middle x _ifor h, at f ₃middle x _ifor d, at f ₄middle x _ifor t, at f ₅middle x _ifor R _i, at f ₆middle x _ifor K _i, at f ₇middle x _ifor H _i.Calculate the fitness value of each particle in change procedure, the optimal location P of each particle in storage change process _bestand the fitness value of correspondence, using the optimal location G of the position of particle maximum for fitness value in population in change procedure as population _bestif, P _best=(p _i1, p _i2..., p _is), G _best=(g _i1, g _i2..., g _is), p _isfor the value of s constraint in particle i change procedure, g _sfor the value of s constraint in particle i change procedure;

S33: the optimal location P that the fitness value of each particle is lived through with it _bestcorresponding fitness value compares, if comparatively large, then using position corresponding for this fitness value as current optimal location;

S34: by the optimal location G of population in change procedure before the fitness value of each particle and its _bestcorresponding fitness value compares, if comparatively large, then using position corresponding for this fitness value as current population optimal location;

S35: judge whether change frequency is greater than preset times, or whether meet: the fitness value of current each particle equals the P in previous change _bestcorresponding fitness value, and the maximum adaptation angle value in current population equals the G in previous change _bestcorresponding fitness value, if change frequency reaches preset times or meets above-mentioned condition, then will obtain many parts of new papers as initial population, otherwise returns step S33.

In the present embodiment, the predation of particle cluster algorithm simulation flock of birds.Imagine such scene: bevy is at random search food, one piece of food is only had in this region, all birds do not know that food there, but they know how far current position also has from food, so finds the optimal strategy of food to be exactly search the peripheral region of the current bird nearest from food.

In particle cluster algorithm, the solution of every part of paper is all a bird in search volume, is referred to as particle.All particles have a fitness value, the direction that each particle also has a speed to determine to circle in the air and distance, and then particles are just followed current optimal particle and searched in solution space.

Paper is initialized as a group random particles by particle cluster algorithm, then finds optimum solution by iteration (change).In each iteration, particle upgrades self by following the tracks of two extreme values, first optimum solution P being exactly particle itself and finding _best; Another extreme value is the optimum solution G that whole population is found at present _best.

Compared with genetic algorithm, in particle cluster algorithm, the shared mechanism of information is different. in genetic algorithm, chromosome shares information mutually, so the movement of whole population to be moved to optimal region more uniformly, in particle cluster algorithm, only has P _bestor G _besttransmission information is to other particle, and this is unidirectional information flow, and whole search renewal process is the process of following current optimum solution, compares with genetic algorithm, and at most of conditions, all particles may converge on optimum solution faster.

Process initial population further by genetic algorithm, genetic algorithm can be searched for from the trail of solution (required paper), instead of from single solution.And traditional optimized algorithm asks optimum solution from single initial value iteration, be easily strayed into locally optimal solution.And genetic algorithm is searched for from trail, wide coverage, is beneficial to the overall situation preferentially.

In addition, genetic algorithm can process the multiple individualities in colony simultaneously, namely assesses the multiple solutions in search volume, decreases the risk being absorbed in locally optimal solution, and algorithm itself is easy to realize parallelization simultaneously.

Preferably, step S32 comprises:

In the t time change of i-th particle, the position of the n-th dimension is speed is i-th particle current location and and its optimal location lived through between distance be i-th particle current location and before it in change procedure population optimum position between distance be the position of i-th particle, the t+1 time change speed as follows:

$v_{in}^{t+1}={\mathrm{wv}}_{in}^t+c_1{r}_1(p_{in}^t-x_{in}^t)+c_2{r}_2(p_{gn}^t-x_{in}^t);$

$x_{in}^{t+1}=x_{in}^t+v_{in}^{t+1},$

Wherein 1≤i≤m, 1≤n≤s, c1, c2 are the nonnegative constant that user sets, r ₁and r ₂for the random number between [0,1] of user's setting, w is inertia weight coefficient, represent the speed of i-th particle, n-th dimension in the t time iteration, as the acceleration of i-th particle in the t+1 time change procedure, as the acceleration of colony in the t+1 time change procedure.

Preferably, step S5 comprises:

Each chromosomal crossover probability P is obtained respectively according to each chromosomal fitness value in two chromosomes of pairing, maximum adaptation angle value and average fitness value calculation _cwith mutation probability P _m, wherein,

$P_c=\left\{\begin{array}{cc}{c}_1\frac{(f_{\max }-f_c)}{f_{\max }-f_{avg}}& f_c\&GreaterEqual;f_{avg}\\ c_2& f_c<f_{avg}\end{array}\right.,P_m=\left\{\begin{array}{cc}{k}_1\frac{(f_{\max }-f_m)}{f_{\max }-f_{avg}}& f_m\&GreaterEqual;f_{avg}\\ k_2& f_m<f_{avg}\end{array}\right.,$

F _cfor participating in the larger fitness in the fitness of the two parts of papers intersected, f _mfor the individual fitness value that makes a variation, f _maxfor the maximum adaptation angle value in population, f _avgfor the average fitness value of population, c ₁, c ₂, k ₁, k ₂be user set between (0,1] constant.

Pass through the present embodiment, the individuality that fitness is high can be made, select lower crossover probability and mutation probability, thus this individuality is more easily retained and enters the next generation, and the individuality making fitness low, select higher crossover probability and mutation probability, thus this individuality is more easily eliminated, and not easily remain into the next generation, therefore can ensure that rear life is for the needs of the paper in population closer to user during evolution.

Preferably, pre-conditioned comprise following at least one or its combination:

User's initial conditions terminates heredity;

In maximum adaptation degree in current population and former generation population, fitness difference is less than preset value;

Reach default evolutionary generation.

Although describe embodiments of the present invention by reference to the accompanying drawings, but those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, such amendment and modification all fall into by within claims limited range.

Claims (9)

1., based on an Intelligent Auto-generating Test Paper method for Genetic Particle Swarm Algorithm, it is characterized in that, comprising:

S1, the constraint condition according to corresponding to paper attribute information generates objective function corresponding to each constraint condition, calculates the fitness function of paper according to the objective function of each constraint condition;

S2, obtains examination question and forms many parts of initial papers, carry out chromosome coding to every part of paper from exam pool, every part of corresponding chromosome of paper, chromosome comprises multistage, every section of corresponding class examination question of chromosome, every section of chromosome comprises multiple gene, and each gene pairs should one examination question;

S3, adopts particle cluster algorithm, using every part of paper as a particle, calculates many parts of papers, to obtain many parts of new papers as initial population;

S4, the fitness value of every part of paper in initial population is calculated according to described fitness function, sort according to fitness value, from multiple fitness value, select fitness value to be greater than the paper heredity of default fitness value to the next generation, to generate first generation population according to default select probability;

S5, in first generation population, by the chromosome in the population selected, match between two at random, crossover probability and mutation probability is obtained respectively according to each chromosomal fitness value in two chromosomes of pairing, maximum adaptation angle value and average fitness value calculation, by described crossover probability and mutation probability, two chromosomes to pairing carry out interlace operation

Described interlace operation comprises:

Using pairing two chromosomal multiple corresponding section at least one section as transposition section, exchange the gene of the corresponding transposition section of two chromosomes, to generate two new chromosome, will be new individual as second generation population to the multiple chromosomes generated after chromosome pairings all in first generation population;

S6, in second generation population, arranges at least one change point in each chromosomal every section of chromosome, obtains the examination question that the examination question identical with this change point type replaces this change point, to generate third generation population as new population from exam pool;

S7, calculates the fitness value of multiple individuality in new population, judge whether new population meets pre-conditioned, if meet, then exports corresponding paper, otherwise, new population is returned step S4 as initial population.

2. method according to claim 1, it is characterized in that, described paper attribute information comprises: the score value ratio of paper total score, topic type, Reaction time, degree-of-difficulty factor, discrimination, the distribution proportion comprising the A to Z of point, knowledge point score value ratio, cognitive level, topic type, the score value ratio of each difficulty examination question.

3. method according to claim 2, is characterized in that,

The constraint of Distribution of knowledge gists wherein, t is the number ratio that knowledge point that paper relates to accounts for the A to Z of point, and T is the ratio that knowledge point that paper that user sets relates to accounts for the A to Z of point;

Degree-of-difficulty factor retrains wherein, h is the actual degree-of-difficulty factor of paper, and H is the degree-of-difficulty factor that user sets;

Discrimination retrains wherein, d is paper actual zone calibration, and D is the discrimination that user sets;

Reaction time retrains wherein, t is the actual Reaction time of paper, and T is the Reaction time that user sets, t≤T;

Cognitive level constraint $f_5=\frac{\underset{i=1}{\overset{n}{\&Sigma;}}\left|\frac{R_i-L_i}{L_i}\right|}{n},$ Wherein, $R_i=\underset{j=1}{\overset{m}{\&Sigma;}}{r}_{j,i},$ and f ₆be less than or equal to 1, R _irepresent the actual total score of examination question of the cognitive level of the i-th class in paper, L _irepresent the examination question total score of the cognitive level of the i-th class of user's setting, n represents the number of cognitive level, the number of M to be cognitive level be whole examination questions of i, r _j,ibe the examination question mark of the jth of i for cognitive level, represent that cognitive level is the total score of whole examination questions of i;

The score value ratio constraint of topic type $f_6=\frac{\underset{i=1}{\overset{n}{\&Sigma;}}\left|\frac{K_i-L_i}{L_i}\right|}{n},$ Wherein, $K_i=\underset{j=1}{\overset{m}{\&Sigma;}}{k}_{j,i},$ and f ₇be less than or equal to 1, K _irepresent the actual total score of examination question of the i-th class examination question type in paper, L _irepresent the examination question total score of the i-th class examination question type of user's setting, n represents the number of examination question type.The number of M to be examination question type be whole examination questions of i, k _j,ibe the examination question mark of the jth of i for examination question type, represent that examination question type is the total score of whole examination questions of i;

The score value ratio constraint of each difficulty topic type wherein, and f ₈be less than or equal to 1, H _irepresent the actual total score of examination question of the i-th class examination question type in paper, L _irepresent the examination question total score of the i-th class examination question type of user's setting, n represents the number of examination question type, the number of M to be examination question type be whole examination questions of i, h _j,ibe the examination question mark of the jth of i for examination question type, represent that examination question type is the total score of whole examination questions of i.

4. method according to claim 3, is characterized in that, described step S1 comprises:

To institute's Constrained weighted sum, i=1 ..., 7,

The target generic function of f expression group volume, w _irepresent the weights of i-th index, w _i> 0, w _if _irepresent the objective function of i-th paper attribute information constraint.

5. method according to claim 4, is characterized in that, described step S1 comprises: generate fitness function F=1/ (1+f).

6. method according to claim 5, is characterized in that, described step S3 comprises:

S31: the speed v of each particle in initialization population m ₁, position x ₁, population size, be x to the position of arbitrary particle i and dimension s, particle i _i=(x _i1, x _i2..., x _is), pace of change is v _i=(v _i1, v _i2..., v _is), particle is at position range [-x _max, x _max] in obey be uniformly distributed produce x _is, to arbitrary particle i and dimension s, at velocity range [-v _max, v _max] in obey be uniformly distributed produce v ⁱ _s, wherein m is the quantity of paper in population, and s equals the quantity of constraint condition, x _maxand v _maxfor the position that presets and velocity amplitude;

S32: each particle according to initialized position and velocity variations, by the x of particle i _isubstitute into fitness function and calculate corresponding fitness value, at f ₁middle x _ifor t, at f ₂middle x _ifor h, at f ₃middle x _ifor d, at f ₄middle x _ifor t, at f ₅middle x _ifor R _i, at f ₆middle x _ifor K _i, at f ₇middle x _ifor H _i, calculate the fitness value of each particle in change procedure, the optimal location P of each particle in storage change process _bestand the fitness value of correspondence, using the optimal location G of the position of particle maximum for fitness value in population in change procedure as population _bestif, P _best=(p _i1, p _i2..., p _is), G _best=(g _i1, g _i2..., g _is), p _isfor the value of s constraint in particle i change procedure, g _isfor the value of s constraint in particle i change procedure;

S33: the optimal location P that the fitness value of each particle is lived through with it _bestcorresponding fitness value compares, if comparatively large, then using position corresponding for this fitness value as current optimal location;

S34: by the optimal location G of population in change procedure before the fitness value of each particle and its _bestcorresponding fitness value compares, if comparatively large, then using position corresponding for this fitness value as current population optimal location;

S35: judge whether change frequency is greater than preset times, or whether meet: the fitness value of current each particle equals the P in previous change _bestcorresponding fitness value, and the maximum adaptation angle value in current population equals the G in previous change _bestcorresponding fitness value, if change frequency reaches preset times or meets above-mentioned condition, then will obtain many parts of new papers as initial population, otherwise returns step S33.

7. method according to claim 6, is characterized in that, described step S32 comprises:

In the t time change of i-th particle, the position of the n-th dimension is speed is i-th particle current location and and its optimal location lived through between distance be i-th particle current location and before it in change procedure population optimum position between distance be the position of i-th particle, the t+1 time change speed as follows:

$v_{in}^{t+1}={\mathrm{wv}}_{in}^t+c_1{r}_1(p_{in}^t-x_{in}^t)+c_2{r}_2(p_{gn}^t-x_{in}^t);$

$x_{in}^{t+1}=x_{in}^t+v_{in}^{t+1};$

Wherein 1≤i≤m, 1≤n≤s, c1, c2 are the nonnegative constant that user sets, r ₁and r ₂for the random number between [0,1] of user's setting, w is inertia weight coefficient, represent the speed of i-th particle, n-th dimension in the t time iteration, as the acceleration of i-th particle in the t+1 time change procedure, as the acceleration of colony in the t+1 time change procedure.

8. method according to claim 7, is characterized in that, described step S5 comprises:

Each chromosomal crossover probability P is obtained respectively according to each chromosomal fitness value in two chromosomes of pairing, maximum adaptation angle value and average fitness value calculation _cwith mutation probability P _m, wherein,

$P_c=\left\{\begin{array}{cc}{c}_1\frac{(f_{\max }-f_c)}{f_{\max }-f_{avg}}& f_c\&GreaterEqual;f_{avg}\\ c_2& f_c<f_{avg}\end{array}\right.,P_m=\left\{\begin{array}{cc}{k}_1\frac{(f_{\max }-f_m)}{f_{\max }-f_{avg}}& f_m\&GreaterEqual;f_{avg}\\ k_2& f_m<f_{avg}\end{array}\right.,$

F _cfor participating in the larger fitness in the fitness of the two parts of papers intersected, f _mfor the individual fitness value that makes a variation, f _maxfor the maximum adaptation angle value in population, f _avgfor the average fitness value of population, c ₁, c ₂, k ₁, k ₂be user set between (0,1] constant.

9. method according to claim 8, is characterized in that, described pre-conditioned comprise following at least one or its combination:

User's initial conditions terminates heredity;

In maximum adaptation degree in current population and former generation population, fitness difference is less than preset value;

Reach default evolutionary generation.