JPH06222710A - Operation training evaluating device - Google Patents

Abstract

PURPOSE:To accurately evaluate the skillfulness of a trainee from the operation result of the trainee by eliminating the need for an instructor who evaluates the skillfulness. CONSTITUTION:When operation is performed through a console panel 1, its operation signal is converted by a controller 2 into a control signal, which is sent to a process simulator 3 as a simulation plant. The process simulator 3 performs process simulation according to the control signal and outputs a process signal indicating the operation result to a neural network 4. The neural network 4 performs operation corresponding to the skillfulness in advance to learn an operation pattern. The neural network 4 compares the learnt operation pattern with an operation pattern sent from the process simulator 3, evaluates the skillfulness of operation technique from their difference, and outputs the evaluation result to the console panel 1 and displays it.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an operation training evaluation device for evaluating the skill level of operation technology in operation training of a plant or the like.

[0002]

2. Description of the Related Art Conventionally, operation training for plants and the like has been carried out using an operation training device. In this conventional driving training device, the target of performance evaluation is displayed in the pattern of the sensor signal.Therefore, a trained instructor is present during the driving training, and the skill level is evaluated by observing the pattern of the signal waveform. There is.

[0003]

The evaluation of the skill level from the pattern of the signal waveform as described above has to rely on a trained instructor, unlike the simple evaluation by the numerical value or the XX formula. I don't get it. However, it takes a great deal of effort for the instructor to attend all the training scenes of many trainees with different learning levels.

If, for example, a range in which a correct answer is to be obtained is set for an operation result and whether or not the range is within the range is evaluated by the XX formula, the range is close to the boundary of the range without depending on the instructor. However, there is a problem in that the degree of proficiency cannot be accurately grasped because the evaluation greatly changes depending on whether or not

In order to solve the above-mentioned problems, if a plurality of ranges for various evaluations are to be provided, the amount of data becomes huge due to the combination of various conditions, which causes problems in data input and management.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a driving training evaluation device that does not require an instructor for evaluating the skill level and can accurately and easily evaluate the skill level from the driving result. And

[0007]

A driving training evaluation apparatus according to the present invention comprises a console for performing a driving operation, a process simulator for performing a process simulation based on an operation signal input from the console, and A neural network for learning and storing a driving pattern according to the skill level, and comparing the driving pattern by the trainer outputted from the process simulator with the previously stored driving pattern to judge the skill level, and the neural network. And a display unit for displaying the determination result according to.

[0008]

When the operator operates the operation console, the operation signal is sent to the process simulator. The process simulator performs a process simulation based on the operation signal and outputs the process signal to the neural network. This neural network has
The driving pattern according to the skill level is stored in advance by learning. Then, the neural network evaluates the skill level of the driving technique by comparing the driving pattern of the trainee output from the process simulator with the driving pattern stored in advance, and displays the evaluation result on the display unit.

With the above configuration, the trainee is
After performing the driving operation, the user can immediately know his / her skill level by looking at the result displayed on the display unit. Therefore, an instructor for evaluating the skill level is unnecessary, and the trainee can check the skill level as many times as he / she learns.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a driving training evaluation apparatus according to an embodiment of the present invention. In the figure,
Reference numeral 1 denotes a console for operating a plant or the like, and an operation signal from the console 1 is sent to the control device 2. The control device 2 generates a control signal based on an operation signal from the console 1 and outputs the control signal to the process simulator 3, which is a simulated plant. This process simulator 3
Performs a process simulation according to a control signal from the control device 2 and outputs a process signal indicating the operation result to the control device 2 and the neural network 4. When the process signal from the process simulator 3 is sent, the control device 2 operates as a monitor signal on the console 1
Output to.

On the other hand, the neural network 4 is made to previously learn and store driving models of advanced driving, intermediate driving, and beginner driving. For example, each of the representatives of advanced driving person, intermediate driving person, and beginner driving person performs driving operation with the console 1, and the neural network 4 is made to learn the driving pattern. The neural network 4 compares the process signal output from the process simulator 3 during training with the stored operation pattern to evaluate the skill level of the driving technique, and outputs a signal indicating the evaluation result to the console 1. To do. The console 1 displays the monitor signal sent from the control device 2 on the display unit DP, and also displays the evaluation result and the certainty factor sent from the neural network 4 on the display unit DP after the operation is completed.

The neural network 4 is composed of an input layer A, an intermediate layer (hidden layer) B, and an output layer C as shown in FIG. The input layer A is the process simulator 3
The process signal sent from is captured as a plurality of amplitude signals at each time interval Δt and input to the intermediate layer B. The intermediate layer B first multiplies each input value by a weight coefficient Wi initially initialized at random, and sends them to the output layer C.
The output layer C further multiplies the input value by another weighting coefficient Vi to obtain output values Z1, Z2 and Z3.

On the other hand, as expected output values, teacher signals D1, D2, D for "advanced", "intermediate", and "beginner" respectively.
Set 3 in advance. For example, if "advanced" is selected, "D1 = 1", "D2 = 0", and teacher signals D1 to D3,
"D3 = 0" is given. In addition, when "Intermediate" is selected, "D1 = 0", "D2 = 1", "D3 = 0", and when "Beginner" is selected, "D1 = 0", "D2 = 0", "D3
= 1 ”is given.

If "advanced" is selected, the neural network 4 determines the difference between the output values Z1, Z2, Z3 and the teacher signals "D1 = 1", "D2 = 0", "D3 = 0". Each of them is taken by the so-called back propagation method, that is, the reverse of the network is traced from the output layer C to the input layer A, and the weighting coefficient is repeatedly corrected so that the error becomes sufficiently small,
Form the expected signal waveform.

FIG. 3 shows a specific configuration example of the neural network 4 in the case of two output units. This neural network 4 has 1
Input layer A having _one input unit 11 _{1 to} 11 ₁₁
Intermediate layer B, 2 with _one hidden unit 12 1-12 ₁₁
The output layer C has a number of output units 13 ₁ and 13 ₂ and an error detection unit 14.

The input layer A receives the input signal X _m (m: 1 to 11) from the process simulator 3 as an input unit 11 _m.
It is taken in by ( _m : 1-11) and sent to the intermediate layer B. The hidden unit 12 _n (n: 1 to 11) of the intermediate layer B is
Multiplied by a weighting factor W _nm to signals X _m from the input unit 11 _m, and sends the calculated output signal Y _n in the output layer C. Output layer C
Output unit 13 _{k (k:} 1,2) is multiplied by a weighting factor V _kn output value Y _n from the hidden units 12 _n, and sends the calculated output signal Z _k to the error detection unit 14. The error detection unit 14 obtains propagation errors δ _k and δ ′ _n between the output signal Z _k of the output unit 13 _k and the teacher signal D _k, and the output layer so that the errors δ _k and δ ′ _n are sufficiently small. The error is backpropagated from C to the input layer A, and the weighting coefficient of the unit in each layer is calculated and corrected.

FIG. 4 shows a configuration example of the _eleventh hidden unit 12 ₁₁ of the intermediate layer B in FIG. The hidden unit 12 ₁₁ receives input values X _{1 to} X from the input layer A.
_{When 11} is given, the weighting factors W _{11,1 to} W respectively
The sum of the product of _11,11 and the product of the input bias of "X ₀ = 1" multiplied by the weighting coefficient W _11,0 "S ₁₁ = Σ (W _nm
× X _m ) ”, and the sum S ₁₁ is further substituted into the sigmoid function“ Y = (1 + e ^−s ) ⁻¹ ”to obtain the output value Y ₁₁ , which is output to the output layer C. The other hidden units 12 _{1 to} 12 ₁₀ of the intermediate layer B have the same configuration as the hidden unit 12 ₁₁ . Next, the operation of the above embodiment will be described.

Prior to the training, the neural network 4
In advance, learn driving models for advanced driving, intermediate driving, and beginner driving. That is, the learning mode is designated, and each of the representatives of the advanced person, the intermediate person, and the beginner person performs a driving operation on the console 1, so that the neural network 4 learns the driving model. In this learning,
"Advanced", "Intermediate", "Beginner" according to the operator's level
Either of them is selected and designated. The corresponding teacher signal D _k (D1, D2, D3) is given by this level designation. If the learning mode is specified, the neural network 4
Starts the learning operation of the driving model according to the back propagation learning algorithm shown in FIG. Neural network 4
First determines the provisional weighting coefficient initial values W ⁰ _nm and V ⁰ _kn (step A 1).

First, for example, a representative of an "advanced person" performs a driving operation using the console 1. By this driving operation, an operation signal is sent to the control device 2. This control device 2
Generates a control signal based on the operation signal from the console 1 and outputs it to the process simulator 3. The process simulator 3 performs a process simulation according to the control signal, and outputs a process signal indicating the operation result to the neural network 4.

The neural network 4 processes the process signal output from the process simulator 3 at time intervals Δ.
At every t, the output value is calculated (learned) for the input value X _m , and the output values Y ′ _m , Y _n , and Z _k of each layer are _obtained (step A2). The error sum of squares E is obtained from the output value Z _k of the output layer C and the teacher signal D _k, and it is determined whether or not the value is sufficiently small (step A3). The teacher signal D _k
In order to obtain the error sum of squares E between the output value Z _k and the output value Z _k , the error sum of squares shown in the following equation is calculated. E = 1/2 · Σ _k (Z _k −D _k ) ² … (1)

If it is determined that the error obtained by the above calculation is sufficiently small, the processing is terminated as it is, but if it is not small, it is necessary to modify the weighting coefficient of each layer and each unit in order to reduce the error. Then, the back-propagation of the error from the output layer C to the input layer A is performed to perform the calculation for determining the weighting coefficient. To modify the weighting factors described above, first, performs propagation error calculation as follows, obtaining the propagation error δ _k, δ _'n (step A4).
The propagation error δ _k of the output layer is δ _k = (Z _k −D _k ) · Z _k · (1−Z _k ) ... (2) The propagation error δ ′ _n of the intermediate layer is δ ′ _n = Σ _k ( δ _k · V _kn ) · Y _n (1-Y _n ) ... (3) Next, the weight coefficient correction amount calculation shown in the following equation is performed to obtain the coefficient correction amounts ΔV _kn and ΔW _nm (step A5). ΔV _kn = −δ _k · Y _n (4) ΔW _nm = −δ ′ _n · X _m (5) Next, the weighting factors V _kn and W _nm are corrected by the following equation (step A6). V _kn = V _kn + α · ΔV _kn (6) W _nm = W _nm + α · ΔW _nm (7) where α = 1 (learning coefficient).

Then, it is judged whether or not the above calculation is completed for all layers and units (step A7).
If not completed, the process returns to step A4 to repeat the same processing. When each layer and each unit is completed, the process returns to step A2 and the output value is calculated again.

Thereafter, the same processing is repeated, and if it is determined in step A3 that the error sum of squares E is sufficiently small, the processing ends. In this way, each layer and each unit weight coefficient are determined. After the above-mentioned "advanced person" has been learned, the "intermediate person" and "beginner" representatives are similarly learned.

As described above, the driving pattern according to the skill level is learned by the neural network 4 and stored. If the learning of the neural network 4 is performed once, no work is required thereafter.

After completing the above learning, the trainee operates the console 1
When the driving operation is performed by the process simulator 3, the process simulator 3 performs the process simulation according to the operation,
A process signal indicating the operation pattern is output to the neural network 4. This neural network 4
Determines the skill level by comparing the driving pattern output from the neural network 4 with the driving pattern stored in advance by learning. That is, the neural network 4
Takes in the process signal output from the process simulator 3 calculates an output value Z _k, "advanced" by the output value Z _k, "intermediate", it is determined which of "beginner". For example, if the output values Z1 to Z3 of the output layer C are "Z1
= 1, Z2 = 0, Z3 = 0 "indicates" advanced "and" Z1
= 0, Z2 = 1 and Z3 = 0 ", then" intermediate "," Z1
If = 0, Z2 = 0, Z3 = 1 ", it is judged that" beginner ". Then, the neural network 4 outputs the above judgment result to the console 1 together with the certainty factor.

The console 1 displays the judgment result sent from the neural network 4 on the display section DP. The trainee can immediately know his level by looking at the display result.

[0028]

As described above in detail, according to the present invention, it is possible to operate the neural network in accordance with the characteristics of the plant to which it is delivered and to learn the neural network, and thereafter, an instructor for evaluating the skill is unnecessary. Therefore, the skill level can be accurately evaluated from the driving results of the trainee. In addition, it is possible for self-study only by trainees, and the skill level can be evaluated many times by this self-study.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a driving training evaluation apparatus according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram of a neural network in the same embodiment.

FIG. 3 is a configuration diagram of a neural network in the embodiment.

FIG. 4 is a configuration diagram of a hidden unit in FIG.

FIG. 5 is a flowchart showing a back propagation learning operation of the neural network according to the embodiment.

[Explanation of symbols]

1 Operator's console 2 Control device 3 Process simulator 4 Neural network 11 _{1 to} 11 ₁₁ Input unit 12 _{1 to} 12 ₁₁ Hidden unit 13 ₁ , 13 ₂ Output unit 14 Error detector

Claims (1)

[Claims] 1. An operating console for performing a driving operation, a process simulator for performing a process simulation based on an operating signal input from the operating console, and an operating pattern previously learned and stored according to the skill level, A neural network for judging the skill level by comparing the driving pattern output from the process simulator by the trainee with the driving pattern stored in advance, and a display means for displaying the judgment result by the neural network. Characteristic driving training evaluation device.