TW200823776A - Case-based reasoning and learning method and the device for implementing the method - Google Patents

Abstract

The present invention relates to a case-based reasoning and learning method and the device for implementing the method in order to obtain a new solution of a new problem, including the steps: analyzing a whole cases in a database and modifying plural similar parameters and plural adapting parameters accordingly; performing a recall process so as to retrieve at least a similar case similar to the new problem from the database based on the plural similar parameters; and performing an adapting process so as to adapt the at least a similar case to the solution of the new problem.

Description

200823776 IX. Invention: [Technical field of invention] This case proposes a learning reasoning method applied in the field of artificial intelligence, especially a case-based learning reasoning method based on case and its implementation device [prior art] in artificial intelligence In the field of (Artificial Intelligence, AI), Case-Based Reasoning (CBR) is a method to solve existing problems through past experience, and is also a method closer to human learning knowledge: CBR method is the past Experience is stored as knowledge in the case base. 'When a new problem needs to be solved, similar cases can be obtained from the case base, and the information carried in these similar cases can be used to solve new problems from the case base. The process of finding the most similar case is called recall. ' Just like a person's _ to a question, it will immediately be the most similar to the previous one. _ The whole process can solve the new question of self-referencing, called adapt. Just as humans have used past experience in new problems. ...Γίΐ第图图' is the basic structure of CBR. ========================================================================================== The procedure is to compare the differences between the new cases and the similar cases, and then correct the most similar _ Qiao ^ = in the library answer - solution). If this new answer is additionally detailed: become the recognition of the new solution expert, 贞 存 _ _ _ _ _ 丨 = correction or volume, with the number of cases in the case library continues to accumulate ^ 200823776

Sit looks like a higher _, will

Cases that can be used in the case library, so the learning mechanism of case-based reasoning mainly relies on the _quantity (four) product to achieve a mechanism of learning. Know the CBR map. Refer to the section "®, which is the detailed structure of the CBR machine. The Γλ% feed-step subdivision is 5Uindex>·, and the seiect phase is obtained. The index stage is important to consider which parameters or nature of tea or sex f in the new case and the cases that have been encountered in the past. The acquisition stage is to confirm that "two, the case needs to be proposed for advancement - step-by-step procedure: the order of the cases selected in the answer page is selected to select the closest one or a few pens to adjust the order. It can be further subdivided into the repair and odi^b segment and the evaluation phase. The modification phase will be modified after the comparison of the selected case with the new case, and evaluate the case of • MJ. Whether it is appropriate compared with the new case. Eight to +, the algorithm of the CBR mechanism in the recall procedure can be roughly divided into two categories: r method and similarity algorithm. Among them, the classification tree algorithm will be wood All cases of the library are classified according to the importance of the parameters or the nature of the property. 'When a new case is encountered, the relevant parameters or properties are established; index, to find the case in the case library. The similarity algorithm is as follows The similarity shown is mainly calculated as: Σϋπ (1) ', medium Wl is the weight of the parent parameter, and the similarity equation is the same as the new case (inputted case) and capture Case 200823776 (retrievedcases After each parameter value is obtained, the weight of each parameter is multiplied by the sum of the equations of each parameter, and the sum of the weights of the parameters of the silky term is used to obtain the similarity between the new case and the case. When the case is the most important of her, the CBR mechanism will be adapted to the procedures, and the selected cases will be appropriately adjusted to provide answers to the new cases. There are many methods, and the adjustment methods commonly used in the adjustment procedure are large, and the age is _: struetural adaptation and derivative adjustment (d_t_l adaptati〇n). Structural adjustment is the established adjustment method. Or the rule can be directly used in the case of the recall program, and the new answer can be obtained quickly. The derivative adaptation is to redo the process of the new case to the answer of the case--the new answer, so If the Tszaki side shot method requires a miscellaneous A_social (four) case of the library towel all the cases to generate a solution to the process of storage 'also requires a detailed description of the job, also (4) day ^•. This simple adjustment method directly in the m In the case of the excavation, it is more time-saving', but the adjustment method used will change with the increase of the number of cases in the case library to improve the rationality and accuracy of the adjustment results. The learning method is also reviewed here. Aamodt (1994) proposed the 4R learning cycle (retrieve_ reuse_ review-retainleamingcycle), which is to obtain the new problem from the case library. Call the most similar case, use the information in this most similar case to solve the new problem, and use other more detailed methods to revise the new case and save the new case in (a) ain) case library. In the future, when it comes to marriage, it will become a case that can be used in the 7 200823776 case library. In the study of CBR learning methods, there is also a hybrid architecture that combines other artificial intelligence-learning mechanisms. For example, c_neetaL (1999) establishes a mixed law structure that combines (mle induction) and CBR, and uses the weight of the law of the case county riding. Malek (2000) combines the inverse transfer method (backpr〇pagati〇n) of CBR with a neural network (such as Xunca ne^n) to correct the weight of similarity calculation. Juell and Paulson (2003) using reinforcement learning in the CBR system is also used to improve the weighting of similarity calculations in order to find the case in the case base that is most similar to the new case. Due to the problems encountered in various engineering practice areas, they often have repetitiveness or similarity, that is, they often need to deal with similar problems that have been solved before, and each time may also take a lot of time and cost to solve these similar problems. The problem, if every time you encounter a problem that has already been treated, it takes the same time and cost to deal with it, which is very uneconomical. So after accumulating a certain number of cases or experiences, these engineering problems are quite suitable for solving new problems using case-based reasoning. Case-based reasoning has also been widely used in various engineering fields. Fenves et al. (1995) used the CBR mechanism to build a building structure design support system. V〇ng et aL (2002) assisted the design of the hydraulic circuit design with the CBR mechanism. Maher and Garza (1996) implemented a four-structured case-based reasoning expert system based on the CBR mechanism: CaseCAD, CADsyn, Win, and Demex. Chua et al (2001) constructed a bidding system CASEBID using the CBR mechanism. Morcous et al. (2002a) constructed a CBR system CBRMID that simulates a public construction damage scenario and applies it to the bridge damage 200823776 bad simulation (Morcous et al. 2002b). Andrade et al. (2003) proposed a CBR-assisted design architecture and implemented a bridge-assisted auxiliary system. Karim and Adeli (2003) built a highway construction area traffic management system using a hierarchical object-oriented case model. Yeh (1997) ffi cbr mechanism built a steel frame design assistance system to assist engineers in estimating the section size in the design of the steel structure and reducing the working time of repeated approaching the section size. In summary, the current case In the learning mechanism, the mechanism of learning is only to increase the number of cases in the case library to achieve the purpose of learning. When the number of cases increases, most CBR systems and systems use the method of Will change, and in the case of case similarity calculation or classification, and case _ timely 'have a lot of people to listen to the factors that determine the method of adjustment. However, after the number of cases in the case library gradually increased, reasoning The method should be adjusted accordingly, and if the problem is encountered, the correct solution can be obtained by parameterized experiment or analysis method, == Γ Γ! Apply higher cost and more time The case-style reasoning secret - the human decision and classification method, or the green determined in advance by humans, is not necessarily reasonable for the problem that can be correctly answered by parameter analysis. Cut,,, Knife analysis seeks to use the wisdom of technology to touch CBR transfer or system, but the financial platform is too over-exposed to the true similarity of high m cases. It is easier to find 200823776 number ^ method to achieve the purpose of learning f, when the case library inside the case Gu Yi finds a case that is more similar to the case to be evaluated. It is also the result of 2 well-conformed peaks. However, in the case of _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Heheshengren kind of pure CBR system, but did not make corresponding adjustments or corrections to the recall or adjustment king. In fact, when the number of cases increased, = part of CBR paved in the whispering i change, κ 'and will add a lot of people to pre-determine the weight and adjust the factors of the Henry in its towel. Therefore, as far as the current CBR mechanism, there is a suitable case study (Case_Based Leaming, cbl mechanism necessary As for the academic publication, patent announcement, and related case-telling commercial software, CBR has not yet been issued, and CBL can be used in conjunction with CBL.

Reasoning and

Learning Mechanism 'referred to as CBRL method), the reason is that the application of ^ Zhong Yuzhi knows the above-mentioned lack of the technical towel, _ careful test and research, and a perseverance spirit, finally conceived the case "case-based learning reasoning The method and its implementation device can overcome the shortcomings of the prior art described above. SUMMARY OF THE INVENTION The present invention provides a CBRL method, which can solve the problems existing in the prior art, make the case-based reasoning process more objective and more accurate, and improve the existing CBR learning mechanism, which is very suitable for processing. Repeatability, high similarity, the problem of parameter simulation, and the significant advantages of improving the CBR learning mechanism of 200823776. In addition to the learning mechanism that gradually increases the number of cases, this method can also be used in the automation mechanism when the number of cases increases. Next, the weight of each parameter calculated by the similarity calculation used by the recalling program and the influence degree formula used by the adapting program are adjusted. Therefore, in the process of reasoning, the subjective factor of artificially determining the weight of the parameter to calculate the similarity can be avoided. In addition, by increasing the number of cases to achieve the learning function, the content of recall and adjustment can be adjusted to enhance the learning when the number of cases increases. The effect and improve the accuracy of the evaluation results.

Firstly, according to the concept of the present invention, a case-based learning reasoning method is proposed to obtain a solution to a new problem, including the steps of: parsing a whole case ship included in a case library, and correcting several similar parameters and several adjustment parameters accordingly. Performing a recall procedure based on the similar parameters to take at least one similar case from the case base = similar to the new question; and adjusting the solution to _ at least - similarly based on the adapted parameter binding program . ~ 乂iizJiz is also the case-based learning reasoning method proposed by the local government, the HMrctain) program to update the solution to more packages; =: title, case-based learning reasoning method , Correction and acquisition of several impact factors 1 . The bauxite also referred to this facet 200823776 where the recall procedure involves performing a similarity computation to evaluate a similarity between each full case and the new question. Preferably, the case-based learning inference method proposed by the present invention, wherein the weight (W) required to perform the similarity operation is dynamically determined by the influence factor. Preferably, the case-based learning inference method proposed by the present invention, wherein the adapting procedure comprises performing an adapting operation. According to the concept of the present invention, an electronic device implementing the above method is proposed. According to the concept of the present invention, a private device for implementing a case-based learning and reasoning method is provided, which is used to obtain a solution to a new problem, which includes: an input device that is ribbed to the subject; a storage device, For storing the wood case library, the combination of the different unit and the input device and the storage device are electrically connected to perform the following steps: parsing the similarity and touch parameters included in the case; based on the second and second仃Recalling (recaii) program to obtain M〇eve from the case library) at least one similarity to the new question ^ parameter execution - tune Zhengdapati 〇 该;; ^^^^^^^^^^^^^^ In order to electrically connect the device to the _ computing unit, the electronic device for implementing the Lai learning reasoning method, which is proposed by the present invention, is displayed in the present invention. The answer was verified by experts 12 200823776. The invention relates to an electronic device for implementing a case-based learning and reasoning method, wherein the input device is selected from one of a keyboard, a mouse and a touch screen. Preferably, the present invention provides a private device for implementing a case-based learning and inference method. The storage device is selected from one of a hard disk, a flash drive, and other non-volatile storage devices. Preferably, the present invention is an electronic device that implements a detailed learning and inference method. The arithmetic unit is combined with a central processing unit (CPU) of the electronic computer to match a plurality of memories. Preferably, the present invention provides an electronic device for implementing a method of inference, wherein the f recording device is selected from one of a coffee screen, a rib screen, and a projector. [Embodiment] It can be understood from the implementation of the town's implementation ship, so that the person without the S technology can complete it. However, the implementation of this case is not limited to the implementation form by the following implementation cases. Please refer to the 帛三目 '林(四) jobaki cbrl method compilation not intended. The (4) learning and numeracy method of the present invention is to analyze the (a) program to analyze the influence of each parameter on the evaluation result, the 'box improvement _ the quantity increases the ethical solution, the financial improvement CBR machine _ study method. ^In the CBRL method proposed by Erming, the following part of the CBRL method: Recalling the program phase: the weight used in the degree calculation and the parameters used in the adjustment procedure. The vibration force is calculated by parameter analysis. When the case library (10) _ increased the number of ¥ 'small by the record and the _ recorded section of the seam and shadow plastic production =, making the CBR method face the number of cases in the case library increase this enough to adjust the adjustment and adjustment _ capacity, to mention Learning ability. Mechanism ▲ Set an engineering or scientific problem with n influencing parameters, that is, after inputting the value of one of the numbers, you can make one or several mixed results. When a new case is entered, the similarity calculation is performed in the recall program: the most similar case in the case library can be found, and after comparing the most important cases with the new cases, the touch program can be the most (10) The touch touches out a new result. In the process of recalling and adapting, the weights required for the similarity calculation and the influence degree formulas required by _ can be obtained by parameter analysis, which can not only reduce the subjective factors of the general decision-making weight of the general CBR mechanism, but also increase the subjective factors. The accuracy of the assessment results. The Cbrl method still uses the 4R learning method of retrieve-reuse_i*evie\VHretdii learning eyele, retains the validated cases, and joins the case library to gradually increase the case to achieve the learning effect. When the number of cases increases, The CBRL method can be used to obtain more accurate evaluation results. Therefore, in the CBRL method, the parameter analysis and the added case are the mechanism of learning (Case-Based Learning (CBL), and the recall and adaptation is the mechanism of the reasoning (Case-Based Reasoning, CBR for short). The parameter anaiySis method proposed by the present invention will be described in detail later. If there are a total of N cases in the case library that meet the CBRL method hypothesis, and the engineering problem is simulated by η parameters, the solution to the engineering problem is ρ. When analyzing the ith parameter, if the strict item 14 200823776 The value of (4) 'saki 丨 丨 项 项 项 项 项 项 项 项 项 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四 四$value=.. The gamma value of the case with the smaller answer in the second case. · The value of the case with the larger answer in the case of the second / the case, the case of the jth answer to the smaller case The value of the & value may be different for each tea number. Therefore, the present invention uses the average value of all the cells in the phase, that is, the influence factor of the j solution: 3, ,, Item > Number pair (m~l)x-^ Σ\, (3) The meaning of (3) is the relative Wei in the case base, the sternness rate on the fth solution (4)^ The relative change rate of J is caused by the shadow 2, which means that when the value of the parameter of the ^ item is increased, the number of the ^^^^ is also increased. If v is negative, the ith item of the table is increased by ith. The answer to the lion is reduced. Then divide the ^= 15 200823776 parameters to get the influence factor iirinj of all parameters. For example, the parameter of the tenth item is taken as the example of concrete strength. If there is 1_ pen in the case database, then N=1_, and the possible concrete strength is 21〇kg/cm2, 18〇kg/ 5 kinds of possibilities such as em2, 15〇kg/cm2, i2〇kg^ 峨gw, etc. (4), and in the analysis of the parameter of concrete strength, all cases* can be i嶋/5=fine group. The jth answer on the right is the ultimate strength of the beam section bending moment. When the parameters of the estuary _ estimation county # mine answer are recorded, the concrete strength of the concrete will be analyzed for the bending moment strength of the beam section. Case order and classification, when all other parameters are the same (other parameters may have steel strength, number of steel bars, number of numbers, section size, etc.), all cases are divided into 2〇0 groups, then each group has other The parameters are the same, only the concrete strength is 210kg/cm2^180kg/^12〇kg/cm^^ 90^^ 5 different cases. Therefore, comparing the two cases in which one of the five cases is adjacent, that is, the concrete strength is equal to 210 peak/〇1112 compared with 18〇kg/cm2, 180kg/cm2 compared with 150kg/cm2, I50kg/ Compared with 120kg/cm2, 120kg/cm2 and 90kg/cm2, the total influence of m_l=4 concrete strength on the ultimate strength of beam section bending moment can be obtained. At kg/cm2, the beam section strength is equal to 30t-m, and the concrete strength is equal to 18〇/(if 2, the beam section strength is equal to 27t-m, then one of the groups is ^·· δ one (WXYj _ (30 - 27)/30 "(X "Xj/X Factory Reverse 10 - 180" / 210 =0·7 (4) Since it is divided into 200 groups, each group has 4 3 纟 ·, so for the concrete strength to the beam The relative influence degree of the ultimate moment of the section bending moment can be obtained as a total of 16 200823776 / m = _ solid. The average of this _ heart can be obtained! (3), that is, the case of * Parameter sub-item 』 2: The impact factor of the answer, because the parameters of the second brother II = recorded are called, for a certain parameter will be two cases, so = the problem is to n the number of bribes, job _ Examples shore to the h = time, in order to carry out parametric analysis. If the case of the library case the ship at the time of the amount of cut into the Ding analysis, (iv) entry she will encounter no two cases may

Second: the law produces the impact factor V. When this situation is encountered, the parameter should be removed first for m parameters __, if the _ tea number is an important parameter, the job record is complemented, and the i number analysis is performed. . J ^ ^ In the subsequent recall and adjustment procedures, the weights required for the similarity calculation and the shadow formulas required for the light adjustment are obtained. n When entering the recall procedure, the similarity blush of the case to be evaluated and the case in the case library is compared by similar similarity (similar^y c〇mpu such as i〇n), and the similarity is calculated as follows:

Similarity (new case, case in case base) i=1 (5) The highest degree of similarity is divided into 100 points, which means that the case in the case library is exactly the same as the case S. The lowest score is 0, which means that the ^ cases in the case library are completely different from the case to be sought. f(neweaseeasein_base) is the degree of similarity between a case in the case library and the parameters of the Shanghai item to be solved. The range is 0~1. When the strict parameter of the case in the case library is the same as the parameter f of the case to be solved, fi=1; when the case is in the case; the number of teas in the 17th 200823776 item and the case to be solved _th ith parameter When completely different, ίΗ)〇 is calculated as follows: A =1.

Maxpq

K Λ new case —X;

i, case in case base I t max valuein case base l, new case l, new case ~ min valuein case (6) If the 1th parameter is the concrete strength as in the previous example, the possible values are 210, 180, 150 , 120 disk 90, then X. /, max value in case baSf21 〇, Ximin μ - 仏 _. If the case to be solved is case=180 ’, then when the similarity calculation is performed, the case to be solved is for the Shanghai

The fi parameter of the item parameter (concrete strength) compared with the five concrete strength values in the case library is calculated as follows: f i(Xi,case in case base = 210) = 1. |180-210|

Max (|210 a 180|, |l80 - 9〇|)f sink-off = 1 = i - 18 〇丨 180) = 1- ^ i i, case in case base — 150) = 1 - =0.667 1.000

Fi(Xi,case in case base — 120) = 1 · fi(Xi, case in case base - 9〇) - 1 -

Max(|210 — 180|,|l80 — 90|) ll80-150| Max (|210 one 180丨, 丨180 - 9〇|) |180~120| "Max(|210-180|5|l80 -90[)= 〇333 |180-90| ^ Max (|210-18〇yi80~9〇|)= ° =0.667 (7) (8) (9) (10) (11) It can be seen that In the case that the value of the strict parameter of the case in the case library is the same as the value of the parameter of the ith item to be solved, fi=1, and the value of the parameter and the maximum value of the parameter in the case library (Ximax) The difference between value ώ base-210) and the small value (Xi, min value in case base-90) is set to be completely different, which will make f^O. Therefore, in the above example, the case to be solved, Xi, neW eaSe=18〇, is significantly different from the minimum value of the parameter Xi>min vake ^ CaSebase=90 in the case library (丨18〇-9〇卜9〇), Therefore, the parameter in the case library 18 200823776 (four), and in the case library case, the parameter is the maximum value, the difference is only mixed, so (four). surgery. The method does not determine the weight of the CBR mechanism when calculating the similarity. 减 减 = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = The parameter analysis will adjust the record reductions _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Transmit bribe _ i, the relative influence of the parameters on the broad answer; generation; =] = value multiplied by the case to be solved i, the parameter "case: the item parameters in this case library every change _; ^^ _ The effect of the first production of the solution. Xi.t ί spacing 'willing distance, if the first!, the possible heart of the parameter:;, = If the first item is as ridiculous as before, can == 210 , 180, 150, 120 and 90, then χ·.卞 肊 ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' 2,···,1G, then ^, the change rate of wood case plus 1 is fresh/0, but if the case of the second day of the priest: the rate of change of 1 is - you can see the first parameter 2:: 200823776 ==? Effect phase Therefore, it is more reasonable to use the relative change rate to calculate the relative change rate. Therefore, it is more reasonable to calculate the absolute change amount for all the n parameters in the case library. ;曰巧' also represents the relative weight of each parameter to the jth mine. This is Fy as the first production parameter. The weight of the first answer is _ cap. The weight of each parameter is equal to the sum of the two. (nomallze) process 'even if ίχ, so that the similarity calculation of formula (5) can be calculated between Μ(8). 便式(5) All the similarity calculations in the case library are completed, and from the case library = evaluation _ The most sensible case can make the impact analysis formula of the financial analysis to make a fine touch. The adaptation method used in the adjustment procedure is also proposed by the invention. Degree of influence of parameters _ Applicable nose (ad_ion c〇mputati〇 method. According to the parameter analysis, the degree of influence of the case Ynewc_ by the ρth parameter is · case in case base one 210) : _7 (1:

Iij is the relative influence of the ith parameter on the fth solution by parameter analysis, so: the 3⁄4 axis to be evaluated tentativity parameter

The difference between the Xi and Xi of the most similar case is ν ν W pm , ', V ljnewcase~Xi, most Similar case)/Xi,n_se 'You can get the value of the $ parameter pair between the two cases, the influence of Yj ,. Therefore, 々 is the widest answer to the case to be solved. The difference between Ynew cased and the most similar case is the percentage of the P parameter. The reasoning result of the CBRL method is: 曰20 200823776 new caseJ most similar casej^y^y^xy^ (lg) is also about to be solved by the case and the most similar case in the case library. With the formula (16), you can find the fth solution of the case to be solved. When a case is evaluated by the CBRL method proposed by the present invention, the new case may be verified or verified by a special test (revise(| s〇luti〇n) and then stored (or called update). The case library, this can also be obtained (the ruler mechanism increases the number of cases to set the learning effect, in line with the 4R learning cycle of the CBR mechanism. After the case 2 increase, you can also do parameter analysis to get the new parameters Impact factor Ιΰ. CBR contains two stages of recall and adaptation. In the recall process, the most similar case is obtained by the similarity calculation case. The adjustment program adjusts the most similar case to a new result. The recall procedure in CBRL The similarity calculation method is similar to that of most S Nearest Neighbor, and the similar method is similar to the formula (1), but not limited to the formula (1). When the solution is obtained, the retaining program can be used. After the verification, the solution is stored in the case library to update to the case library, and the number of cases in the case library is increased. As the number of cases in the case library continues to accumulate, subsequent new problems are encountered. It is easy to find a case with higher similarity from the case base, and improve the accuracy of the solution. Based on the case-based learning reasoning method described by Lu Yikai, the applicant re-examines this disclosure--can implement the Lai-style (4) The electronic device of the method, to obtain a solution to the new problem. Please refer to the fourth figure, which is a schematic diagram of an electronic device of a practical case-like method. The electronic device 4 in the fourth figure includes an input device. The device a, the computing unit combination 43 and the output clothing 44, wherein the electronic device 4 is various types of electronic meters 21 200823776 special machines such as desktop personal computers, notebook personal computers and even handheld personal computers (PDAs) Further, the input device 41 can be a keyboard, a mouse, a touch screen or other kinds of input devices for inputting the new problem, and the storage device 42 is a hard disk, a flash drive, or other non-volatile (n 〇n_v〇latile) storage device for storing a case library, and the output device 44 is a CRr screen, an LCD screen or a projector 'to display the solution.

The electronic device 4 〇 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Including the foregoing parameter analysis, back-age sequence, adaptation program, similar operation included in the recall program, adaptation operation in the adaptation program, and deposit and processing materials. In terms of inclination, the arithmetic unit combination 43 is: various types of electronic calculation bribes are processed and post-suppressed with a plurality of memories (m_y_彡成. The county unit electric person with the deduction of Wei is set in the electronic device 40) In the middle of the 'electronic device 4〇卽忐' field T electric clothing set 40 becomes a device that can be used to learn the case-based learning reasoning method. Based on the above, the method of the invention is based on the method of _t_ She studied the branch network of the gods in his crane, or left the _ calculation _, the key is to know the CBR; the applicability of the function, as described later. Class:: The road is using a large number of Neurons are used to simulate network models, which are suitable for predicting multivariables and complex between variables and solutions: Linear relationship, honest green computing, large source, and the process of neural network training can not know how much God = 22 200823776 Numbers, too many or too few neurons often need to get the second chance in the way of trial and error: the accuracy of the system, so the law is the same, the neural network needs 匕 == number, and the genetic algorithm knows the whole Axis miscellaneous 2. ^:= Turning turns the user method, if the helmet is half broken - 々a pass The different method is the optimal solution (4). The influence of the lion's answer is reduced or the importance is matched. However, the parameter analysis used in the invention 1 The influence degree of the parameter on the solution is u, the financial output - a 壹士士戾This represents the weight of each parameter, and it has the weight of the rn 解 面 面 重要性 , , , , , , , , , , , , , , , 合 合 合 合 合 合 合 合 合 合 合 而 而 而 而 而 而 而 而 而 而 而The two proposed by the present invention can not only help the (10) machine to make adjustment weights, but also learn to adjust the adjusted content, and more transparently let the user serve all the information of the learning and reasoning process. It is only the preferred embodiment of the present invention, and is not limited to the scope of the present invention. That is, the equal variation and modification of the patent application of the present invention should still belong to the invention. Within the $ 明贝番查委明鉴, and pray for the right, is the prayer. The case must be modified by the people who are familiar with the craftsmanship, but they are not removed as the scope of the application. Brief description of the formula] The figure is the basis of CBR Schematic diagram of the architecture; the first diagram is a more detailed architecture diagram of the CBR mechanism; 23 200823776 The third diagram is a schematic diagram of the architecture of the CBRL method proposed by the present invention; and the fourth diagram is an electronic method for implementing the case-based learning reasoning method Schematic diagram of the device. [Main component symbol description] 40: Electronic device 41: Input device 42: Storage device 43: Arithmetic unit combination 44: Output device 24

Claims (1)

200823776 X. Applying for the patent Fan Park·· The cow is a wooden case-type self-inference method to obtain a solution to a new problem, including the case and correcting several similar parameters to analyze the total number included in the case library. Adjusting parameters; performing - recalling __ presenting based on the similar parameters to retrieve at least one similar case from the case library that is similar to the new question; and kewei et al. (adaptati(m) The procedure fine-tunes the at least φ-similar case to the answer to the new question. 2. The method of claim 2, further comprising the step of: performing a retain procedure to solve the solution Updated to the case library. 3. The method described in claim 2, wherein the transfer is verified by an expert. 4. The method described in claim 1 further includes the steps of: 5. The method of claim 1, wherein the step of analyzing further comprises: obtaining a plurality of influence factors (I) 〇 6. The method of claim 1, wherein the recall procedure Including A similarity computation to evaluate a similarity between each full case and the new question. 7. The method of claim 5 or 6, wherein the similarity is performed The required weight (W) is determined by the influence factor (jynamicaiiy). 8. The method of claim 1, wherein the adapting procedure comprises performing an adaptation computation. An electronic device implementing the method of claim 1 of the patent scope. 25 200823776 1 ' l0. An electronic device implementing a case-based learning inference method for obtaining a solution to a new problem, comprising: an input device, To enter the new problem; a storage device for storing a case library; an arithmetic unit combination electrically connected to the input device and the storage device, the execution comprising the steps of: parsing the fineness and the silk included in the sample a number of similar parameters and a plurality of adaptation parameters; _ executing a recall procedure based on the similar mics to retrieve and new questions from the case library At least one similar case of similarity; performing an adaptation (&(叩叩丨〇11) procedure based on the 荨5 week-adapted parameter to adjust the at least one similar case to the answer to the new question; and executing one Retaining a program to store the solution in the storage device to update the case library; and an out-of-round device electrically connected to the computing unit to display the solution. 11·If the patent application scope is the first item According to the electronic computer, the solution stored in the step is verified by experts. 12. The electronic computer of claim 1, wherein the input farm is selected from the group consisting of a keyboard, a mouse, and a touch screen. 13. The electronic computer of claim 10, wherein the storage device is far from one of a hard disk, a dish, and other non-volatile (dirty storage devices. . . , 14· *Application The electronic computer according to the first aspect of the invention, wherein the computing unit group is equipped with a plurality of memories, and the electronic computer according to the first aspect of the invention, wherein The output device 26 200823776 is selected from one of a CRT screen, an LCD screen, and a projector. 27