JP2002015107A - Method and system for estimating risk and risk estimation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a risk estimating device capable of extremely appropriately deciding or estimating the existence/absence of a prescribed accident factor. SOLUTION: This accident factor estimation system 10 is provided with an input item processing part 22 having a table where items to be inputted which correspond to respective combinations of classified accident factors and classified accident types are enumerated, a neural network 24 for performing a prescribed operation with the data of an item given from the part 22 as an input layer and outputting a value showing the existence/absence of a prescribed accident factor to an output layer, and a results analyzing part 26 for preparing an index showing the existence/absence of the prescribed accident factor on the basis of an output value appearing on the output layer.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for analyzing a cause of an accident using a neural network and estimating a risk of an accident caused by the cause.

[0002]

2. Description of the Related Art Conventionally, various estimation systems using a neural network have been proposed, especially for estimating the cause of a power transmission accident. For example, JP-A-5-25
In No. 0593, when an accident occurs in the power system, input accident information and use a neural network,
An accident cause estimating device for estimating an accident cause corresponding to accident information of a past accident having the highest similarity as an accident cause of an occurred accident is disclosed. Also, JP-A-8-65
No. 923 also discloses a device for estimating the cause of a power transmission accident. However, in the above-mentioned conventional apparatus, its use is limited to a power transmission accident in which the cause is relatively easy to elucidate.

On the other hand, about 30% of occupational accidents, particularly 40% or more of fatal accidents, occur in the civil engineering and construction business. In addition, the causes of accidents have become extremely diversified with the sophistication, complexity, and size of technologies and systems. Therefore, as described in the proceedings of the 54th Annual Scientific Lecture Meeting of the Japan Society of Civil Engineers, the present inventor made a database of past accidents of occupational accidents, and based on the information of individual workers, work environment, conditions, etc. We are trying to estimate the causes of fatal accidents using neural networks.

[0004]

In the above-mentioned attempt, the accident factors are classified into four human factors and four physical factors, and data relating to the working environment and workers (worker data) are used as learning data. When the network was trained, relatively good results were obtained for the presence or absence of each factor. However, depending on the factors, as a result of comparison between the estimation result and the teacher data, they may not match, and thus higher accuracy has been desired.

Further, it is desired to make a more accurate estimation by learning a neural network by accumulating accidents. Also in this case, it is desired to select an appropriate input layer item depending on the cause of the accident.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a risk estimating apparatus capable of judging or estimating the presence or absence of a predetermined accident factor very appropriately.

[0007]

SUMMARY OF THE INVENTION An object of the present invention is to classify an accident factor and an accident type, respectively, and to provide a neural network with respect to a work environment and workers according to a combination of the accident factor and the accident type. Determining an item to be an input layer of the above, receiving a value of the item relating to the work environment and the worker, and using the input value as an input layer to execute an operation using a neural network; A method of estimating a risk level, comprising: a step of representing data indicating the presence / absence of an accident in an output layer; and a step of creating an index indicating the presence / absence of a predetermined accident factor based on the data represented in the output layer. Achieved.

According to the present invention, an optimum input item is determined according to an accident factor and an accident type. Therefore, since the calculation using the neural network is performed using the optimum input item value as the input layer, it is possible to obtain an output that accurately estimates the presence or absence of the cause of the accident. Further, since it is only necessary to give a value to only the selected input item, the procedure of the operator can be simplified.

In a preferred embodiment of the present invention, the step of accumulating accident data, the step of classifying the accumulated accident data according to the accident factor and the accident type, and the step of Extracting the values of the items related to the working environment and the worker according to the combination of the accident factor and the type of accident, and executing the operation using the neural network using the extracted item values as an input layer. ,
A step of representing in the output layer a value indicating the presence or absence of the classified accident factor; a step of comparing the data of the output layer with data relating to the accident factor corresponding to the extracted item as teacher data; Modifying the weights in the neural network so as to minimize and the data of the output layer, performing the operation, comparing the teacher data with the data represented in the output layer, and modifying the weights. Repeatedly storing the obtained weights in association with the accident cause and the accident type. According to the above embodiment, the weight of the neural network can be modified as the data is accumulated, so that a more accurate output can be obtained.

In a more preferred embodiment, the method further includes a step of extracting a combination of items relating to a work environment and a worker, executing an operation by using the values of the items constituting the combination as an input layer, and executing the calculation in the teacher data and the output layer. Comparing with the represented data, and repeating the correction of the weight, temporarily storing the corrected error and the obtained weight in association with the accident factor and the accident type and the combination of the items, For each of the extracted combinations, repeat the execution of the operation, comparison of the teacher data and the data represented in the output layer, correction of the weight, temporary storage of the corrected error and the acquired weight,
Comparing the errors, identifying the smallest one, and storing the weight associated with the smallest error, in association with the accident factor and accident type, and a combination of the items. Have.

According to the above embodiment, when data is accumulated, for each combination of the accident factor and the accident type,
The optimal combination of items relating to the working environment and the worker is found again. The values of the items that make up this combination become the new input layer.

In a preferred embodiment of the present invention, the items are at least one of a work time, weather at the time of work, a work position, a work type, and a contract order of a company performing the work. And a worker item including at least one of the worker's age, skill, and job type. In a further preferred embodiment of the present invention, the types of accidents are categorized into accidents relating to transportation and machinery, accidents relating to falls, accidents relating to collapse / collapse, accidents relating to flying and falling, and other accidents.

The accident factors are classified into human factors and physical factors, and the human factors are the position of the worker, the position of the worker, the handling of the machine / apparatus of the worker, and the worker. The physical factor may be related to any of the experiences, and the physical factor may be related to any of the deficiency of the prevention equipment, the deficiency of the machine / apparatus, the deficiency of the working environment, and other factors.

[0014] In a more preferred embodiment of the present invention, the step of receiving information indicating at least one of the classified accident types, the information indicating at least one of the classified accident factors is provided. And a step of selecting an item to be input based on each of the combination of the accident type and the accident factor,
Thereby, the item to be the input layer is determined. For example, only one accident type may be selected, and a plurality (for example, two) of accident factors may be selected. In this case, the union of the items to be input determined by the type of accident and one factor of the accident and the items to be input determined by the type of accident and the other factor of the accident is calculated. What is necessary is just to determine the item to be input.

[0015] In another preferred embodiment, the method further comprises the step of receiving information indicating at least one of the classified accident types, and the information on the received work environment and workers. The method includes a step of selecting an item serving as an input layer for each combination of a type and an accident factor, and a step of receiving a value of the selected item for each combination.

Another object of the present invention is to provide an input item processing means having a table in which items to be input are listed according to each combination of a classified accident factor and a classified accident type; Based on an output value expressed in the output layer, a neural network that performs a predetermined operation and outputs a value indicating the presence or absence of an accident factor to an output layer, using the data of the item provided from the item processing means as an input layer, and The present invention is also attained by a risk estimating apparatus comprising: an analyzing unit that creates an index indicating the presence or absence of a predetermined accident factor.

In a preferred embodiment, with respect to the accident database for accumulating accident data and the accumulated accident data, the accident factor and the accident type are classified by referring to the table of the input item processing means. A value extracting means for extracting values of items relating to a work environment and a worker according to a combination of the accident cause and the type of the accident, and a neural network comprising the values of the items extracted by the value extracting means as an input layer. Data comparison means for comparing data indicating the presence or absence of an accident factor, which is represented in an output layer as a result of performing an operation using a net, with data relating to the accident factor corresponding to the extracted item serving as teacher data, Correcting means for correcting weights in the neural network so as to minimize teacher data and output layer data; And weight storage means for storing the weight obtained by repeating the calculation using the data, the data comparison by the data comparison means, and the correction of the weight by the correction means in association with the accident factor and the accident type. .

In a more preferred embodiment, furthermore, an item combination extracting means for extracting a combination of items relating to a work environment and a worker, and executing an operation, using teacher data as an input layer using the values of the items constituting the combination. By repeating the comparison with the data represented in the output layer and the correction of the weight, the corrected error and the obtained weight are temporarily stored in association with the accident factor and the accident type and the combination of the items. And, for each of the extracted combinations, the execution of the operation, the comparison between the teacher data and the data represented in the output layer, the correction of the weight, the temporary storage of the corrected error and the obtained weight are repeated. And a minimum error specifying unit for comparing the errors to specify a minimum error, wherein the weight storage unit associates the error with the minimum error. The weights, and type of the accident factor and accidents, and is configured to store in association with a combination of said items.

Further, the object of the present invention is to
A risk estimation program for operating a computer for estimating a risk of an accident, comprising: a step of classifying an accident factor and an accident type; and a work environment according to a combination of the accident factor and the accident type. And for the operator, determining an item to be an input layer of the neural network; and accepting the values of the item related to the work environment and the operator, and using the input value as an input layer to execute an operation using the neural network. Executing, on the computer, a step of representing data indicating the presence / absence of the classified accident factor in an output layer, and a step of creating an index indicating the presence / absence of a predetermined accident factor based on the data represented in the output layer. This is also achieved by a risk estimation program characterized by causing

[0020]

Embodiments of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a block diagram showing a configuration of the estimation system according to the embodiment of the present invention. As shown in FIG. 1, the estimation system 10 includes an apparatus main body 12, an interface 14, a storage device 16, and an input device 18 including a keyboard and a mouse.
And a display device 20 including a CRT or an LCD.

As shown in FIG. 2, the apparatus main body 12 selects an input item for specifying the cause of the accident, and executes an input item for executing a process for presenting the selected item on the display device 20 through the interface 14. A processing unit 22 and a hierarchical neural network processing unit 24 that receives an input item input by a user and executes processing by a neural network
And a result analysis unit 26 that receives a calculation result from the neural network processing unit 24, creates a processing result in a predetermined format, and outputs the result to the display device 20. The hierarchical neural network processing unit 24 includes an input layer 32,
An intermediate layer 34 and an output layer 36 are provided. Device body 12
The program (learning program or analysis program) for operating is stored in the storage device 16. These programs may be read from a CD-ROM and read into the storage device 16. Alternatively, it may be supplied from outside via a communication line (not shown).

In the present embodiment, data relating to the work environment and the workers are given to the input layer.
In addition, accident factors were classified into four human factors and four physical factors, and the presence or absence of these accident factors was shown in the output layer. As the data to be given to the input layer, the work time, weather at the time of work, work position, work type, and work environment items including the contract degree of the operator performing the work, and the age, skill level and Worker items including occupations are used. More specifically, in the present embodiment, an input value is given to each of the items as shown in FIG.

In “1. month”, input values from “January” to
One of “December” is given, and “2.
Any of “0:00” to “23:59” is given.
Hereinafter, the input numerical values are defined for each item. 3. Weather: "1: sunny", "2: cloudy" or "3:
Any of rain ” Occupations: "1: earthwork", "2: staff", "3: operator", "4: device, drilling, shielding, boring", "5: formwork, rebar, carpenter, formwork" Demolition, Blacksmith, Tobiko, Molding Turf, 6: Logging, Slope, Diver, Chief Engineer, 7: Pile, Anchor, Pile Driving, or 8: Spraying, pumping, PC, painting, paving ” 5. Contract order: any of "0th order (primary contract)" to "4th order" 6. Days after admission: "0 day (first day)" or any number of days Cause: "1: vehicle, train", "2: member",
"3: Machine", "4: Crane", "5: Drowning",
"6: Structure", "7: Slope", "8: Opening",
7. Any one of "9: oxygen deficiency", "10: earth and sand", "11: face" or "12: rock" Age: age of worker Engagement work: "1: light work, preparation work, clearing, cleaning work, walking, resting, waiting", "2: assembling and dismantling work", "3: unloading and loading work", "4: Survey, surveying, monitoring, inspection work, confirmation work, guidance work "," 5: operation work "," 6: machine work "," 7: cutting work ",
"8: logging, diving, pavement, repair, plumbing, painting work",
Either “9: concrete placing operation” or “10: transporting operation” Height: Any of "-17m" to "26m"

Further, as shown in FIG. 4, "human factors 1-4" and "physical factors 1"
"4" includes the following contents. Human factor 1: Entry into danger area, unsafe posture, work in dangerous position, etc. Human factor 2: Violation / error of work procedure, misjudgment, instruction /
Human factors such as ignoring signals / guidance, panic / irritation, shortcut actions, ignoring or overlooking prohibited items, etc. Human factors 3: Misoperation of machinery, improper use of tools / machines, non-use / misuse of protective equipment, etc. 4: Unqualified work, inexperienced / unskilled work, etc.

Physical factor 1: Defects and non-installation of opening prevention / gripping prevention equipment, defective safety equipment, defective safety belt mounting equipment, etc. Physical factor 2: Failure / damage of machinery / equipment, machinery / equipment Defects in the shape and structure of the equipment, etc. Physical factor 3: Irregularity of materials and remaining materials, inadequate work paths,
The physical factor 4 includes items that are not included in the physical factors 1 to 3 described above, such as unsecured, inadequate environment such as lighting, bad weather due to rain or wind, and the like.

Next, in this embodiment, in order to optimize the data to be given to the input layer by the input item processing unit 22, it is necessary to train the estimation system 10. An explanation will be added for this learning. First, as shown in FIG. 5, actual accident cases were classified according to accident types. In the present embodiment, five types of accidents are provided: “transportation / machinery”, “fall”, “collapse / collapse”, “flight” and “other”. For example, the data value of the worker data in FIG. 3 is specified for each of the 58 accidents included in the type “transportation / machine”. In addition, for each of these accidents, "human factors 1-4" and "physical factors 1-4"
4) (for example, a value “1” when present, and a value “0” when not present). Similarly, accidents included in other types (30 cases included in "fall", 19 cases included in "collapse / collapse", 13 cases included in "flying fall", 12 cases included in "other")
The data values of the worker data and
Values for each factor were given.

Next, as shown in FIG. 5, one of the accident cases used for learning was specified (see the item “learning” in FIG. 5). In addition, from 11 items of the work data (see FIG. 3), 2 items each including 1 to 11 items.
A combination of 058 items was used as an input layer.

In this embodiment, the neural network includes an input layer 32 and an output layer 3 as shown in FIG.
6, a structure having a hierarchized intermediate layer 34 is provided, and the value (learning data) input to the input layer is weighted as a connection weight and transmitted to the next layer. You. The value appearing in the output layer is compared with a desired output value (teacher data), and the connection weight is corrected so that the square error between the two becomes small. By repeating such a procedure, that is, back propagation, an error between the value appearing in the output layer and the teacher data is reduced. In the present embodiment, work data composed of a combination of predetermined items is used as the learning data. As the teacher data, data indicating the presence / absence of “human factors 1 to 4” and “physical factors 1 to 4” is used. Necessary data such as the weights are sequentially stored in the storage device 14.

As a result of verifying an input item that minimizes an estimation error with unlearned data (see the item “unlearned” in FIG. 5), the result is as shown in FIG. Was. FIG. 6 shows “human factors 1 to 4” for each type of accident.
4 shows preferred input items for knowing the presence / absence of “4” and “physical factors 1-4”. For example, regarding the accident type “transportation / machinery”, in order to know the existence / absence of “human factor 1”, “3. weather”, “5. contract order”, “6. Days after admission ”,“ 7.
The most accurate output can be obtained if "causes", "9. years of experience", "10. engaged work", and "11. height" are given. In FIG. 6, a column marked with “-” indicates that a particularly preferable input item has not been selected. Therefore, in the present embodiment, the input item processing unit 22 is configured as shown in FIG.
In addition, a table is provided for specifying items of the worker data according to the cause of the accident.

Next, the actual analysis processing in the estimation system 10 will be described. FIG. 7 is a flowchart illustrating a processing procedure of the estimation system according to the present embodiment. First, the operator operates the input device 18 to select an accident type and an accident factor whose presence or absence is to be known (steps 701 and 702). For example, if the operator says "crash"
Regarding the accident, "Human factor 2 (problem of worker's awareness)"
If you want to know the existence of a crash,
The accident factor “human factor 2” is selected.

The input item processing section 22 of the estimation system specifies an item in the table specified by the selected accident type and the accident cause in response to the input from the operator, and displays the item on the screen of the display device 20. Presented above. FIG. 8 is a diagram illustrating an example of an image presented on the screen of the display device 20 in the above example (an example in which the accident type “fall” and the accident factor “human factor 2” are selected). . in this way,
On the display device 20, among the items of the work data, "2. time", "3. weather", "6. days after entrance", "7. cause", "9. years of experience", and "10. Prompt the operator to enter only "Work engaged" and "11. Height". Therefore, the operator only needs to input a necessary value for the presented item.

Next, the input item processing section 2 of the estimation system
2 receives the input value for the predetermined item (step 704), and
4 is input to the input layer 32 (step 705). The neural network 24 starts processing in response to the input of a value to the input layer 32, and sequentially transmits the weighted value to the next layer (step 706).

When a value appears in the output layer, the result analyzer 26
Acquires this value (output value) (step 707), and executes analysis based on the acquired value (step 708).
In the present embodiment, the accident factors (“human factors 1 to
4 ”and“ physical factors 1 to 4 ”), a predetermined value (for example,“ 1 ”) when the factor exists, and another predetermined value (for example,“ 0 ”) when the factor does not exist. )). Therefore, referring to the output value of the accident factor selected in step 702, it is determined whether or not the accident factor exists (or whether the degree of the existence is large or small) based on this value. Create that indicator. In the above determination, for example, a threshold “x” that satisfies “0 <x <1” may be set, and the presence or absence of an accident factor may be determined based on whether the output value is larger or smaller than the threshold. .
Of course, it is needless to say that the existence may be determined by another method. Next, the result analysis unit 26 outputs the obtained index to the display device 20 (Step 709).

This index is used when an accident actually occurs.
It is possible to indicate what kind of human factor has occurred with respect to the worker who was a party to the accident, or what kind of physical factor has occurred with respect to the accident.

In addition, the accident type and the cause of the accident are selected so as to match the situation of the work site, and the work related to the person who is scheduled to work and the situation of the work day according to the selected accident type and the cause. By inputting data, it is also possible to indicate how likely the accident of the accident type is caused by an accident factor. For example, at a work site such as a slope reinforcement or a tunnel construction, a predetermined worker data is input to determine whether any of the accidents may cause the “collapse / collapse” accident. By operating, it is possible to know the possibility of a “collapse / collapse” accident occurring due to the selected accident factor.

That is, the estimation system 10 according to the present embodiment can be used to analyze the accident factor by inputting the accident type of the accident that actually occurred and the situation of the site and the worker. It is also possible to input the situation of the site or the worker and to estimate the possibility of an accident due to the factor.

Next, in the estimation system 10 according to the present embodiment, the input item processing section 22 selects an optimum input item from the work data, inputs a value relating to the item, and activates the neural network. An explanation will be added for the example that was made. In this example, among the unlearned cases shown in FIG. 5, regarding the cases in which the accident type is “collapse / collapse” (5 cases), the existence or nonexistence of “human factor 3 (problem in handling machinery and equipment of workers)” Verified.

For each case, based on the processing of the input item processing section 22, the input items of the worker data displayed on the screen of the display device 20 include "4.
"5. Contract degree", "6. Days after admission", "8. Age", "9. Years of experience" and "10.
You have entered the required values. After giving values to necessary items, the neural network 24 was activated to obtain output values. FIG. 9 is a diagram illustrating output values, teacher data, and accuracy determination of each case. As shown in FIG. 9, in any case, the output value substantially matches the data value of the teacher data. Thus, it can be understood that the presence or absence of an accident factor can be determined extremely accurately by inputting a value into an appropriately selected input item.

According to the present embodiment, the type of accident is classified, the cause of the accident is classified into a predetermined type, and the neural network is formed based on the values of appropriate input items according to the type of accident and the cause of the accident. It drives to obtain an output value indicating the presence or absence of an accident factor. Therefore, since the input item is selected, the operation of the operator can be simplified, and the presence or absence of the accident factor can be determined or estimated with extremely high accuracy.

Next, a second embodiment of the present invention will be described. In the second embodiment, the input values and the output values are the same as those in the first embodiment. In this embodiment, data relating to various accidents that have occurred are accumulated, and the neural network is re-learned based on the accumulated data for a predetermined period to compensate for the lack of data and sequentially respond to various patterns. Accident estimation for estimating the possibility that an accident will occur for each accident factor by extracting a weight in which the error in the neural network converges and extracting an estimated value from the input value. Section.

In this embodiment, the estimation system 1
The hardware configuration of 00 is the same as that shown in FIG. FIG. 10 is a block diagram for explaining functions of a processing apparatus main body of the estimation system according to the second embodiment and components related thereto. As shown in FIG. 10, a processing apparatus main body 1 according to the second embodiment
Reference numeral 12 denotes an input user interface (I / F) 122 for executing a process for inputting necessary information by operating the input device 18 while referring to the display device 20.
From the input user I / F 122 to the input layer 32 of the neural network processing unit 24.
An estimation result analysis unit 126 that analyzes data output through the intermediate layer 34 and the output layer 36 and outputs an estimation result indicating the possibility of an accident with respect to the data input by the operator. User I / F1 for these inputs
The accident estimating unit includes the neural network processing unit 24, the neural network processing unit 24, and the estimation result analyzing unit 126.

Further, the processing unit main body 112 has a data extraction unit 140 for extracting data to be given to the input layer 32 from an accident data storage DB 142 storing various data relating to an accident that has actually occurred. It receives data that is provided to the input layer 32 of the neural network processing unit 24 and output through the input layer 32, the intermediate layer 34, and the output layer 36, compares the received data with teacher data, calculates an error, and calculates an error between each layer. Weight (weight) of a certain synaptic connection
And an optimum weight accumulation DB 146 that stores the optimum weight and the like obtained as a result of learning. The data extraction unit 140, the neural network processing unit 24, and the error analysis unit 144 constitute a data storage / addition unit.

In the present embodiment, data on a predetermined number (for example, 200) of accidents is accumulated in the accident data accumulation 142 in advance, and the weight of the neural network is optimized based on the data.

First, the operation of the data storage / addition unit will be described in more detail. FIG. 11 is a flowchart showing an outline of the processing in the data accumulation / addition unit. Figure 1
As shown in FIG. 1, when an operator operates an input device (not shown) or receives data (accident data) relating to an actually caused accident from external data via a network, this is the accident data. Storage DB1
42 (step 1101).

In the accident data related to an accident,
It is desirable that the type of accident (accident type) shown in FIG. 5, the learning item shown in FIG. 3, and the accident factor shown in FIG. 4 be specified. This makes it possible to easily generate data to be input to the input layer and teacher data. When the number of accumulated accidents reaches a predetermined number (for example, 50, 100, etc.) (Yes in step 1102)
s)), the data extraction unit 140 extracts items of data to be input to the input layer from the data (step 110).
3).

In this embodiment, accidents are accumulated in five types of accidents (accident types), and any one of eight accident factors is given to the accident. Therefore, for each of the 40 combinations, the data to be input to the input layer is extracted from the accident data of the accidents belonging to the combination, and a separately formed neural network is learned, and the weight is again optimized for each combination. Become That is, the data extraction unit 14
0 is a combination of the type of accident to be processed and the cause of the accident (for example, "transportation / machine" accident and "human factor 1").
In combination with).

Next, a learning process is executed in which the newly extracted data is input to the input layer of the neural network 24 to which the weight based on the data originally stored is given (step 1105). FIG. 12 is a flowchart showing the processing of step 1105 in more detail. As shown in FIG. 12, the data extraction unit 140 determines, for a specific combination, items of data to be input to the input layer of the neural network 24 (in the above combination, “3, 5, 6, 7, 9, 10, 11 ") is extracted (step 1201). Further, data to be compared with the output layer (in the above combination, the “human factor” is “1”) is extracted (step 1202).

When the data extraction unit 140 inputs predetermined data into the input layer 32 (step 1203), the neural network 24 transmits a value from the input layer to the output layer via the intermediate layer (step 1204). ). The error analysis unit 144 compares the value appearing in the output layer with the value of the learning data provided from the data extraction unit 140, and corrects the weight so that the error is reduced (step 1).
206). The processing of steps 1204 and 1205 is as follows:
Iterate until the error is considered to be minimal (step 1
207). Steps 1203 to 1207 are repeated for all data to be input to the input layer (see step 1208).

As described above, when the learning process is completed for a certain combination, verification may be performed using the unlearned data that has been left in advance (step 110).
6). The weight obtained in this way is the optimum weight accumulation D
In B146, the information is stored in association with the combination (step 1107). In the present embodiment, 40
The processing from step 1104 to step 1107 is executed for all combinations of the types (step 1108).
reference).

The data accumulation / addition unit according to the present embodiment is configured to perform learning and optimize the weight again in response to the accumulation of a predetermined number of data. This makes it possible to construct a more accurate neural network in accordance with the accumulation of actual accidents. The learning may be performed not in response to accumulation of a predetermined number of data but in response to passage of a predetermined period (for example, six months, one year, or the like). Needless to say.

Next, the processing by the accident estimating unit according to the present embodiment will be described. FIG. 13 is a flowchart illustrating a process performed by the accident estimation unit. Here, by inputting predetermined information by the user, it becomes possible to estimate what factors are likely to cause an accident or the like.

First, when the operator operates the input device 18 to obtain an accident type list, the input user I / F 1
22 displays the list on the screen of the display device 20 in response to this. The operator refers to the list and selects the type of accident to be estimated (step 1301). FIG. 14A is a diagram illustrating an example of the accident type list displayed on the display device screen. In this list 1401, the user operates the input device 18 such as a mouse to turn on a button arranged before each accident, whereby a desired accident type (incident type) can be selected.

Next, the input user I / F 122 presents a list of items to be input, and the user
Is operated to receive a previously entered list in which predetermined items have been entered or entered (step 1302). This list is the same as that shown in FIG. 8, but in the second embodiment, it is desirable to request input of all items shown in FIG. This is because it is sufficient to first obtain all inputs and select an item for each factor to be estimated.

When the necessary information is obtained in this way, the following processing is executed for each of the eight types of accident factors. When an accident factor to be processed (for example, “human factor 1”) is specified (step 1303), items to be input to the input layer 32 are extracted from the type of accident and the accident factor (step 1304). These are input layer 2
32 (step 1305). A value is obtained from the output layer 36 via the transmission in the hidden layer (step 1306) (step 1307). Based on this value, the estimation result analysis unit 126 estimates the degree of risk related to the accident factor that is the processing target (step 1308).

In the present embodiment, the degree of risk of the obtained value “v” is evaluated in multiple stages as follows. 0.9 ≦ v risk level 5 0.7 ≦ v <0.9 risk level 4 0.5 ≦ v <0.7 risk level 3 0.3 ≦ v <0.5 risk level 2 v < 0.3 Danger level 1

Therefore, the estimation result analysis unit 126 obtains the level of the danger level relating to the accident factor to be processed based on the obtained value. Such processing from step 1303 to step 1308 is executed for all accident factors (see step 1309).
When the levels of the danger levels are acquired for the eight accident factors (four human factors and four physical factors), an image showing the estimation result is generated and displayed on the screen of the display device 20 ( Step 1310). FIG. 14 (b)
It is a figure showing the example of the obtained picture. According to the present embodiment, it is possible to obtain a multi-stage evaluation of the degree of risk for each of the accident factors.

Next, a third embodiment of the present invention will be described. In the third embodiment, an optimum combination of items as an input layer is acquired again according to accumulation of accident data. The configuration of the estimation system according to the third embodiment is the same as that shown in FIG.
FIG. 15 is a flowchart illustrating a learning process performed by the data storage / addition unit according to the third embodiment. This learning process can be considered as an extension of the process shown in FIG.

The data extraction unit 140 specifies a combination of items considered as an input layer. In the present embodiment, eleven types of items are provided (see FIG. 3). Therefore, (11 + 11C2 + 11C3 + 11C4 + 11C6 + 1
1C7 + 11C8 + 11C9 + 11C10 + 1) = 20
Forty-seven combinations of items can be considered. In the present embodiment, for each combination, step 120
Processes equivalent to those in steps 1 to 1209 are executed, and the optimum error in the combination is temporarily stored (step 1).
503). When the processing for all of the 2047 combinations is completed, the smallest error among the temporarily stored errors is found, and the combination related thereto is specified (step 1505). Thereby, a new combination of items to be considered as an input layer can be obtained. Further, the weight of the neural network constructed using the combination of the items becomes a new weight.

In the third embodiment, the processing by the accident estimating unit is the same as that in the second embodiment.
In the second embodiment, all the eleven items are input. However, in the third embodiment, the optimum combination of the items specified according to the processing in FIG. Is selected. According to the third embodiment, items to be input to the input layer can be optimized according to the accumulation of accident data. This allows
It is possible to estimate the degree of risk with higher accuracy.

The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the invention described in the claims, which are also included in the scope of the present invention. Needless to say, this is done.

For example, in the above-described embodiment, the types of accidents are “transportation / machine”, “fall”, “collapse / collapse”,
Although classified into "Falling and falling" and "Others", it is not limited to this. In addition, accident factors are classified into four human factors and four physical factors, but this is not limited to the above eight factors. Furthermore, in the present embodiment, eleven items of operator data are prepared, but it is needless to say that the present invention is not limited to this.

In the first embodiment,
The operator operates the input device 18 of the estimation system 10 to give an instruction or a value to the device main body 12, but is not limited to this, and is provided from an external computer via a communication line. Instruction etc. is the estimation system 1
0, and in response to this, the input item processing unit 22 of the apparatus main body 12 operates to transmit the input items to the external computer, or to execute the neural network 24 based on the value from the external computer. It may be configured to be driven.

In the first embodiment, a single accident factor is selected for a single accident type, and based on this, a predetermined input item is specified based on a table shown in FIG. Or a plurality of accident factors may be selected. In the latter case, the union of the input items specified by the accident type and the accident factor may be set as the input items.

Furthermore, in the second and third embodiments, the data storage / addition unit and the accident estimation unit are integrated, but the invention is not limited to this. A weight obtained by a computer provided with a data storage / addition unit having a large data processing capability and an input I / F 122
A program that functions as the estimation result analysis unit 126 is stored in a portable storage medium such as a CD-ROM or a DVD-ROM, and is downloaded to a personal computer so that the personal computer can function as the accident estimation device. good. Also in the first embodiment,
Similarly, the weight obtained by learning and the input item processing unit 2
The portable storage medium including the program 2 and the program functioning as the result analysis unit 26 may be downloaded to a personal computer.

With this arrangement, when an operator or the like operates an input device and operates a program at a construction work site, a factory, a work place, or the like, what kind of accident is highly likely to occur. It becomes possible to know. When changing weights or items due to data accumulation, a portable storage medium may be provided again.

Further, the entire estimation system or a server equipped with a program capable of executing processing equivalent to the portable storage medium is connected to a network such as the Internet, and a client machine accesses the server and
By activating the program, the risk of an accident occurring may be acquired. If this mode is adopted, there is no need to provide a portable storage medium. In this specification, the function of one unit may be realized by two or more physical units, or the function of two or more units may be realized by one physical unit.

[0067]

According to the present invention, it is possible to provide a risk estimating apparatus capable of extremely appropriately judging or estimating the presence or absence of a predetermined accident factor.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an estimation system according to an embodiment of the present invention.

FIG. 2 shows an apparatus main body 12 according to the present embodiment.
3 is a block diagram showing the configuration of FIG.

FIG. 3 is a diagram illustrating items included in worker data according to the present embodiment;

FIG. 4 is a diagram illustrating classification of accident factors according to the present embodiment.

FIG. 5 is a diagram for explaining a case used for learning and verification in the estimation system according to the embodiment;

FIG. 6 is a diagram showing input items determined by an accident type and an accident factor in the present embodiment.

FIG. 7 is a flowchart illustrating a process executed by the processing apparatus main body 12 according to the present embodiment.

FIG. 8 is a diagram illustrating an example of an image prompting an input item in the present embodiment.

FIG. 9 is a diagram illustrating a verification example using the estimation system according to the present embodiment;

FIG. 10 is a block diagram for explaining functions of a processing apparatus main body of an estimation system according to the second embodiment and components related thereto;

FIG. 11 is a flowchart illustrating an outline of a process in a data accumulation / addition unit;

FIG. 12 is a flowchart showing the processing of step 1105 in more detail in the second embodiment.

FIG. 13 is a flowchart illustrating a process performed by the accident estimating unit according to the second embodiment;

FIG. 14 is a diagram illustrating an example of an image displayed on a screen of a display device according to the second embodiment.

FIG. 15 is a flowchart illustrating a learning process performed by a data accumulation / addition unit according to the third embodiment;

[Explanation of symbols]

 Reference Signs List 10 estimation system 12 device body 14 storage device 22 input item processing unit 24 neural network 26 result analysis unit 32 input layer 34 middle layer 36 output layer 122 input user I / F 126 estimation result analysis unit 140 data extraction unit 142 accident data accumulation DB 144 Error analysis unit 146 Optimal weight accumulation DB

Claims (19)

[Claims] 1. Classifying an accident factor and an accident type, respectively, according to a combination of the accident factor and the accident type.
For the work environment and the worker, a step of determining an item to be an input layer of the neural network, and accepting the value of the work environment and the item related to the worker,
Using this as an input layer, performing an operation using a neural network; a step of displaying data indicating the presence or absence of the classified accident factor in an output layer; a predetermined accident based on the data shown in the output layer. Creating an index indicating the presence or absence of a factor. 2. A step of accumulating accident data; a step of classifying the accumulated accident data by an accident factor and an accident type; and in the accident data, the accident factor and the accident type. Extracting the values of the items related to the working environment and the worker according to the combination of: the step of executing an operation using a neural network using the values of the extracted items as an input layer; and the classified accident factors. The step of representing a value indicating the presence or absence of the data in the output layer, and the step of comparing the data of the output layer with the data on the accident factor corresponding to the extracted item as teacher data; and Modifying the weights in the neural network so as to minimize Comparing the data with the data represented in the output layer and repeating the correction of the weight, and storing the obtained weight in association with the accident factor and the accident type. Item 6. The risk estimation method according to Item 1. 3. A step of extracting a combination of items relating to a work environment and a worker; and executing an operation using the values of the items constituting the combination as an input layer; Comparing, and repeatedly correcting the weight, temporarily storing the corrected error and the obtained weight in association with the accident factor and the accident type and the combination of the items; and For each of the combinations, repeat the execution of the operation, comparison of the teacher data with the data represented in the output layer, correction of the weight, temporary storage of the corrected error and the acquired weight, and comparing the error. Identifying the smallest one, and the weight associated with the smallest error, the accident factor and accident type, and a combination of the items 3. The risk estimation method according to claim 2, further comprising the step of storing in association with each other. 4. The method according to claim 1, wherein the items include work time, weather at the time of work,
A work environment item including at least one of a work position, a work type, and a contract degree of a company performing the work, and a work including at least one of the worker's age, skill, and job type The risk estimation method according to any one of claims 1 to 3, further comprising: 5. The type of the accidents is classified into accidents relating to transportation / machinery, accidents relating to falls, accidents relating to collapse / collapse, accidents relating to flying and falling, and other accidents. The risk estimation method according to any one of the above. 6. The accident factor is classified into a human factor and a physical factor, and the human factors include a worker position, a worker position, handling of a machine / apparatus of the worker, and a worker. Claims characterized in that the physical factors relate to any of the following experiences, and the physical factors relate to any of the following: inadequate preventive equipment, inadequate machinery / equipment, inadequate work environment, and other factors. 6. The risk estimation method according to any one of 1 to 5. 7. A step of receiving information indicating at least one of the classified accident types; a step of receiving information indicating at least one of the classified accident factors; 7. A step of selecting an item to be input based on each of the combination of the type of accident and the cause of the accident, whereby the item serving as the input layer is determined. The risk estimation method according to any one of the above. 8. A step of receiving information indicating at least one of the classified accident types; and information on the received work environment and the worker.
A step of selecting an item to be an input layer for each combination of the type of accident and the cause of the accident, and a step of receiving a value of the selected item for each combination. 1 to 6
The risk estimation method according to any one of the above. 9. An input item processing means having a table in which items to be input are listed according to each combination of a classified accident cause and a classified accident type; A neural network that performs a predetermined operation using the data of the items as an input layer and outputs a value indicating the presence or absence of an accident factor to an output layer, and the presence or absence of a predetermined accident factor based on the output value expressed in the output layer. A risk estimating apparatus, comprising: an analyzing unit that creates an index that indicates the risk. 10. An accident database for accumulating accident data, and for the accumulated accident data, referring to a table of input item processing means, classifying the accident factors and accident types, and A value extracting means for extracting values of items relating to a work environment and a worker according to a combination of a factor and an accident type, and a neural network is used with the values of the items extracted by the value extracting means as an input layer. Data comparing means for comparing data indicating the presence or absence of an accident factor, which is represented in the output layer as a result of performing the operation, with data relating to the accident factor corresponding to the extracted item serving as teacher data, Correction means for correcting weights in the neural network so as to minimize layer data; A weight storing means for storing the weight obtained by repeating the data comparison by the data comparing means and the correction of the weight by the correcting means in association with the accident factor and the type of the accident. 9. The risk estimation device according to 9. 11. An item combination extracting means for extracting a combination of items relating to a work environment and a worker, and using the values of the items constituting the combination as an input layer, executing an operation, representing the combination in teacher data and an output layer. Comparison with the data and the correction of the weight are repeated, and the error after correction and the obtained weight are temporarily stored in association with the accident factor and the type of the accident and the combination of the items, and extracted. For each of the combinations obtained, repeat the calculation, compare the teacher data with the data represented in the output layer, correct the weight, temporarily store the corrected error and the acquired weight, and compare the errors. And a minimum error specifying means for specifying a minimum error, wherein the weight storage means stores a weight associated with the minimum error with the accident factor or the like. And type of fine accident risk estimation apparatus according to claim 10, characterized in that in association with the combination of the item. 12. A work environment item including at least one of a work time, a weather at the time of work, a work position, a work type, and a contract degree of a company performing the work. The risk estimating apparatus according to any one of claims 9 to 11, further comprising: a worker item including at least one of a worker's age, skill, and job type. 13. The type of the accidents according to claim 9, wherein the types of the accidents are classified into accidents relating to transportation and machinery, accidents relating to falls, accidents relating to collapse / collapse, accidents relating to flying and falling, and other accidents. The risk estimation device according to any one of claims 1 to 4. 14. The accident factor is classified into a human factor and a physical factor, and the human factors include a position of a worker, a position of the worker, handling of a machine / apparatus of the worker, and
It is characterized in that it is related to any of the worker's experiences, and the physical factors are related to any of the following: inadequate prevention equipment, inadequate machinery and equipment, inadequate work environment, and other factors. The risk estimation device according to any one of claims 9 to 13. 15. A risk estimating program for operating a computer for estimating a danger of an accident for each accident factor, wherein the step of classifying the accident factor and the type of the accident respectively comprises: According to the combination with the type of
For the work environment and the worker, a step of determining an item to be an input layer of the neural network, and accepting the value of the work environment and the item related to the worker,
Using this as an input layer, performing an operation using a neural network; a step of displaying data indicating the presence or absence of the classified accident factor in an output layer; a predetermined accident based on the data shown in the output layer. Creating an index indicating the presence or absence of a factor by the computer. 16. Accumulating accident data; classifying the accumulated accident data according to the accident factor and accident type; and calculating the accident factor and accident type in the accident data. Extracting the values of the items related to the working environment and the worker according to the combination of: the step of executing an operation using a neural network using the values of the extracted items as an input layer; and the classified accident factors. The step of representing a value indicating the presence or absence of the data in the output layer, and the step of comparing the data of the output layer with the data on the accident factor corresponding to the extracted item as teacher data; and Modifying the weights in the neural network to minimize Comparing the master data with the data represented in the output layer, and repeating the correction of the weight, and storing the obtained weight in association with the accident factor and the accident type. The risk estimation program according to claim 15, characterized in that: 17. A method for extracting a combination of items related to a work environment and a worker, the method comprising: using the values of the items constituting the combination as an input layer, executing an operation; Comparing, and repeatedly correcting the weight, temporarily storing the corrected error and the obtained weight in association with the accident factor and the accident type and the combination of the items; and For each of the combinations, repeat the execution of the operation, comparison of the teacher data with the data represented in the output layer, correction of the weight, temporary storage of the corrected error and the acquired weight, and comparing the error. Identifying the smallest one, and combining the weight associated with the smallest error with the accident cause and accident type and the item The computer-readable storage medium storing the risk estimation program according to claim 2, wherein the computer is caused to execute the step of storing in association with the risk. 18. A method for receiving information indicating at least one of the classified accident types; and receiving information indicating at least one of the classified accident factors. Selecting the item to be input based on each of the combination of the type of accident and the cause of the accident, causing the computer to execute, thereby determining the item to be the input layer. Claims 15 to 17
The risk estimation program according to any one of the above. 19. A step of receiving information indicating at least one of the classified accident types; and information on the received work environment and workers.
Selecting an item to be an input layer for each combination of the type of accident and the cause of the accident; and accepting a value of the selected item for each combination. The risk estimation program according to any one of claims 15 to 17, wherein