Machine learning device, electronic device, machine learning program, and simulation device

Abstract

The machine learning device has room for further consideration regarding confirmation of reliability of the output data.SOLUTION: The machine learning device includes a model holding unit, a data storage unit, and a model computation unit. The model storing unit stores a first machine learning model trained with supervised learning and a second machine learning model trained with unsupervised learning. The model computation unit inputs the input data to the first machine learning model to generate first output data, and inputs the input data to the second machine learning model to generate accuracy data. The reliability data is calculated using a value of at least one of an input layer, an intermediate layer, or an output layer of the second machine learning model so that the tendency is displaced to correspond to the first output data.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The invention disclosed in this specification relates to a machine learning device, an electronic device, a machine learning program, and a simulation device. [Background technology]

[0002] In the past, in the field of industrial factory equipment maintenance, AI (artificial intelligence) has been increasingly applied to condition-based maintenance of mechanical systems.

[0003] As an example of the related prior art, Patent Document 1 can be mentioned. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2019 / 035279

[0005] [overview] The machine learning device disclosed in Patent Document 1 and used in such mechanical systems and the like leaves room for further consideration in terms of verifying the reliability of its output data (inference data).

[0006] The machine learning device disclosed herein includes a model holding unit, a data storage unit, and a calculation unit. The model holding unit is configured to hold a first machine learning model that has undergone supervised learning and a second machine learning model that has undergone unsupervised learning. The data storage unit is configured to store input data to be input to the first machine learning model and the second machine learning model. The calculation unit is configured to input the input data to the first machine learning model to generate first output data, and to input the input data to the second machine learning model to generate accuracy data. The second machine learning model includes an input layer, an output layer, and at least one intermediate layer disposed between the input layer and the output layer. The accuracy data is calculated using at least one of values ​​of the input layer, the intermediate layer, and the output layer so that the trend shifts in response to the first output data.

[0007] The electronic device disclosed in this specification includes the machine learning device having the above-described configuration.

[0008] The machine learning program disclosed in this specification is intended to function as the machine learning device having the above configuration.

[0009] The simulation device disclosed in this specification calculates accuracy data using the machine learning device having the above configuration. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram showing the configuration of a computer 100. As shown in FIG. [Figure 2] FIG. 2 is a block diagram showing the configuration of the machine learning device 6 according to the first embodiment of the present disclosure. [Figure 3] FIG. 3 is a diagram showing the configuration of the prediction model 80a. [Figure 4] FIG. 4 is a diagram showing the configuration of the verification model 80b. [Figure 5]FIG. 5 is a diagram illustrating a mode in which the first output data do1 is generated and verified using the input data 71, a prediction model 80a, and a verification model 80b. [Figure 6] FIG. 6 is a graph showing an example of the input data 71 and the first output data do1. [Figure 7] FIG. 7 is a graph showing the first output data do1 and the accuracy data da. [Figure 8] FIG. 8 is a block diagram showing the configuration of a machine learning device 6 according to the second embodiment of the present disclosure. [Figure 9] FIG. 9 is a diagram schematically illustrating an aspect in which first output data do1 is generated and verified using input data 71, a prediction model 80a, and a plurality of verification models 80b.

[0011] [Detailed explanation] <About Computer 100> First, a computer 100 that functions as a machine learning device 6 according to the present disclosure will be described. Next, the machine learning device 6 according to the first embodiment of the present disclosure will be described in detail.

[0012] 1 is a diagram showing the configuration of a computer 100. The computer 100 functions as a machine learning device 6, which will be described later. The computer 100 is, for example, a PC (Personal Computer).

[0013] The computer 100 includes a CPU (Central Processing Unit) 100A, a memory 100B, an auxiliary storage device 100C, an operation input unit 100D, and a display unit 100E.

[0014] The CPU 100A has a control device and an arithmetic device (neither of which are shown). The control device interprets program instructions and controls each part of the computer 100. The arithmetic device performs arithmetic processing.

[0015] The memory 100B is a semiconductor storage device that temporarily stores programs or data. The information stored in the memory 100B is erased when the computer 100 is powered off.

[0016] The auxiliary storage device 100C is configured with a hard disk drive (HDD) or a solid state drive (SSD), and stores programs or data. The programs stored in the auxiliary storage device 100C are loaded into the memory 100B. The CPU 100A executes the programs loaded into the memory 100B.

[0017] Here, the auxiliary storage device 100C stores a simulation program P. The simulation program P is a program for causing the computer 100 to function as a machine learning device 6, which will be described later. The machine learning device 6 will be described in detail later.

[0018] The operation input unit 100D is configured with a keyboard, a mouse, or the like, and is a device that provides operation input to the computer 100. Information input from the operation input unit 100D is sent to the memory 100B.

[0019] The display unit 100E is configured by, for example, a liquid crystal display, and converts the information acquired from the memory 100B into an image and outputs it.

[0020] <Regarding the machine learning device 6 of the first embodiment> Next, a description will be given of the machine learning device 6 according to the first embodiment of the present disclosure. The machine learning device 6 is configured with an MCU (Micro Controller Unit). The machine learning device 6 is incorporated into a predetermined mechanical system (such as a motor device) and controls this mechanical system. In addition to controlling the mechanical system, the machine learning device 6 can also perform machine learning using various data of this mechanical system as input data.

[0021] 2 is a block diagram showing the configuration of the machine learning device 6 according to the first embodiment of the present disclosure. As shown in FIG. 2, the machine learning device 6 includes a data storage unit 7, a model holding unit 8, and a calculation unit 9.

[0022] The data storage unit 7 stores input data 71 and initial value data 72. The input data 71 is, for example, time-series data output from a mechanical system or the like. This time-series data may be subjected to preprocessing such as normalization or FFT as necessary. The initial value data 72 is set to an initial value determined by the computer 100 as described above.

[0023] The model storage unit 8 stores a prediction model 80a and a verification model 80b. The prediction model 80a and the verification model 80b are machine learning models that can learn and infer based on input data. Details of the prediction model 80a and the verification model 80b will be described later.

[0024] The calculation unit 9 has a learning calculation unit 91 and an inference calculation unit 92. The learning calculation unit 91 performs supervised learning using the prediction model 80a, input data 71, and initial value data 72. The learning calculation unit 91 also performs unsupervised learning using the verification model 80b, the input data 71, and initial value data 72.

[0025] The inference calculation unit 92 performs inference using the prediction model 80a or the verification model 80b, the input data 71, and the initial value data 72. Inference can be performed during learning and after learning is completed.

[0026] The probability data calculation unit 93 generates probability data da using the input data 71 and the verification model 80b.

[0027] Next, the prediction model 80a will be described in detail. Fig. 3 is a diagram showing the configuration of the prediction model 80a. As shown in Fig. 3, the prediction model 80a includes a three-layer neural network 50. The prediction model 80a is a trained inference model that has been trained in advance using predetermined training data.

[0028] The three-layer neural network 50 is an AI model having an input layer 50A, a hidden layer 50B, and an output layer 50C. The hidden layer 50B is also called an intermediate layer. Generally, in a three-layer neural network 50, for n-dimensional input data x∈Rk×n with a batch size k, an n'-dimensional inference result y∈Rk×n' is obtained as y=G(x·α+b)β. Here, α∈Rn×m is a weight connecting the input layer 50A and the hidden layer 50B. β∈Rm×n' is a weight connecting the hidden layer 50B and the output layer 50C. Furthermore, b∈Rm is a bias for the hidden layer 50B. G is an activation function for the hidden layer 50B. For example, a Sigmoid or ReLU function can be used as the activation function.

[0029] As described above, prediction model 80a is a trained inference model that has been trained in advance using predetermined training data and has been weighted by predetermined weights for α and β. In Figure 3, the weights set by training are indicated by the thickness of the lines connecting input layer 50A and hidden layer 50B, and the thickness of the lines connecting hidden layer 50B and output layer 50C.

[0030] The input data 71 is input to the prediction model 80a and inferred, thereby generating the first output data do1.

[0031] <Considerations on the accuracy of machine learning model outputs> Generally, supervised learning using machine learning models is used to predict future time-series data, predict specific parameters that are difficult to sense, and so on. However, the accuracy (i.e., the degree of reliability) of predicted data obtained through supervised learning is unknown. This has led to a problem in that it is difficult for users of machine learning models to decide whether to accept or reject predicted data generated by the machine learning model, whether to perform re-learning, and so on.

[0032] To address this issue, the machine learning device 6 of the present disclosure allows a user of the machine learning device 6 to acquire accuracy data da. The accuracy data da is data indicating the accuracy of the output result (first output data do1) generated by the prediction model 80a (details will be described later). By referring to the acquired accuracy data da, the user can easily determine whether to accept the output result or whether to perform re-learning. The machine learning device 6 according to each embodiment of the present disclosure will be described in more detail below.

[0033] 4 is a diagram showing the configuration of the verification model 80b. As shown in FIG. 4, the verification model 80b includes a three-layer neural network 51.

[0034] The three-layer neural network 51 is an AI model having an input layer 51A, a hidden layer 51B, and an output layer 51C. The hidden layer 51B is also called an intermediate layer. Generally, in the three-layer neural network 51, for n-dimensional input data x∈Rk×n with a batch size k, an n'-dimensional inference result y∈Rk×n' is obtained as y=G(x·γ+b)ι. Here, γ∈Rn×m is a weight connecting the input layer 51A and the hidden layer 51B. ι∈Rm×n' is a weight connecting the hidden layer 51B and the output layer 51C. Also, b∈Rm is a bias for the hidden layer 51B. G is an activation function for the hidden layer 51B. For example, a Sigmoid or ReLU can be used as the activation function.

[0035] The three-layer neural network 51 employs an algorithm that allows sequential learning with any batch size. When the i-th training data {xi∈Rki×n, ti∈Rki×n'} of batch size ki is obtained, it is necessary to find βi that minimizes the error shown in the following equation (1).

number

[0036] The optimized weight βi is calculated by the following equation (2). P i =P i-1 -P i-1 H i T (I+H i P i-1 H i T ) -1 H i P i-1 β i =β i-1 +P i H i T (t i -H i β i-1 ) (2)

[0037] Here, P0 and β0 are obtained by the following equation (3). P0=(H0 T H0) -1 β0=P0H0 T t0(3)

[0038] The learning algorithm is as follows: (1) The weight α and bias b are initialized with random numbers. (2) Calculate H0 for x0, and calculate P0 and β0. (3) Batch size k i Each time the i-th training data of P is obtained, i and β i It should be noted that instead of using the formula for calculating β0 in equation (3), a value initialized by a random number may be used as β0.

[0039] The computational bottleneck in the above equation (2) is (I+H i P i-1 H i T ) -1 However, (I+H i P i-1 H i T) has a matrix size of k × k, so when k = 1, the inverse matrix operation can be replaced with the reciprocal operation. Therefore, by fixing the batch size to k = 1, the operation becomes easy even for a computing device such as a microcomputer.

[0040] Furthermore, the verification model 80b according to this embodiment performs learning using an autoencoder. The autoencoder uses input data as training data and learns to reconstruct the input data as an inference result. In other words, in the above example, it learns with t = x. Because an autoencoder does not require separate training data to be created, it is a type of unsupervised learning algorithm. Furthermore, by making the number of nodes in the hidden layer smaller than the number of nodes in the input and output layers, the hidden layer matrix can be considered a dimensionally compressed form of the input data once the error between the input data and the inference result has converged. That is, the encoding result of input data x is H = G(x·γ + b), and the decoding result of H is obtained as y = H·ι.

[0041] FIG. 5 is a diagram illustrating a mode in which the first output data do1 is generated and verified using the input data 71, a prediction model 80a, and a verification model 80b.

[0042] 5, when input data 71 (corresponding to input data x shown in FIG. 3) is input to a prediction model 80a and inference is performed by an inference calculation unit 92, first output data do1 is generated as an inference result. When input data 71 (corresponding to input data x shown in FIG. 4) is input to a verification model 80b and inference is performed by an accuracy data calculation unit 93, accuracy data da is generated as an inference result.

[0043] When the above-described computer 100 is made to function as the machine learning device 6, the CPU 100A displays the generated first output data do1 and accuracy data da on the display unit 100E.

[0044] As described above, the accuracy data da is data indicating the accuracy of the first output data do1 generated by the prediction model 80a. The accuracy data da is comprehensive data including various indicators, numerical values, graphs, etc. Details of the accuracy data da will be described later.

[0045] A user of the machine learning device 6 can compare the generated first output data do1 with the accuracy data da to analyze and confirm the accuracy (=degree of reliability) of the first output data do1. Hereinafter, a specific example will be described.

[0046] 6 is a graph showing an example of input data 71 and first output data do1. The input data 71 and first output data do1 are time-series data with the horizontal axis representing time and the vertical axis representing the value of an arbitrary data. The input data 71 is predetermined data actually output by a mechanical system incorporating a machine learning device 6. The first output data do1 is data of the inference result obtained by inputting the input data 71 into a prediction model 80a and performing inference.

[0047] As shown in Fig. 6, the input data 71 has an abnormality in the latter half of the data (the area circled by a dashed line in Fig. 6, hereinafter also referred to as "abnormal area Ap"). Here, it is assumed that the mechanical system has entered some abnormal state, and the input data 71 output by the mechanical system contains an abnormal value.

[0048] Here, the start point of the input data 71 is set to time t1. The time when an abnormality occurs in the input data 71 is set to time t2. The end point of the input data 71 is set to time t3. The period from time t1 to time t2 is set to normal period T1. Normal period T1 is a period in which no abnormal values ​​appear in the input data 71. The period from time t2 to time t3 is set to abnormal period T2. Abnormal period T2 is a period in which abnormal values ​​appear in the input data 71.

[0049] As described above, the prediction model 80a is an inference model that has been trained in advance. The training data used in the pre-training does not contain any abnormal values ​​that would correspond to the abnormal location Ap, and is basically normal. Therefore, when input data 71 that includes the abnormal location Ap is input to the prediction model 80a and inference is performed, the first output data do1 that is the inference result does not contain any abnormal values ​​that would correspond to the abnormal location Ap.

[0050] That is, even if inference is performed using the prediction model 80a, abnormal values ​​corresponding to the abnormal point Ap are not included in the inference result. Therefore, the first output data do1 generated at this time has low accuracy in the latter half of the data (especially in the abnormal period T2). However, as described above, it is difficult for a user to grasp the accuracy of the first output data do1 simply by referring to the first output data do1. For this reason, it is not possible to make decisions such as re-learning the prediction model 80a or reconsidering the pre-processing of the input data 71.

[0051] Figure 7 is a graph showing the first output data do1 and the accuracy data da. Times t1 to t3 in Figure 7 correspond to times t1 to t3 in Figure 6. The accuracy data da includes an input / output error da1, a first hidden layer anomaly degree da2, a first hidden layer error da3, and a second hidden layer error da4.

[0052] The input / output error da1 is calculated based on the input data 71 and the second output data do2. The first hidden layer anomaly degree da2, the first hidden layer error da3, and the second hidden layer error da4 are calculated based on the hidden layer 51B. The input / output error da1, the first hidden layer anomaly degree da2, the first hidden layer error da3, and the second hidden layer error da4 are data with different tendencies. The calculation methods for the input / output error da1, the first hidden layer anomaly degree da2, the first hidden layer error da3, and the second hidden layer error da4 will be described later.

[0053] As shown in Figure 7, assume that an abnormality occurs in the input data 71 during the abnormal period T2. The accuracy data da exhibits a specific tendency depending on the input data 71. By checking and analyzing this accuracy data da, the user can grasp the degree of deviation between the input data 71 and the first output data do1, and thus the accuracy of the first output data do1. Figure 7 will be explained in more detail as follows.

[0054] In the accuracy data da shown in Figure 7, it is difficult to find any particular trends at first glance for the first hidden layer anomaly degree da2 and the second hidden layer error da4 during the normal period T1. For this reason, it is difficult to confirm any change in trends between the normal period T1 and the abnormal period T2 for the first hidden layer anomaly degree da2 and the second hidden layer error da4.

[0055] On the other hand, the input / output error da1 and the first hidden layer error da3 in the normal period T1 are lower than the input / output error da1 and the first hidden layer error da3 in the abnormal period T2. In this way, at a glance, the input / output error da1 and the first hidden layer error da3 show a change in trend between the normal period T1 and the abnormal period T2.

[0056] When the probability data da shown in FIG. 7 is generated, the user can infer that some abnormality has occurred in the input data 71 by checking the change in the trend of the probability data da described above.

[0057] Even if no particular trend is apparent at first glance, it is possible to identify a particular trend based on the user's experience when generating the accuracy data da multiple times. The accuracy data da shown in Figure 7 is an example. Depending on the input data 71, a particular trend may emerge in the first hidden layer anomaly degree da2 and the second hidden layer error da4.

[0058] <About input / output error da1> Next, we will explain the input / output error da1, the first hidden layer anomaly degree da2, the first hidden layer error da3, and the second hidden layer error da4. The input / output error da1 represents the error between the second output data do2 and the input data 71. The second output data do2 is data of the inference result obtained by inputting the input data 71 into the verification model 80b and performing inference. The input / output error da1 is calculated by a loss function based on the input data 71 and the second output data do2. Specifically, it is as follows.

[0059] Each input value included in the input data 71 is defined as an input value x. Each output value (each value of the inference result) included in the second output data do2 is defined as an output value y. As a loss function for calculating the input / output error da1, for example, MAE (Mean Absolute Error) or MSE (Mean Squared Error) can be used. When the loss function is MAE, the loss function L is expressed as in the following equation (4).

number

number

[0060] <About the first hidden layer anomaly score da2> The first hidden layer anomaly degree da2 is calculated based on the first hidden layer vector ha. More specifically, the first hidden layer anomaly degree da2 represents the normalized distance between the first hidden layer vector ha and the mean vector of the first hidden layer vector ha. The first hidden layer vector ha is a vector derived based on the input data 71 and the hidden layer 51B. Specifically, it is as follows.

[0061] The first hidden layer vector ha is a feature vector of the hidden layer 51B when the input data 71 is input to the verification model 80b and inferred. The first hidden layer vector ha is expressed by the following equation (6).

number

[0062] The mean vector of the first hidden layer vector ha is expressed by the following equations (7) and (8).

number

number

[0063] The first hidden layer abnormality degree da2 is calculated by the following equation (9).

number

[0064] <About the first hidden layer error da3> The first hidden layer error da3 is calculated by a loss function based on the first hidden layer vector ha and the second hidden layer vector hb. The second hidden layer vector hb is the feature vector of the hidden layer 51B when the second output data do2 is input to the verification model 80b and inferred. The second hidden layer vector hb is expressed by the following equation (10):

[0065]

number

[0066] The first hidden layer error da3 is calculated by a loss function L that represents the error between the first hidden layer vector ha and the second hidden layer vector hb. When the loss function is MAE, the loss function L is expressed by the following equation (11).

number

number

[0067] <About the second hidden layer error da4> The second hidden layer error da4 is calculated using a loss function based on the first hidden layer anomaly degree da2 and the second hidden layer anomaly degree db1. The second hidden layer anomaly degree db1 represents the normalized distance between the second hidden layer vector hb and the mean vector of the second hidden layer vector hb.

[0068] The mean vector of the second hidden layer vector hb is expressed by the following equations (13) and (14).

number

number

[0069] The second hidden layer abnormality degree db1 is calculated by the following equation (15).

number

[0070] As a loss function for calculating the hidden layer error da4, for example, MAE can be adopted. When the loss function is MAE, the hidden layer error da4 is calculated by the loss function L of the following equation (16).

number

number

[0071] <Regarding the Machine Learning Device 6 of the Second Embodiment> Next, a machine learning device 6 according to a second embodiment of the present disclosure will be described. Note that the machine learning device 6 of this embodiment basically has a common configuration with the machine learning device 6 of the first embodiment described above. Therefore, the common configurations are denoted by the same reference numerals and will not be described here.

[0072] 8 is a block diagram showing the configuration of a machine learning device 6 according to the second embodiment of the present disclosure. The machine learning device 6 includes a data storage unit 7, a calculation unit 9, and an anomaly detection unit 10, similar to those described above. In addition, the machine learning device 6 includes a model holding unit 20.

[0073] The model storage unit 20 stores a prediction model 80a and multiple (here, two) verification models 80b. The number of nodes in the input layer 50A of the two verification models 80b is the same. The number of nodes in the hidden layer 50B of the two verification models 80b may be the same or different. The activation functions of the two verification models 80b may be the same or different.

[0074] FIG. 9 is a diagram schematically illustrating an aspect in which first output data do1 is generated and verified using input data 71, a prediction model 80a, and a plurality of verification models 80b.

[0075] 9, one of the verification models 80b is an inference model trained using normal input data 71 as training data. The other of the verification models 80b is an inference model trained using a specific state (e.g., a specific state including abnormal data) as training data.

[0076] The machine learning device 6 inputs input data 71 to a prediction model 80a and a plurality of verification models 80b, and generates first output data do1 and a plurality of accuracy data da (here, accuracy data da' and accuracy data da'').

[0077] As described above, one verification model 80b is an inference model that has learned a specific state. Therefore, the accuracy data da" generated by inputting input data 71 into this verification model 80b exhibits a relatively significant tendency when a state equivalent to the specific state described above occurs in the input data 71. The user can infer the accuracy of the first output data do1 by checking and analyzing the accuracy data da' and the accuracy data da".

[0078] <Modification> The present disclosure is not limited to the above embodiments, and various modifications are possible within the scope of the present disclosure. For example, although the accuracy data da is comprehensive data including various indicators, numerical values, graphs, etc., it may be one predetermined data (for example, any one of the above-mentioned input / output error da1, the first hidden layer anomaly degree da2, the first hidden layer error da3, and the second hidden layer error da4).

[0079] In the second embodiment described above with reference to FIG. 9, the model storage unit 8 stores two verification models 80b. However, the model storage unit 8 may store three or more verification models 80b. In this case, the number of nodes in the input layer 50A of each verification model 80b is the same. The number of nodes in the hidden layer 50B of each verification model 80b may be the same or different. The activation functions of each verification model 80b may be the same or different. In this case, each verification model 80b may or may not have been trained using predetermined training data, and may be trained (or retrained) along with inference.

[0080] <Additional Notes> The machine learning device (6) includes a model holding unit (8) configured to hold a first machine learning model (80a) that has undergone supervised learning and a second machine learning model (80b) that has undergone unsupervised learning, a data storage unit (7) configured to store input data (71) to be input to the first machine learning model (80a) and the second machine learning model (80b), and a data storage unit (7) configured to input the input data (71) to the first machine learning model (80a) to generate first output data (do1), and to store the input data (71) in the second machine learning model (80b). and a calculation unit (9) configured to input the first output data (do1) to the second machine learning model (80b) to generate accuracy data (da), and the second machine learning model (80b) includes an input layer (51A), an output layer (51B), and at least one intermediate layer (51C) arranged between the input layer (51A) and the output layer (51B), and the accuracy data (da) is calculated using at least any one of the values ​​of the input layer (51A), the intermediate layer (51C), and the output layer (51B) so that the tendency shifts in response to the first output data (do1) (first configuration).

[0081] In the machine learning device (6) according to the first configuration, the accuracy data (da) includes an input / output error (da1) calculated by a loss function based on the values ​​of the input layer (51A) and the values ​​of the output layer (51B) (second configuration).

[0082] In the machine learning device (6) according to the second configuration, the calculation unit (9) is capable of generating second output data (do2) by inputting input data (71) to a second machine learning model (80b) and performing inference, and is capable of calculating an input / output error (da1) by a loss function based on the input data (71) and the second output data (do2), and is capable of generating a first hidden layer vector based on a value of the hidden layer (51C) and a first hidden layer anomaly degree (da2) based on the first hidden layer vector, and is capable of generating an input / output error (da1) by inputting the second output data (do2) to the second machine learning model (80b) based on the value of the hidden layer (51C). a second hidden layer vector based on the first hidden layer vector and a second hidden layer anomaly degree (db1) based on the second hidden layer vector; a first hidden layer error (da3) can be calculated by a loss function based on the first hidden layer vector and the second hidden layer vector; and a second hidden layer error (da4) can be calculated by the loss function based on the first hidden layer anomaly degree (da2) and the second hidden layer anomaly degree (db1), and the accuracy data (da) includes at least one of the input / output error (da1), the first hidden layer anomaly degree (da2), the first hidden layer error (da3), and the second hidden layer anomaly degree (db1) (third configuration).

[0083] The electronic device (100A) includes a machine learning device according to any one of the first to third configurations (fourth configuration).

[0084] An electronic device (100) according to a fourth configuration includes a display unit (100E) configured to display the first output data (do1) and the accuracy data (da) (fifth configuration).

[0085] The machine learning program (P) is for functioning as the machine learning device (6) according to any one of the first to third configurations (sixth configuration).

[0086] A simulation device (100) calculates first output data (do1) and accuracy data (da) using a machine learning device (6) according to any one of the first to third configurations (seventh configuration). [Explanation of symbols]

[0087] 6 Machine Learning Devices 7 Data storage unit 8 Model holding section 9 Arithmetic section 10. Anomaly detection unit 20 Model holding section 50 Three-layer neural network 50A input layer 50B Hidden layer 50C output layer 51 Three-layer neural network 51A Input layer 51B Hidden Layer 51C output layer 71 Input Data 72 Initial Value Data 80a Prediction Model 80b Validation Model 91 Learning calculation unit 92 Inference calculation section 93 Accuracy data calculation section 100 Computer (simulation device) 100A CPU (electronic equipment) 100B memory 100C auxiliary storage 100D Operation input section 100E Display section Ap Abnormal location L loss function P Simulation program (machine learning program) T1 normal period T2 Abnormal period da Accuracy data da1 input / output error da2 1st hidden layer anomaly score (1st hidden layer anomaly score) da3 1st hidden layer error (1st intermediate layer error) da4 Second hidden layer error (second intermediate layer error) db1 Second hidden layer anomaly score (second intermediate layer anomaly score) do1 First output data do2 Second output data ha 1st hidden layer vector (1st intermediate layer vector) hb Second hidden layer vector (second intermediate layer vector)

Claims

1. a model storage unit configured to store a first machine learning model that has undergone supervised training and a second machine learning model that has undergone unsupervised training; a data storage unit configured to store input data to be input to the first machine learning model and the second machine learning model; a calculation unit configured to input the input data to the first machine learning model to generate first output data, and to input the input data to the second machine learning model to generate accuracy data; Equipped with The second machine learning model is an input layer; an output layer; at least one hidden layer disposed between the input layer and the output layer; Including, A machine learning device in which the accuracy data is calculated using at least one of the values ​​of the input layer, the intermediate layer, and the output layer so that the tendency shifts in response to the first output data.

2. The machine learning device according to claim 1 , wherein the accuracy data includes an input / output error calculated by a loss function based on the values ​​of the input layer and the values ​​of the output layer.

3. The calculation unit generating second output data by inputting the input data into the second machine learning model and performing inference; and the input / output error can be calculated by a loss function based on the input data and the second output data; and generating a first hidden layer vector based on the hidden layer value and a first hidden layer anomaly degree based on the first hidden layer vector; inputting the second output data into the second machine learning model; A second hidden layer vector based on the hidden layer value and a second hidden layer anomaly degree based on the second hidden layer vector can be generated; and a first hidden layer error can be calculated based on the first hidden layer vector and the second hidden layer vector using a loss function; and a second intermediate layer error can be calculated in a loss function based on the first intermediate layer anomaly degree and the second intermediate layer anomaly degree; The machine learning device according to claim 2 , wherein the accuracy data includes at least one of the input / output error, the first hidden layer anomaly degree, the first hidden layer error, and the second hidden layer anomaly degree.

4. An electronic device comprising the machine learning device according to claim 1 .

5. The electronic device according to claim 4 , further comprising a display configured to display the first output data and the accuracy data.

6. A machine learning program for causing the machine learning device according to any one of claims 1 to 3 to function.

7. A simulation device that calculates the first output data and the accuracy data using the machine learning device according to claim 1 .

Images