JP2005165624A - Bayesian network approximation processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a bayesian network approximation processor for performing bayesian network approximation processing. <P>SOLUTION: The processor comprises: a sensitivity data input receiving means for receiving the input of sensitivity data; a validity reference data input receiving means for receiving the input of validity reference data; an input record input receiving means for receiving the input of an input record; an experience record input receiving means for acquiring an experience record; an arithmetic means for calculating the total of multiplication by weighting with the sensitivity data given at every field concerning the input record and the experience record, as validity relative to each experience record at every experience record, classifying the experience records at every kind of validity reference data, and defining the value of an output value field in the experience record classified to high validity reference as output data; and an output means for receiving the output data from the arithmetic means and outputting it. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

The present invention relates to a Bayesian network approximation processing apparatus that performs Bayesian network approximation processing.

  A processing method called a Bayesian network may be used to process various events that occur in machines, devices, systems, and the like. A Bayesian network is a probabilistic model with a graph structure in which random variables are represented by nodes and links are established between variables that show dependency relationships such as causality and correlation, and this link is in the direction of causality. This is a model that has a directivity and is represented by an acyclic directed graph in which a path that follows the link does not circulate (the Bayesian network is detailed in Non-Patent Document 1 below).

  Patent Documents 1 and 2 listed below disclose inventions that use a Bayesian network in an automatic diagnosis system.

JP 2001-75808 A JP 2002-318691 A

Yoichi Motomura, “Probability Network and Application to Knowledge Information Processing”, [online], January 24, 2001, Internet <URL: http://staff.aist.go.jp/y.motomura/DS/DS .html / ＞

  In the processing by the Bayesian network described above, a table based on experience (for a variety of past events, a table of correspondences between input values to the system and output values when those events occur) is created, and a certain input When a value is received, an output value corresponding to the event is searched from the table and output. Therefore, the input value to the system for an event and the output values in all cases (that is, all output values of all Bayes functions) must be stored in the table in advance.

For example, if there are 10 events that are expected to occur in the system and there are 4 possible outcomes for each event (in the case of 4 values and 10 inputs), there are 4 ¹⁰ items (about 1 million items) as a table. Necessary.

  However, it is very difficult to prepare such a table in advance, and even if such a table can be created, the amount of data becomes large. A lot of resources are required. Therefore, when there is only a small hardware resource such as in the ubiquitous field, there is a problem that processing using a Bayesian network cannot be performed.

  It is also difficult to update the table as described above frequently.

  Therefore, the inventor of the present invention is based on a record composed of input values (input record) and a record (experience record) that combines an input value field to the system and an appropriate output value field corresponding thereto for various past events. Bayesian network is approximated by calculating the similarity as effectiveness, classifying experience records by effectiveness, and outputting the output value field of experience records with high effectiveness as output values. A network approximation processing device was invented.

  Further, in order to solve the second problem using the Bayesian network, the Bayesian network approximation processing device is implemented as hardware to increase the processing speed, and even with a small hardware resource. Processing that approximates a Bayesian network is made possible.

The invention of claim 1
A Bayesian network approximation processing device that performs Bayesian network approximation processing, sensitivity data input receiving means for receiving input of sensitivity data, which is data that gives weight to each field, and effectiveness reference data for receiving input of effectiveness reference data Input accepting means; input record input accepting means for accepting input of an input record acquired outside the Bayesian network approximation processing apparatus; preset input value field to the Bayesian network approximation processing apparatus and output thereto An experience record input receiving means for acquiring an experience record composed of a value field, and for each experience record, a sum of multiplications weighted by sensitivity data given to each of the input record and the experience record for each field The Effectiveness calculating means for calculating the effectiveness for each experience record, and classifying the experience record by the effectiveness based on the effectiveness reference data, and outputting from the value of the output value field in the classified experience record A Bayesian network approximation processing apparatus comprising: an effectiveness classification unit that calculates data; and an output unit that receives and outputs the output data from the effectiveness classification unit.

The invention of claim 2
A Bayesian network approximation processing device that performs Bayesian network approximation processing, sensitivity data input receiving means for receiving input of sensitivity data, which is data that gives weight to each field, and effectiveness reference data for receiving input of effectiveness reference data Input accepting means; input record input accepting means for accepting input of an input record acquired outside the Bayesian network approximation processing apparatus; preset input value field to the Bayesian network approximation processing apparatus and output thereto An experience record input receiving means for acquiring an experience record composed of a value field, NF is the number of input value fields in the experience record, and NI is the number of elements of the input value field. The effectiveness of each record Effectiveness calculating means for calculating and effectiveness classification for classifying the experience records by the effectiveness based on the effectiveness reference data and calculating output data from the value of the output value field in the classified experience records A Bayesian network approximation processing device comprising: means; and output means for receiving and outputting the output data from the effectiveness classification means.

The invention of claim 9
The Bayesian network approximation processing device includes at least data storage means for storing the sensitivity data, the validity reference data, and the input record, and the validity calculation means is configured to validate each calculated experience record. Degree is stored in the data storage means, and the effectiveness classification means obtains each validity stored in the data storage means when classifying the experience records for each validity reference data, and then the experience record Is a Bayesian network approximation processing device.

  According to these inventions, by calculating the effectiveness for each experience record as in the present invention without storing in advance all the values of all the Bayes functions constituting the Bayesian network as in the prior art, It is possible to automatically select an experience record and use it as output data. As a result, the output value for the required past event can be reduced, and processing similar to that of the Bayesian network can be performed even with a smaller resource than before.

The invention of claim 10
A Bayesian network approximation processing device that performs Bayesian network approximation processing, acquired from outside the Bayesian network approximation processing device, sensitivity data input receiving means for receiving input of sensitivity data that is data giving weights to each field An input record input receiving means for receiving input of an input record; an experience record input receiving means for acquiring an experience record composed of a preset input value field to the Bayesian network approximation processing device and an output value field thereto; Effectiveness calculating means for calculating, for each experience record, the sum of multiplications weighted with the sensitivity data given for each field for the input record and the experience record, as the effectiveness for each experience record; Above The test record is classified according to the effectiveness based on a predetermined criterion, the effectiveness classification means for calculating the output data from the value of the output value field in the classified experience record, the effectiveness classification means from the A Bayesian network approximation processing device comprising output means for receiving and outputting output data.

The invention of claim 11
A Bayesian network approximation processing device that performs Bayesian network approximation processing, acquired from outside the Bayesian network approximation processing device, sensitivity data input receiving means for receiving input of sensitivity data that is data giving weights to each field An input record input receiving means for receiving input of an input record; an experience record input receiving means for acquiring an experience record composed of a preset input value field to the Bayesian network approximation processing device and an output value field thereto; Effectiveness calculation means for calculating the effectiveness of each experience record based on Equation 1, where NF is the number of input value fields in the experience record and NI is the number of elements of the input value field, and the experience record Based on predetermined criteria. And an effectiveness classification means for calculating the output data from the value of the output value field in the classified experience records, and an output means for receiving and outputting the output data from the effectiveness classification means; Is a Bayesian network approximation processing device.

The invention of claim 17
The Bayesian network approximation processing device includes at least data storage means for storing the sensitivity data and the input record, and the validity calculation means calculates the calculated validity of each experience record as the data storage means. And the effectiveness classification means is a Bayesian network approximation processing device that classifies the experience records after obtaining each validity stored in the data storage means when classifying the experience records.

  According to these inventions, by calculating the effectiveness for each experience record as in the present invention without storing in advance all the values of all the Bayes functions constituting the Bayesian network as in the prior art, It is possible to automatically select an experience record and use it as output data. As a result, the output value for the required past event can be reduced, and processing similar to that of the Bayesian network can be performed even with a smaller resource than before. Further, unlike the cases of claims 1 and 2, it is not necessary to use the effectiveness reference data.

  When the effectiveness classification means performs the classification of the effectiveness, various methods can be taken.

  For example, an average value of an output value field in an experience record whose effectiveness calculated by the effectiveness calculation means is higher than the effectiveness reference data can be used as the output data.

  In addition, the effectiveness calculated by the effectiveness calculation means is higher than the effectiveness reference data, and the average value of the output value field in the experience record corresponding to the upper nth predetermined is output as the output It is good as data.

  Further, based on the effectiveness calculated by the effectiveness calculation means, the experience record is classified for each effectiveness reference data, the value of the output value field in the experience record classified in the highest effectiveness criteria, The output data may be used.

  In this case, as described in claim 6, when there are a plurality of experience records classified into the highest effectiveness criterion data, the average value of the output value field in the experience records classified there May be the output data, and when there are a plurality of experience records classified as the highest effectiveness criterion data as described in claim 7, the most out of the experience records classified there The value of the output value field in the experience record with high effectiveness may be the output data, or there are a plurality of experience records classified into the highest effectiveness reference data as described in claim 8. Alternatively, an output value field value in an arbitrary experience record among the experience records classified therein may be used as the output data.

  Further, the effectiveness classification when the effectiveness reference data is not used (in the case of claims 10 and 11) can be performed by various methods as described above.

  For example, the effectiveness sorting is performed, and the value of the output value field in the experience record having the highest effectiveness can be used as the output data.

  Further, the effectiveness sorting may be performed, and the average value of the output value field in the experience records corresponding to the predetermined top n may be used as the output data.

  Further, the sorting of the effectiveness may be performed, and the values in the output value fields in the experience records corresponding to the predetermined upper nth may be used as output data.

  In addition, the value of the output value field in an arbitrary experience record may be used as the output data among the experience records corresponding to the predetermined top n.

  Further, the effectiveness value of the experience record is weighted with respect to the output value field of the experience record selected as in claims 12 to 15, and the average value of the effectiveness after the weighting is used as output data. good.

  The Bayesian network approximation processing device according to the present invention is preferably realized on hardware, but the probability distribution constituting the field at that time is preferably an integer array rather than a decimal array for hardware processing. Therefore, it is better to use an integer array.

  When the Bayesian network approximation processing device is hardware, as described in claims 19 to 23, the Bayesian network approximation processing device is incorporated directly into a control chip or the like, or incorporated into an external coprocessor for an embedded CPU. It is suitable to be incorporated in the hardware of an expansion board, incorporated in a mobile terminal as a microcontroller, or incorporated as an all-in-one product for a small-scale control device.

  According to the present invention, it is not necessary to create a table having all values of each event as elements in the Bayesian network. In the present invention, since similar experience records are extracted from a table (experience record) having only some values, resources are not required as compared with the conventional case. Therefore, the Bayesian network approximation processing can be performed even with a small hardware resource.

  Conventionally, since it takes time to generate a table, it is difficult to immediately reflect the latest experience. However, in the present invention, since the table is not used, it is possible to immediately reflect the latest experience.

Furthermore, since the hardware is adapted to SDRAM (Synchronous DRAM) burst transfer (a method of transferring a plurality of data continuously by one address designation), an effect of high-speed processing can be obtained. In other words, in the case of an SDRAM that performs burst transfer, it is possible to perform an operation in which data is continuously returned after an elapse of a certain time when an address is sent to the memory. Can be received without waiting time, enabling high-speed processing. Since the present invention is designed so that the operation can be performed in synchronization with the burst transfer, the operation is performed on the spot each time one piece of data is input, and the operation is completed only in the time required for reading the memory. Therefore, high-speed processing can be realized.

  An example of the system configuration of the Bayesian network approximation processing device 1 of the present invention will be described with reference to the system configuration diagram of FIG.

  The Bayesian network approximation processing device 1 includes sensitivity data input reception means 2, validity standard data input reception means 3, input record input reception means 4, experience record input reception means 5, data storage means 6, validity calculation means 7, validity It comprises degree classification means 8 and output means 9.

  Bayesian network approximation processing device 1 is a result of processing performed by Bayesian network approximation processing device 1 based on input record 10 (described later) acquired from an external sensor or the like and preset experience record 11 (described later). Is output as output data 12, and control of, for example, a motor and various control devices is performed.

  A field is probability distribution data corresponding to one variable. For example, an input value from a temperature sensor or an output value to a motor is represented by a probability distribution as a field. In this embodiment of the Bayesian network approximation processing device 1, the probability distribution is an integer array in which the respective probabilities are represented by integers for a plurality of representative values. For example, as shown in FIG. 5C, the input value from the temperature sensor has a probability of “˜20 ° C.” of 20%, a probability of “20 ° C.-30 ° C.” is 70%, and “30 ° C.-40 ° C.” Is 10% and the probability of “40 ° C.˜” is 0%, the input value field from the temperature sensor is an integer array of {20, 70, 10, 0}.

  Since this array is originally a probability distribution, decimal points are displayed (in the case of the above-described integer array, {0.2, 0.7, 0.1, 0.0}), but a Bayesian network approximation process is performed. When the apparatus 1 is provided as a circuit in hardware such as a control chip, since the integer processing is more effective, the probability is multiplied by 100 and used as an array element. Therefore, an array of decimal numbers may be used instead of an integer array multiplied by 100.

  A record is a collection of a plurality of fields corresponding to one event. For example, "input value field from temperature sensor", "input value field from humidity sensor", "input value field from people sensor", "output value field to heater", etc. It is handled as a record. Records composed of the same field can be compared with each other. In the Bayesian network approximation processing device 1 of the present invention, the current input value to the processing device is the input record 10, and events that have occurred in the past are recorded. An experience record 11 is selected, and an appropriate experience record 11 is selected by comparing the records.

  The output data 12 is a field that the Bayesian network approximation processing device 1 outputs to the outside. Therefore, in comparison with the past experience record 11 (described later), an appropriate experience record 11 is selected, the output data 12 is calculated therefrom, and the output data 12 is output. Become.

  The input record 10 is a record indicating an input to the current Bayesian network approximation processing device 1 acquired from a sensor or the like.

  The experience record 11 is a record obtained by combining an input value field to the Bayesian network approximation processing device 1 and an appropriate output value field corresponding thereto for various past events set in advance. The Bayesian network approximation processing device 1 compares the plurality of experience records 11 with the input record 10 to select an appropriate experience record 11 and obtain an output value field. Therefore, as the number of experience records 11 increases, a more accurate result can be output.

  The sensitivity data input receiving unit 2 is a unit that receives (receives and receives the sensitivity data) from the Bayesian network approximation processing device 1 and stores it in the data storage unit 6. Sensitivity data is data that specifies how much weight is given to each field when comparing records. The sensitivity data has a number of elements obtained by multiplying the number of elements of the array indicating the probability distribution of the corresponding field of the input record 10 and the number of elements of the array of the probability distribution of the corresponding field of the experience record 11. For example, if the input value field is an array of 4 elements as indicated by {20, 70, 10, 0}, if the corresponding output value field is also an array of 4 elements, the sensitivity data is 16 elements. (4 × 4 elements). Sensitivity data is input for each field.

  The validity criterion data input receiving unit 3 is a unit that receives (receives and receives the same) the input of validity criterion data from the Bayesian network approximation processing device 1 and stores it in the data storage unit 6. The validity reference data is a numerical value that specifies how much validity (described later) the output data 12 is to be obtained. Therefore, the output value field in the experience record 11 having an effectiveness higher than the effectiveness reference data is obtained as the output data 12. When a plurality of validity standard data are designated, the experience records 11 are classified according to the validity, and an output value field is obtained for each classification.

  The effectiveness is a numerical value indicating the degree of similarity when the input record 10 and an experience record 11 are compared. By comparing the input record 10 with a certain experience record 11 based on the sensitivity data, it is possible to indicate to what extent the experience record 11 is similar to the input record 10 as the effectiveness. The Bayesian network approximation processing device 1 collects the output value fields of the experience records 11 having high effectiveness and calculates the output value.

When the number of elements in a certain field of the input record 10 is NI, the array of the input record 10 is an array of 1 × NI, the array of the experience record 11 is an array of NI × 1, and the sensitivity data is an array of NI × NI. Therefore, the effectiveness is calculated by Equation 1. The calculation of the effectiveness is performed by the effectiveness calculation means 7, and the calculation of the effectiveness using Equation 1 will be described later.

  The input record input receiving means 4 receives (receives, and so on) an input record 10 acquired outside the Bayesian network approximation processing device 1 such as a sensor as an input to the Bayesian network approximation processing device 1, and stores it as data storage Means for storing in means 6.

  The experience record input receiving unit 5 is a unit that acquires (receives and receives the same) the preset experience record 11 from the Bayesian network approximation processing device 1 and sends it to the effectiveness calculation unit 7.

  The data storage means 6 stores the sensitivity data received by the sensitivity data input receiving means 2, the validity reference data received by the validity reference data input receiving means 3, the input record 10 received by the input record input receiving means 4, and the like. The storage means is, for example, a register, a cache, a memory, or the like. However, the present invention is not limited to these, and any means can be used as long as it can store data.

  The effectiveness calculation means 7 calculates, for each input record 10 and experience record 11, for each experience record 11, the sum of multiplications weighted by the sensitivity data given for each field of the experience record 11 as the respective effectiveness. It is a means to calculate.

  The effectiveness classification means 8 classifies the effectiveness calculated by the effectiveness calculation means 7 for each validity reference data, and sets the value of the output value field in the experience record 11 with high effectiveness (if the effectiveness is high, Since there is a high possibility that the experience records 11 are similar, the value of the output value field in the experience record 11 classified as the highest validity criterion is preferably sent to the output means 9 as output data 12). It is means to do. In addition, when there are a plurality of experience records 11 classified by this effectiveness, the average value of the output value field in each experience record 11 classified there may be output data 12.

  FIG. 3 shows a conceptual diagram of processing of the Bayesian network approximation processing device 1 of the present invention.

  Specifically, the effectiveness calculation means 7 calculates the effectiveness for each experience record 11 by using the above-mentioned formula 1, and classifies it for each validity reference data by the effectiveness classification means 8. Then, the value of the output value field in the experience record 11 classified as the highest effectiveness criterion is sent as output data 12 to the output means 9. In addition, the classification of effectiveness may be performed for each effectiveness. If a plurality of experience records 11 are classified, the average value of the output value field of each experience record 11 is sent to the output means 9 as output data 12.

  In the effectiveness calculation means 7, sensitivity data is generally fixed when calculating the above-described effectiveness, and a large number of experience records 11 are input per one input record 10. Therefore, the frequency at which the value changes is “sensitivity data <input record 10 <experience record 11”. Using this, when the input record 10 is received, the multiplication of the sensitivity data and the input record 10 is first calculated and stored as intermediate data in the data storage means 6 or the like, and the experience record 11 May be used to calculate the effectiveness by multiplying the intermediate data stored in the data storage means 6 or the like and each element of the experience record 11. This is advantageous in terms of processing speed and circuit scale.

  That is, instead of calculating “effectiveness = sensitivity data × input record 10 × experience record 11” collectively, “intermediate data = sensitivity data × input record 10” is calculated in advance and stored in the data storage means 6. By storing and calculating the experience record 11 when “effectiveness = intermediate data × experience record 11” is calculated, a large number of experience records 11 can be processed with fewer resources.

  The output unit 9 is a unit that outputs the output data 12 classified by the effectiveness classification unit 8 to the outside of the Bayesian network approximation processing device 1 as a processing result.

  The Bayesian network approximation processing apparatus 1 according to the present invention can perform processing using either hardware or software, but the effect is particularly remarkable when hardware is used.

  In the case of software for processing a conventional Bayesian network, as shown in FIG. 26 (a), after reading on the memory, the CPU performs the calculation, and after the calculation on the CPU is completed, the reading operation on the memory is performed again. Running.

  On the other hand, when the Bayesian network approximation processing device 1 according to the present invention is implemented as hardware, it conforms to SDRAM burst transfer as shown in FIG. 26 (b). Calculated in hardware. That is, since pipeline processing can be performed in parallel between the memory and the hardware, the effect of high-speed processing can be obtained as compared with the prior art.

  The Bayesian network approximation processing device 1 is incorporated directly into a control chip or the like as a circuit, incorporated into an external coprocessor for an embedded CPU, incorporated into hardware of an expansion board, or used for ubiquitous fields (for example, cellular phones, PHS, PDAs). It is preferable to incorporate it as a microcontroller in a mobile terminal) or as an all-in-one product for a small-scale control device.

  FIG. 4 shows a conceptual diagram when the Bayesian network approximation processing device 1 is incorporated in hardware. 4A shows a case where an external coprocessor for the CPU is used. FIG. 4B shows a case where the CPU is built in an expansion board. FIG. 4C shows a built-in microcontroller for the ubiquitous field. FIG. 4D shows a case where it is incorporated as an all-in-one product for a small-scale control device.

  FIG. 4A shows a case where the arithmetic unit of the Bayesian network approximation processing device 1 is connected to the CPU, and the experience record 11 and the input record 10 accept (receive) input from the CPU. . Therefore, the Bayesian network approximation processing device 1 is recorded on the chip of the arithmetic unit.

  In the case of FIG. 4B, a controller, SDRAM, and an arithmetic unit of the Bayesian network approximation processing device 1 are provided on the expansion board, and the host writes experience records 11 in the SDRAM through the controller. When the host gives the input record 10 to the controller, the controller sets the arithmetic unit, transfers the experience record 11 from the SDRAM to the arithmetic unit, and the arithmetic unit performs Bayesian network approximation processing. During this time, the host can perform another process, and the expansion board has the advantage of being able to operate at high speed regardless of the host.

  In the case of FIG. 4C, the method of FIG. 4B is combined with the method of FIG. 4B in order to increase the speed of FIG. 4A, and the arithmetic unit and controller of the Bayesian network approximation processing device 1 are integrated. An SDRAM is connected to the outside and further connected to the CPU. Here, the SDRAM is dedicated to the arithmetic unit, and is characterized in that it can only be accessed from the CPU via the controller. This also has an advantage that the SDRAM and the arithmetic unit can operate independently of the host.

  In the case of FIG. 4D, the controller shown in FIG. 4C is strengthened so as to have a function as a one-chip microcomputer and to control the entire device.

Examples of the expansion board include PROCStar25-1C, PROCStar25-1, PROCStar25-2, PROCStar25-3 from Gidel, and KAC-02A from Mitsubishi Electric Engineering Co., Ltd. As the controller, for example, ARM922T, ARM926EJ-S, ARM1026EJ-S, ARM1136J (F) -S, ARM7TDMI, ARM7TDMI-S, SC100, SC200, ARM7EJ-S, ARM946E-S, ARM966E-S, Renesas Technology's M32R / E series (M32102S6FP, M32104S6FP, M32121FCAWG, M32121MCB-XXXWG, etc.), M32R / ECU series (32170 group (M32170F3VFP, M32170F4VFP, M32170F6VFP), 32171 group (M32171F2VFP, M32171F3V172, 172F3V172, 172MFP) ), 32173 group (M32173F2VFP), 32174 groups (M32174F3VFP, M32174F4VFP), 32176 groups (M32176F2VFP, M32176F2VWG, M32176F2TFP, M32176F2TWG, M32176F3VFP, M32176F3VWG, M32176F3TFP, M32176F3TWG, M32176F4VFP, M32176F4VWG, M32176F4TFP, M32176F4TWG), 32180 groups (M32180F8TFP, M32180F8UFP , M32180F8VFP), 32182 group (M32182F3TFP, M32182F3UFP, 32182F3VFP, M32182F8TFP, M32182F8UFP, M32182F8VFP), and M32R / I series.

  Next, the processing flow in the Bayesian network approximation processing apparatus 1 will be described in detail with reference to the flowchart of FIG. 2 and the system configuration diagram of FIG.

  It is assumed that at least one experience record 11 used in the Bayesian network approximation processing device 1 is set in advance.

  First, the Bayesian network approximation processing device 1 accepts sensitivity data input by the sensitivity data input accepting means 2, and stores it in the data storage means 6 (S100). An example of sensitivity data for receiving input at this time is shown in FIGS. 5 (a1) to 5 (a3). FIGS. 5A1 and 5A2 show a case where sensitivity data is set symmetrically, and FIG. 5A3 shows an example where the sensitivity data is set asymmetrically.

  As described above, the sensitivity data indicates how much the field similarity affects the overall similarity. For example, when the sensitivity data of field A is set to twice the sensitivity data of field B, the effectiveness of field A is higher, so the designation that the similarity of field A is more important than field B is specified. Will be. By setting such sensitivity data, it can be used to emphasize a field whose value does not fluctuate very much or to weaken a field whose value fluctuates greatly. On the other hand, by setting the base of the sensitivity data widely as shown in FIG. 5 (a2), it is possible to specify that the fields are effective even if they are not completely the same. This is effective in that fluctuations can be absorbed in the case of data with a lot of noise components. Also, as shown in FIG. 5 (a3), when sensitivity data is set asymmetrically, for example, as shown in FIG. 5 (a3), it is effective if the input record 10 is larger than the experience record 11 side. It can also be set as.

  Since the sensitivity data is weighted, any numerical standard (for example, a negative number) may be used.

  Next, the validity reference data input for classifying the experience records 11 is received by the validity reference data input receiving means 3, and stored in the data storage means 6 (S110). An example of the effectiveness standard data is shown in FIG.

  It should be noted that either the sensitivity data in S100 or S110 and the input of validity standard data may be received first.

  After accepting the input of sensitivity data and validity standard data, the Bayesian network approximation processing device 1 accepts an input record 10 acquired from outside the Bayesian network approximation processing device 1 such as a sensor by the input record input acceptance means 4, This is stored in the data storage means 6 (S120).

  FIG. 5C shows an example of the input record 10, and FIG. 5D shows an example of the experience record 11. In the example of FIG. 5C, the input record 10 represents temperature data, and “˜20 ° C. is a value 0”, “20 ° C. to 30 ° C. is a value 1”, and “30 ° C. to 40 ° C. is a value. 2 ”,“ 40 ° C. to value 3 ”, the probability of taking the value 0 is 20%, the probability of taking the value 1 is 70%, the probability of taking the value 2 is 10%, the probability of taking the value 3 Indicates 0%. Further, in the example of FIG. 5D, the relationship between temperature, humidity, and wind speed is shown as the experience record 11, and “the probability that the value is 0 is 20% for temperature, 0% for humidity, and 90% for wind speed. “The probability of a value of 1 is 70% for temperature, 10% for humidity, and 10% for wind speed.” “The probability that the value is 2 is for 10% for temperature, 20% for humidity, and 0% for wind speed.” "The probability that the value is 3 indicates that the temperature is 0%, the humidity is 70%, and the wind speed is 0%."

  The experience record input receiving means 5 acquires each experience record 11 one by one (S130), and sends the acquired experience records 11 one by one to the effectiveness calculation means 7. Then, the effectiveness calculation means 7 acquires the input record 10 and sensitivity data stored in the data storage means 6, compares the field of the experience record 11 with the field of the input record 10, and determines the validity of this field. A numerical value is calculated (S140).

Specifically, the effectiveness calculated for each field is obtained by multiplying the field of the input record 10 corresponding to the field of the acquired experience record 11 and its sensitivity data, and adding the multiplication result of each element. And the effectiveness for each experience record 11 is calculated. That is, the processing of Formula 1 is performed.

For example, the field f of the experience record 11 is {20, 80, 0, 0}, and the field f of the input record 10 corresponding to the field f of the experience record 11 is {0, 100, 0, 0} Furthermore, sensitivity data (= s [f] [NI] [NI])
s [f] [0] = {100, 20, 0, 0}
s [f] [1] = {20, 100, 20, 0}
s [f] [2] = {0, 20, 100, 20}
s [f] [3] = {0, 0, 20, 100}
Therefore, the effectiveness is added by multiplying the corresponding data of each element and sensitivity data (that is, for each element, the field of the input record 10 × the field of the experience record 11 × the sensitivity) Data, and adding all the results), (20 × 0 × 100) + (80 × 0 × 20) + (0 × 0 × 0) + (0 × 0 × 0) + (20 × 100 × 20) + (80 × 100 × 100) + (0 × 100 × 20) + (0 × 100 × 0) + (20 × 0 × 0) + (80 × 0 × 20) + (0 × 0 × 100) + (0 × 0 × 20) + (20 × 0 × 0) + (80 × 0 × 0) + (0 × 0 × 20) + (0 × 0 × 100) = 840000.

  The field f of the experience record 11 is {20, 80, 0, 0}, and the field f of the input record 10 corresponding to the field f of the experience record 11 is {20, 80, 0, 0}. Thus, the effectiveness when the sensitivity data is the same as described above is (20 × 20 × 100) + (80 × 20 × 20) + (0 × 20 × 0) + (0 × 20 × 0) + (20 × 80 × 20) + (80 × 80 × 100) + (0 × 80 × 20) + (0 × 80 × 0) + (20 × 0 × 0) + (80 × 0 × 20) + (0 × 0 × 100) + (0 × 0 × 20) + (20 × 0 × 0) + (80 × 0 × 0) + (0 × 0 × 20) + (0 × 0 × 100) = 744000.

  Since the effectiveness can be calculated for each field as described above, in order to calculate the effectiveness for the experience record 11, the total value is added for each field. For example, in the case of the above-described example, the effectiveness for each field obtained in S140 is added. Therefore, the effectiveness in this case is 840000 + 744000 = 1584000.

  In this way, the effectiveness classification means 8 classifies the effectiveness of each experience record 11 calculated in Equation 1 for each validity criterion data received in S110, and the experience corresponding to the highest effectiveness criterion data is obtained. The value of the output value field in the record 11 is set as output data 12 (S150). When a plurality of experience records 11 are classified in the validity standard data, the average value of the output value field in each experience record 11 classified there is calculated and used as output data.

  In the case of the effectiveness standard data shown in FIG. 5B (assuming that this is the highest validity standard data), since the above-mentioned validity is 1584000, the standard is “0”, and this experience Since the record 11 is the highest (in this case, since there is only one experience record 11, the value of the output value field of this experience record 11 is always), the output value field of this experience record 11 is sent to the output means 9 as an output value.

  The processing in S140 and S150 is represented in FIG. 24 using C language which is one of computer languages. Needless to say, when the Bayesian network approximation processing device 1 is realized as hardware such as a control chip, this processing is realized on hardware resources.

The effectiveness classification means 8 sends the output data 12 processed by the effectiveness classification means 8 in S150 to the output means 9, and the output means 9 outputs the output data 12 as an output value (S160).

  Next, a processing flow of the Bayesian network approximation processing device 1 will be described using a specific example. It is assumed that failure prediction is performed as a model to be processed by the Bayesian network approximate processing apparatus 1, and that two inputs, oil temperature and oil amount, are input. That is, there are two oil records and oil quantity as the input record 10, and output data 12 relating to failure prediction is output.

  As shown in FIG. 6 (a), the oil temperature field is “value 0 when it is ˜20 ° C.”, “value 1 when it is 20 ° C. to 40 ° C.”, “value 2 when it is 40 ° C. to 60 ° C.”, “60 As shown in FIG. 6 (b), the oil amount field is “value 0 when 800 ml˜, value 1 when 600 ml˜800 ml”, “400 ml”. It is a field that takes a value of 2 ”when it is ˜600 ml and a value of 3 when it is ˜400 ml. Here, for ease of explanation, the value is assigned as abnormal if the value is large, but may be assigned arbitrarily. In addition, as shown in FIG. 6C, the failure prediction field has “value 0 when normal”, “value 1 when attention is required”, “value 2 when inspection is necessary”, and “value 3 when emergency stop is required”. The field to be taken.

  Further, as the experience record 11, a past input record 10 and an output value field to be output at the time of the event are set. In this example, a case where there are five experience records 11 shown in FIGS. 7A to 7E will be described. FIG. 7A shows a sample experience record 11 consisting of an input record 10 and its output value field when the event is normal, and FIG. 7B shows an input record 10 and its output value field when the event is attention. Sample 1 of experience record 11 consisting of, input record 10 in the event of attention in FIG. 7C, sample 2 of experience record 11 consisting of its output value field, and event in need of inspection in FIG. 7D The sample of the experience record 11 consisting of the input record 10 and its output value field in the case of FIG. 7, and the sample of the experience record 11 consisting of the input record 10 and its output value field in the case of the emergency stop in FIG. Is the case.

  The Bayesian network approximation processing device 1 receives the oil temperature field sensitivity data and the oil quantity field sensitivity data shown in FIG. 8 by the sensitivity data input receiving means 2 of the Bayesian network approximation processing device 1 and receives them. (S100).

  Next, the validity criterion data input acceptance unit 3 of the Bayesian network approximation processing device 1 accepts the input of the validity criterion data having the data shown in FIG. 9, and stores it in the data storage unit 6 (S110).

  After receiving the input of the sensitivity data and the effectiveness reference data, the failure prediction of the apparatus is performed from the oil temperature and the oil amount in the present embodiment. Therefore, the Bayesian network approximation processing apparatus 1 includes the oil temperature sensor and the oil that detect the oil temperature. The input record 10 shown in FIG. 10 obtained by the oil amount sensor for detecting the amount is received by the input record input receiving means 4 and stored in the data storage means 6 (S120).

  Next, the experience record input receiving means 5 acquires each experience record 11 one by one (S130), and sends the acquired experience records 11 one by one to the effectiveness calculation means 7. Then, the effectiveness calculation unit 7 acquires the input record 10 and sensitivity data stored in the data storage unit 6. Then, the effectiveness calculation means 7 calculates, for each experience record 11, the sum of multiplications weighted by the sensitivity data given for each field for the input record 10 and each experience record 11 as the respective effectiveness (that is, (Equation 1 is calculated) (S140).

  And the effectiveness classification means 8 classifies the effectiveness for each experience record 11 based on the effectiveness calculated by the effectiveness calculation means 7 for each effectiveness reference data, and high effectiveness reference data (preferably the highest) The value of the output value field in the experience record 11 classified as (effectiveness reference data) is set as output data 12 output by the output means 9 (S150). If there are a plurality of experience records 11 classified as the highest validity standard data, the average value of the output value fields of the experience records 11 classified there is set as the output data 12. Note that the processing of S140 and S150 can be expressed as shown in FIG. 24 in C language.

  When the processes of S140 and S150 are applied to the specific example of the second embodiment, the following calculation is performed.

  First, when the input record 10 and the experience record 11A are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil amount field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 680000, and the effectiveness of the oil amount field is 300000. Therefore, the effectiveness when compared with the experience record 11A is 980000. The effectiveness in the case of this experience record 11A is stored in the data storage means 6.

  When the input record 10 and the experience record 11B are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil quantity field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 680000, and the effectiveness of the oil amount field is 50000. Accordingly, the effectiveness when compared with the experience record 11B is 730000. The effectiveness in the case of this experience record 11B is stored in the data storage means 6.

  When the input record 10 and the experience record 11C are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil quantity field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 320,000, and the effectiveness of the oil amount field is 300000. Therefore, the effectiveness when compared with the experience record 11C is 620000. The effectiveness in the case of this experience record 11C is stored in the data storage means 6.

  When the input record 10 and the experience record 11D are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil amount field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 320,000, and the effectiveness of the oil amount field is 50000. Therefore, the effectiveness when compared with the experience record 11D is 370000. The effectiveness in the case of this experience record 11D is stored in the data storage means 6.

  When the input record 10 and the experience record 11E are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil quantity field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 40000, and the effectiveness of the oil amount field is 300000. Therefore, the effectiveness when compared with the experience record 11E is 340000. The effectiveness in the case of this experience record 11E is stored in the data storage means 6.

  Thus, the effectiveness classification means 8 extracts the validity of the experience record 11E from each experience record 11A and the validity reference data received in S110 from the data storage means 6, and the effectiveness classification means 8 is effective. The data is classified for each degree reference data, and the value of the output value field in the experience record 11 classified as the highest validity reference data is sent to the output means 9 as output data 12 (S150). Here, since there is one experience record 11 classified as the highest validity reference data, the average value of the output value field need not be calculated.

  Therefore, in the above-mentioned case, since the effectiveness of the experience record 11A is 980000, it corresponds to the value 2 (800001 to 1000000) of the effectiveness reference data.

  Since the validity of the experience record 11B is 730000, it corresponds to the value 3 (600001 to 800000) of the validity standard data.

  And since the effectiveness of the experience record 11C is 620000, it corresponds to the value 3 (600001 to 800000) of the effectiveness reference data in the same manner as the experience record 11B.

  Since the effectiveness of the experience record 11D is 370000, it corresponds to the value 5 (20,0001 to 400000) of the effectiveness reference data.

  Since the validity of the experience record 11E is 340000, it corresponds to the value 5 (20,0001 to 400000) of the validity reference data as in the experience record 11D.

  FIG. 16 schematically shows what is classified for each validity criterion data as described above. Then, since it is the experience record 11A of the reference value “2” that is classified into the highest validity reference data, the failure value of the experience record 11 in the experience record 11A (FIG. 7A). The value of “100, 0, 0, 0” is referred to. Since this is the only experience record 11 classified as the reference value “2”, the value of the output data 12 is also “100, 0, 0, 0”. Accordingly, the effectiveness classification means 8 sends “100, 0, 0, 0” to the output means 9 as the output data 12 output by the output means 9.

When the output means 9 receives the output data 12 “100, 0, 0, 0” from the effectiveness classification means 8, it outputs it. Therefore, it can be determined that the current output result is normal.

  Next, experience data, sensitivity data, and effectiveness standard data are the same as in the second embodiment (that is, experience data is FIG. 7, sensitivity data is FIG. 8, validity standard data is FIG. 9), and input record input acceptance means Assume that the input record 10 received in 4 (S120) and stored in the data storage means 6 is the case shown in FIG. S100 and S110 are the same as those in the second embodiment.

  Similarly to the second embodiment, the experience record input receiving unit 5 acquires each experience record 11 one by one (S130), and sends the acquired experience records 11 one by one to the effectiveness calculation unit 7. Then, the effectiveness calculation unit 7 acquires the input record 10 and sensitivity data stored in the data storage unit 6. The effectiveness calculation means 7 calculates, for each experience record 11, the sum of multiplications weighted by the sensitivity data given for each field for the input record 10 and each experience record 11 as the respective effectiveness (ie, the number). 1 is calculated) (S140).

  The effectiveness classification means 8 classifies the effectiveness for each experience record 11 based on the effectiveness calculated by the effectiveness calculation means 7 for each validity reference data, and the experience classified as the highest effectiveness reference data. The value of the output value field in the record 11 is set as output data output by the output means 9 (S150). If there are a plurality of experience records 11 classified as the highest validity standard data, the average value of the output value fields in the experience records 11 classified there is calculated and set as output data 12.

  When the processes of S140 and S150 are applied to the specific example of the third embodiment, the following calculation is performed.

  First, when the input record 10 and the experience record 11A are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil amount field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 320,000, and the effectiveness of the oil amount field is 20000. Therefore, the effectiveness when compared with the experience record 11A is 340000. The effectiveness in the case of this experience record 11A is stored in the data storage means 6.

  When the input record 10 and the experience record 11B are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil quantity field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 320,000, and the effectiveness of the oil amount field is 340000. Accordingly, the effectiveness when compared with the experience record 11B is 660000. The effectiveness in the case of this experience record 11B is stored in the data storage means 6.

  When the input record 10 and the experience record 11C are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil amount field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 680000, and the effectiveness of the oil amount field is 160000. Therefore, the effectiveness when compared with the experience record 11C is 840000. The effectiveness in the case of this experience record 11C is stored in the data storage means 6.

  When the input record 10 and the experience record 11D are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil amount field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 680000, and the effectiveness of the oil amount field is 34000. Therefore, the effectiveness when compared with the experience record 11D is 1020000. The effectiveness in the case of this experience record 11D is stored in the data storage means 6.

  When the input record 10 and the experience record 11E are compared and the effectiveness is calculated by the effectiveness calculation means 7, the effectiveness of the oil temperature field and the effectiveness of the oil amount field can be calculated as shown in FIG. In this case, the effectiveness of the oil temperature field is 320,000, and the effectiveness of the oil amount field is 20000. Therefore, the effectiveness when compared with the experience record 11E is 340000. The effectiveness in the case of this experience record 11E is stored in the data storage means 6.

  Thus, the effectiveness classification means 8 extracts the validity of the experience record 11E from each experience record 11A and the validity reference data received in S110 from the data storage means 6, and the effectiveness classification means 8 is effective. The data is classified for each degree reference data, and the value of the output value field in the experience record 11 classified as the highest validity reference data is sent to the output means 9 as output data 12 (S150). Here, since there is one experience record 11 classified as the highest validity reference data, the average value of the output value field need not be calculated.

  Therefore, in the above-described case, the validity of the experience record 11A is 340000, which corresponds to the value 5 (0 to 200000) of the validity reference data.

  Since the effectiveness of the experience record 11B is 660000, it corresponds to the value 3 (600001 to 800000) of the effectiveness reference data.

  Since the validity of the experience record 11C is 840000, it corresponds to the value 2 (800001 to 1000000) of the validity reference data.

  Since the effectiveness of the experience record 11D is 1020000, it corresponds to the value 1 (1000001 to 1200000) of the effectiveness reference data.

  Since the validity of the experience record 11E is 340000, it corresponds to the value 5 (20,0001 to 400000) of the validity reference data as in the experience record 11A.

  FIG. 23 schematically shows the data classified for each validity standard data as described above. Then, since it is the experience record 11D of the reference value “1” that is classified into the highest validity reference data, the failure value of the experience record 11 in the experience record 11D (FIG. 7D) is predicted. The value of “0, 0, 100, 0” is referred to. Since this is the only experience record 11 classified as the reference value “1”, the value of the output data 12 is also “0, 0, 100, 0”. Accordingly, the effectiveness classification means 8 sends “0, 0, 100, 0” to the output means 9 as output data output from the output means 9.

When the output means 9 receives the output data 12 “0, 0, 100, 0” from the effectiveness classification means 8, it outputs it. Therefore, it can be determined that the current output result is a time of inspection.

  In the Bayesian network approximation processing device 1, the effectiveness classification means 8 classifies the effectiveness calculated by the effectiveness calculation means 7 for each validity reference data, and a plurality of experience records 11 belonging to the highest validity reference data. Among them, the value of the output value field in the experience record 11 having the highest effectiveness may be sent to the output means 9 as the output data 12.

  Moreover, the effectiveness classification means 8 classifies the effectiveness calculated by the effectiveness calculation means 7 for each validity reference data, and among the plurality of experience records 11 belonging to the highest validity reference data, an arbitrary experience record 11 The value of the output value field in the above may be sent to the output means 9 as output data 12.

Furthermore, the effectiveness classification means 8 classifies the effectiveness calculated by the effectiveness calculation means 7 for each validity reference data, and among the plurality of experience records 11 belonging to the highest validity reference data, You may send to the output means 9 the average value of the value of the output value field in n experience records 11 as the output data 12.

  Next, FIG. 25 shows a case where the effectiveness standard data input receiving means 3 is not provided in the Bayesian network approximation processing device 1. In this case, processes other than the effectiveness classification means 8 are the same as those in the first to fourth embodiments. Therefore, the process of S110 in FIG. 2 is also unnecessary.

  In this embodiment, based on the effectiveness calculated by the effectiveness calculation means 7, the effectiveness classification means 8 performs sorting (rearrangement) of the effectiveness calculated by the effectiveness calculation means 7, and the highest effectiveness is obtained. The value of the output value field in the experience record 11 is sent to the output means 9 as the output data 12 (preferably the value of the output value field in the experience record 11 classified as the highest effectiveness criterion). It becomes a means.

Further, the effectiveness classification means 8 does not use the value of the output value field of the experience record 11 having the highest effectiveness, but the output value field of the experience record 11 having the validity corresponding to the predetermined top n items. The average value of these values may be used as the output data 12, or the value of the output value field of any one of the top n experience records 11 may be used as the output data 12. Further, when there may be a plurality of output data 12, the values in the output value fields of the top n experience records 11 may be used as the output data 12.

  As described above, when the output data 12 is generated in the first to fifth embodiments, the output value field of each experience record 10 may be an average value weighted according to the effectiveness of the experience record 11. good.

  Specifically, when the output value field of the experience record 11r is indicated by xo [r] and the effectiveness of the experience record 11 calculated by the effectiveness calculation means 7 is indicated by val [r], the effectiveness classification means 8 Calculate (xo [r] × val [r]) ÷ Σ (val [r]) as output data 12. Then, the effectiveness classification means 8 transmits the output data 12 to the output means 9, and the output means 9 outputs the output data 12.

  By further weighting the effectiveness as described above, it is possible to obtain the output data 12 in which the experience record 11 having a low effectiveness is taken in, while emphasizing the experience record 11 having a high effectiveness.

  Also, the weighted average may be calculated using the square of the effectiveness instead of the effectiveness itself. Thereby, it is possible to obtain the output data 12 that places importance on the experience record 11 with particularly high effectiveness.

  Each means in the present invention is only logically distinguished in function, and may be physically or practically the same area.

  In implementing the present invention, a storage medium storing a software program for realizing the functions of this embodiment is supplied to the system, and the computer of the system reads and executes the program stored in the storage medium. Of course.

  In this case, the program itself read from the storage medium realizes the functions of the above-described embodiments, and the storage medium storing the program naturally constitutes the present invention.

  Further, by using a circuit design language such as HDL, each function of the Bayesian network approximation processing apparatus 1 of the present invention may be written and operated in a programmable LSI. For example, each function of the Bayesian network approximation processing device 1 described in a circuit design language such as HDL is written in a programmable logic device called FPGA (Field Programmable Gate Array) to perform calculation.

  As a storage medium for supplying the program, for example, a magnetic disk, a hard disk, an optical disk, a magneto-optical disk, a magnetic tape, a nonvolatile memory card, or the like can be used.

  Furthermore, the program may be downloaded and acquired from a network as well as stored in a storage medium.

  Further, by executing the program read by the computer, not only the functions of the above-described embodiments are realized, but also an operating system running on the computer is one of the actual processes based on the instructions of the program. It goes without saying that the case where the function of the above-described embodiment is realized by performing part or all of the processing and the processing thereof is also included.

Furthermore, after the program read from the storage medium is written in the nonvolatile or volatile storage means provided in the function expansion board inserted into the computer or the function expansion unit connected to the computer, the program is instructed. On the basis of the above, it is natural that the arithmetic processing device or the like provided in the function expansion board or the function expansion unit performs part or all of the actual processing and the functions of the above-described embodiments are realized by the processing.

  According to the present invention, it is not necessary to create a table having all values of each event as elements in the Bayesian network. In the present invention, since similar experience records are extracted from a table (experience record) having only some values, resources are not required as compared with the conventional case. Therefore, the Bayesian network approximation processing can be performed even with a small hardware resource.

  Conventionally, since it takes time to generate a table, it is difficult to immediately reflect the latest experience. However, in the present invention, since the table is not used, it is possible to immediately reflect the latest experience.

Furthermore, since the hardware is adapted to SDRAM (Synchronous DRAM) burst transfer (a method of transferring a plurality of data continuously by one address designation), an effect of high-speed processing can be obtained. In other words, in the case of an SDRAM that performs burst transfer, it is possible to perform an operation in which data is continuously returned after an elapse of a certain time when an address is sent to the memory. Can be received without waiting time, enabling high-speed processing. Since the present invention is designed so that the operation can be performed in synchronization with the burst transfer, the operation is performed on the spot each time one piece of data is input, and the operation is completed only in the time required for reading the memory. Therefore, high-speed processing can be realized.

It is a system configuration | structure figure which shows an example of the system configuration | structure of a Bayesian network approximation processing apparatus. It is a flowchart figure which shows an example of the flow of a process of a Bayesian network approximation processing apparatus. It is a conceptual diagram which shows the concept of a process of a Bayesian network approximation processing apparatus. It is a conceptual diagram when a Bayesian network approximation processing device is incorporated in hardware. It is a figure which shows an example of effectiveness reference | standard data, sensitivity data, an input record, and an experience record. It is an example of the field of oil temperature and oil amount. It is an example of an experience record. It is an example of sensitivity data. It is an example of effectiveness standard data. It is an input record in the case of Example 2. 10 is a table showing the effectiveness of an experience record 11A in Example 2. 10 is a table showing the effectiveness of an experience record 11B in Example 2. 10 is a table showing the effectiveness of an experience record 11C in Example 2. 10 is a table showing the effectiveness of an experience record 11D in Example 2. 10 is a table showing the effectiveness of an experience record 11E in Example 2. It is a table | surface which shows the output value field of each experience record classified for every effectiveness reference | standard data in Example 2, and its average value. It is an input record in the case of Example 3. 10 is a table showing the effectiveness of an experience record 11A in Example 3. 10 is a table showing the effectiveness of an experience record 11B in Example 3. 10 is a table showing the effectiveness of an experience record 11C in Example 3. 10 is a table showing the effectiveness of an experience record 11D in Example 3. 14 is a table showing the effectiveness of an experience record 11E in Example 3. It is a table | surface which shows the output value field of each experience record classified for every effectiveness reference | standard data in Example 3, and its average value. This is an example when S140 and S150 are written in C language. It is an example of the other system configuration of a Bayesian network approximation processing device. It is a figure which shows the concept of the high-speed process at the time of hardware-izing a Bayesian network approximation processing apparatus.

Explanation of symbols

1: Bayesian network approximation processing device 2: Sensitivity data input reception means 3: Effectiveness reference data input reception means 4: Input record input reception means 5: Experience record input reception means 6: Data storage means 7: Effectiveness calculation means 8: Effectiveness classification means 9: Output means 10: Input record 11: Experience record 12: Output data

Claims (23)

A Bayesian network approximation processing device that performs Bayesian network approximation processing,
Sensitivity data input accepting means for accepting input of sensitivity data, which is data giving weights to each field;
Effectiveness reference data input receiving means for receiving the input of the validity reference data;
An input record input receiving means for receiving an input record acquired outside the Bayesian network approximation processing device;
Experience record input receiving means for acquiring an experience record consisting of an input value field to the Bayesian network approximation processing device and an output value field corresponding thereto, which is set in advance,
Effectiveness calculating means for calculating, for each experience record, the sum of multiplications weighted by the sensitivity data given for each field for the input record and the experience record, as the effectiveness for each experience record;
An effectiveness classification means for classifying the experience records by the effectiveness based on the effectiveness reference data and calculating output data from the value of the output value field in the classified experience records;
Output means for receiving and outputting the output data from the effectiveness classification means;
A Bayesian network approximation processing device comprising: A Bayesian network approximation processing device that performs Bayesian network approximation processing,
Sensitivity data input accepting means for accepting input of sensitivity data, which is data giving weights to each field;
Effectiveness reference data input receiving means for receiving the input of the validity reference data;
An input record input receiving means for receiving an input record acquired outside the Bayesian network approximation processing device;
Experience record input receiving means for acquiring an experience record consisting of an input value field to the Bayesian network approximation processing device and an output value field corresponding thereto, which is set in advance,
Effectiveness calculating means for calculating the effectiveness of each experience record based on Equation 1, where NF is the number of input value fields in the experience record and NI is the number of elements of the input value field;
An effectiveness classification means for classifying the experience records by the effectiveness based on the effectiveness reference data and calculating output data from the value of the output value field in the classified experience records;
Output means for receiving and outputting the output data from the effectiveness classification means;
A Bayesian network approximation processing device comprising:

The effectiveness classification means includes:
The average value of the output value field in the experience record in which the effectiveness calculated by the effectiveness calculation means is higher than the effectiveness reference data is used as the output data.
The Bayesian network approximation processing apparatus according to claim 1, wherein the Bayesian network approximation processing apparatus is provided. The effectiveness classification means includes:
The effectiveness calculated by the effectiveness calculation means is higher than the effectiveness reference data, and the average value of the output value field in the experience record corresponding to the upper nth predetermined is defined as the output data. To
The Bayesian network approximation processing apparatus according to claim 1, wherein the Bayesian network approximation processing apparatus is provided. The effectiveness classification means includes:
Based on the effectiveness calculated by the effectiveness calculation means, the experience record is classified for each effectiveness criterion data, and the value of the output value field in the experience record classified as the highest effectiveness criterion is the output Data
The Bayesian network approximation processing apparatus according to claim 1, wherein the Bayesian network approximation processing apparatus is provided. The effectiveness classification means includes:
When there are a plurality of experience records classified into the highest effectiveness criteria data, the average value of the output value field in the experience records classified there is used as the output data.
The Bayesian network approximation processing apparatus according to claim 5. The effectiveness classification means includes:
If there are a plurality of experience records classified as the highest effectiveness reference data, the value of the output value field in the experience record with the highest effectiveness among the experience records classified there is used as the output data. And
The Bayesian network approximation processing apparatus according to claim 5. The effectiveness classification means includes:
When there are a plurality of experience records classified into the highest effectiveness criteria data, among the experience records classified there, the value of the output value field in any experience record is the output data,
The Bayesian network approximation processing apparatus according to claim 5. The Bayesian network approximation processing device is:
At least the sensitivity data, the effectiveness reference data, and a data storage means for storing the input record,
The effectiveness calculation means stores the calculated effectiveness of each experience record in the data storage means,
The effectiveness classification means classifies the experience records after obtaining each validity stored in the data storage means when classifying the experience records for each validity reference data.
The Bayesian network approximation processing apparatus according to claim 1, wherein the Bayesian network approximation processing apparatus is provided. A Bayesian network approximation processing device that performs Bayesian network approximation processing,
Sensitivity data input accepting means for accepting input of sensitivity data, which is data giving weights to each field;
An input record input receiving means for receiving an input record acquired outside the Bayesian network approximation processing device;
Experience record input receiving means for acquiring an experience record consisting of an input value field to the Bayesian network approximation processing device and an output value field corresponding thereto, which is set in advance,
Effectiveness calculating means for calculating, for each experience record, the sum of multiplications weighted with the sensitivity data given for each field for the input record and the experience record, as the effectiveness for each experience record;
The effectiveness classification means for classifying the experience record by the effectiveness based on a predetermined criterion, and calculating output data from the value of the output value field in the classified experience record;
Output means for receiving and outputting the output data from the effectiveness classification means;
A Bayesian network approximation processing device comprising: A Bayesian network approximation processing device that performs Bayesian network approximation processing,
Sensitivity data input accepting means for accepting input of sensitivity data which is data giving weights to each field;
An input record input receiving means for receiving an input of an input record acquired outside the Bayesian network approximation processing device;
Experience record input receiving means for acquiring an experience record composed of an input value field to the Bayesian network approximation processing device and an output value field corresponding thereto, which is set in advance,
Effectiveness calculating means for calculating the effectiveness for each experience record based on Equation 1, where NF is the number of input value fields in the experience record and NI is the number of elements in the input value field;
An effectiveness classification means for classifying the experience records by the effectiveness based on a predetermined criterion, and calculating output data from the value of the output value field in the classified experience records;
Output means for receiving and outputting the output data from the effectiveness classification means;
A Bayesian network approximation processing device comprising:

The effectiveness classification means includes:
The effectiveness is sorted, and the value of the output value field in the experience record having the highest effectiveness is set as output data.
The Bayesian network approximation processing apparatus according to claim 10 or 11, characterized in that: The effectiveness classification means includes:
The effectiveness is sorted, and the average value of the output value field in the experience record corresponding to the upper nth predetermined is set as output data.
The Bayesian network approximation processing apparatus according to claim 10 or 11, characterized in that: The effectiveness classification means includes:
The effectiveness sorting is performed, and the values of the output value fields in the experience records corresponding to the upper nth predetermined are set as output data, respectively.
The Bayesian network approximation processing apparatus according to claim 10 or 11, characterized in that: The effectiveness classification means includes:
Out of the experience records corresponding to the top n defined in advance, the value of the output value field in any experience record is the output data.
The Bayesian network approximation processing apparatus according to claim 10 or 11, characterized in that: The effectiveness classification means includes:
For the value of the output value field, the effectiveness of the selected experience record is weighted, and the average value of the effectiveness after the weighting is set as output data.
The Bayesian network approximation processing apparatus according to claim 12, wherein the Bayesian network approximation processing apparatus is provided. The Bayesian network approximation processing device is:
Having at least data storage means for storing the sensitivity data and the input record;
The effectiveness calculation means stores the calculated effectiveness of each experience record in the data storage means,
The effectiveness classification means classifies the experience records after obtaining each validity stored in the data storage means when classifying the experience records.
The Bayesian network approximation processing apparatus according to claim 10 or 11, characterized in that: The probability distribution of each field is an integer array.
The Bayesian network approximation processing apparatus according to any one of claims 1, 2, 10, and 11. The Bayesian network approximation processing device is:
The Bayesian network approximation processing device according to claim 1, wherein the Bayesian network approximation processing device is incorporated in the control chip as a circuit. The Bayesian network approximation processing device is:
19. The Bayesian network approximation processing apparatus according to claim 1, wherein the Bayesian network approximation processing apparatus is incorporated in an external coprocessor for an embedded CPU. The Bayesian network approximation processing device is:
19. The Bayesian network approximation processing apparatus according to claim 1, wherein the Bayesian network approximation processing apparatus is incorporated as hardware of an expansion board. The Bayesian network approximation processing device is:
The Bayesian network approximation processing apparatus according to claim 1, wherein the apparatus is built in a mobile terminal as a microcontroller. The Bayesian network approximation processing device is:
19. The Bayesian network approximation processing apparatus according to claim 1, wherein the Bayesian network approximation processing apparatus is incorporated in a small-scale control apparatus.

Images