Estimation device, estimation method, and program
The estimation device and method utilize prediction models to estimate item relationships based on prediction accuracy, eliminating the need for mathematical models and enabling efficient, dynamic correlation analysis.
Patent Information
- Application Number
- JP2023518558
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-05-06
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2041-05-06
AI Technical Summary
Existing methods require constructing a mathematical model to estimate relationships between items, which is burdensome and impractical in certain scenarios.
An estimation device and method that uses a prediction model to estimate relationships between items based on prediction accuracy, without the need for a mathematical model, by inputting past values and calculating predicted values to determine correlations.
Enables estimation of relationships between items efficiently and dynamically, reducing the burden of constructing mathematical models and allowing for real-time correlation analysis.
Smart Images

Figure 0007586302000005 
Figure 0007586302000006 
Figure 0007586302000007
Abstract
Description
[Technical field]
[0001] The present invention relates to an estimation device, an estimation method, and program Regarding. [Background technology]
[0002] When controlling objects such as devices and robots in a plant using artificial intelligence (AI), it is necessary to check the validity of the analysis results by the AI, and if valid, to control the objects according to the output of the AI. As one such example, a method has been proposed in which a mathematical formula is constructed that shows the relationship between multiple observation points in an observation object such as a plant, and the correlation between the observation points is analyzed. For example, Patent Document 1 describes a method for estimating the pipelines used for inflow and outflow at a water storage facility based on the water storage volume in the facility and the inflow and outflow volumes of each of multiple pipelines connected to the water storage facility.
[0003] In the method described in Patent Document 1, an equation is constructed for each of a number of times indicating that the total of the inflow and outflow of a pipeline is equal to the change in the amount of water stored in the water storage facility. At this time, the inflow and outflow of the pipeline is multiplied by a coefficient whose value is unknown. Then, in the method described in Patent Document 1, the coefficient value of the inflow and outflow of the pipeline is determined so that the sum of the squared error between the total of the inflow and outflow of the pipeline and the change in the amount of water stored in the water storage facility for all times is minimized. In the method described in Patent Document 1, a pipeline with a determined coefficient value close to -1 or 1 is estimated to be the pipeline used for inflow and outflow of the water storage facility. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] International Publication No. 2016 / 027528 Summary of the Invention [Problem to be solved by the invention]
[0005] If it were possible to estimate the relationships between multiple items without having to construct a mathematical model that shows the relationships between these items, the burden on personnel in constructing the mathematical model could be reduced, and the relationships could be estimated even in cases where it is not possible to construct a mathematical model.
[0006] An example of the object of the present invention is to provide an estimation device, an estimation method, and program The purpose of this project is to provide [Means for solving the problem]
[0007] According to a first aspect of the present invention, an estimation device includes a prediction model that receives past values of a plurality of items or past values of some of the items and outputs a predicted value of one or more of the items. estimates a relationship between the first item and the second item based on a prediction accuracy of a predicted value of the first item output for an input of a past value in a combination of items including the first item and a second item, and a prediction accuracy of a predicted value of the first item output by the prediction model for an input of a past value of an item excluding the second item from the combination of items. .
[0008] According to a second aspect of the present invention, in the estimation method, a computer generates a prediction model that receives past values of a plurality of items or past values of some of the items and outputs a predicted value of one or more of the items. estimates a relationship between the first item and the second item based on a prediction accuracy of a predicted value of the first item output for an input of a past value in a combination of items including the first item and a second item, and a prediction accuracy of a predicted value of the first item output by the prediction model for an input of a past value of an item excluding the second item from the combination of items. This includes:
[0009] According to a third aspect of the present invention, a program causes a computer to generate a prediction model that receives past values of a plurality of items or past values of some of the items and outputs a predicted value of one or more of the items. estimates a relationship between the first item and the second item based on a prediction accuracy of a predicted value of the first item output for an input of a past value in a combination of items including the first item and a second item, and a prediction accuracy of a predicted value of the first item output by the prediction model for an input of a past value of an item excluding the second item from the combination of items. It is a program that performs processing. Effect of the Invention
[0010] According to the present invention, it is possible to estimate the relationships between a plurality of items without the need to construct a mathematical model showing the relationships between these items. [Brief description of the drawings]
[0011] [Figure 1] 1 is a schematic block diagram showing an example of the configuration of a system according to a first embodiment. [Diagram 2] 1 is a diagram illustrating an example of the configuration of an estimation device according to a first embodiment. [Diagram 3] FIG. 2 is a diagram showing a first example of a prediction model according to the first embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of a processing procedure in which the estimation device according to the first embodiment estimates a correlation between items based on prediction accuracy by a prediction model. [Diagram 5] FIG. 11 is a diagram showing a second example of a prediction model according to the first embodiment. [Figure 6] FIG. 11 is a diagram illustrating an example of a processing procedure in which the estimation device according to the first embodiment estimates correlations between items by using a prediction model including weighting nodes for input items. [Figure 7] FIG. 2 is a diagram illustrating an example of a configuration in which a model processing unit according to the first embodiment assigns weights to input items to a prediction model. [Figure 8] FIG. 11 is a diagram illustrating an example of a processing procedure in which the estimation device estimates correlations between items when the model processing unit according to the first embodiment weights input items to a prediction model. [Figure 9] FIG. 11 is a diagram illustrating an example of the configuration of an estimation device according to a second embodiment. [Figure 10] FIG. 13 is a diagram illustrating an example of the configuration of an estimation device according to a third embodiment. [Figure 11] FIG. 13 is a diagram illustrating an example of the configuration of an estimation device according to a fourth embodiment. [Figure 12] FIG. 13 is a diagram illustrating an example of a processing procedure in an estimation method according to the fifth embodiment. [Figure 13] FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0012] The following describes embodiments of the present invention, but the following embodiments do not limit the scope of the invention. Also, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.
[0013] First Embodiment 1 is a schematic block diagram showing an example of the configuration of a system according to the first embodiment. In the configuration shown in FIG.
[0014] The estimation target 900 is a target of estimation by the estimation device 100. The estimation target 900 has a plurality of items from which the estimation device 100 can obtain values, and the estimation device 100 estimates relationships between the plurality of items in the estimation target 900. In the following, a case will be described in which the estimation device 100 estimates correlations between a plurality of items in the estimation target 900. However, the relationships estimated by the estimation device 100 are not limited to specific ones. For example, the estimation device 100 may estimate a causal relationship. The same applies to the estimation device 200 and the estimation device 300. The items in the estimation target 900 are not limited to a specific type as long as the estimation device 100 can obtain the values of the items.
[0015] For example, the items in the estimation target 900 may include items to be measured by a sensor. When an item in the estimation target 900 is an item to be measured by a sensor, the item may be an item of the estimation target 900 itself, such as a current at a predetermined position in the estimation target 900. Alternatively, the item may be an item related to the surrounding environment of the estimation target 900, such as the room temperature of the room in which the estimation target 900 is installed.
[0016] Furthermore, an item in the estimation target 900 may include an operation for the estimation target 900, or a control command for the estimation target 900, or both. When an item in the estimation target 900 is an operation for the estimation target 900, the amount of operation can be used as the item value. When an item in the estimation target 900 is a control command for the estimation target 900, the control command value can be used as the item value. The items in the estimation target 900 are also simply referred to as items.
[0017] The estimation device 100 acquires past values of items in the estimation target 900, and predicts item values in the future from the time when the past values were observed. The past values here are values at times earlier than the time of the prediction target. The predicted item values are also called predicted values.
[0018] In the following, times are represented by consecutive numbers for each time step, such as 1, . . . , t-1, t, t+1. The estimation device 100 acquires past values for each time step, and predicts item values at times in the time step that are future than the times at which the past values were observed. For example, the estimation device 100 acquires time-series data of item values from times 1, . . . , t-1, t, and predicts item values at time t+1. The time does not necessarily have to be expressed in one step, and may be expressed by time steps including mutually different step intervals, such as a mixture of two steps and one step. Alternatively, the time may be expressed by a continuous value. The way of expressing the time is not limited to the above-mentioned example. For convenience of explanation, it is assumed that the time series data of the item value is acquired for each time from time 1, ..., t-1, to t.
[0019] The estimation target 900 can be various items having a plurality of items for which the estimation device 100 can obtain past values, and is not limited to a specific one. For example, the estimation target 900 may be any one of a chemical plant, a distillation tower, an air traffic control system, and a railway system, but is not limited thereto. Furthermore, the estimation target 900 may be a system including multiple devices, a single device, or a part of a device.
[0020] The estimation device 100 calculates a predicted value using a prediction model that receives a past value as an input and outputs a predicted value. The estimation device 100 also estimates a correlation between items using this prediction model. An item that is an input to a prediction model is also called an input item. An item that is an output from a prediction model is also called an output item.
[0021] The prediction model is not limited to a specific type of model as long as it is a model that can be learned. For example, the prediction model may be configured using a machine learning algorithm such as a neural network or a support vector machine, but is not limited thereto. The estimation device 100 is configured using, for example, a computer.
[0022] Fig. 2 is a diagram showing an example of the configuration of the estimation device 100. In the configuration shown in Fig. 2, the estimation device 100 includes a communication unit 110, a display unit 120, an operation input unit 130, a storage unit 180, and a control unit 190. The control unit 190 includes a model processing unit 191 and an estimation unit 192.
[0023] The communication unit 110 communicates with other devices. For example, the communication unit 110 may communicate with the estimation target 900 and receive past values from the estimation target 900 for each time in a time step. Alternatively, the communication unit 110 may receive training data based on past values of items in the estimation target 900 from a server device that provides the training data. This training data may be used to learn a prediction model.
[0024] The display unit 120 includes a display screen such as a liquid crystal panel or an LED (Light Emitting Diode) panel, and displays various images. For example, the display unit 120 may display the correlation estimation result by the estimation device 100. The operation input unit 130 includes input devices such as a keyboard and a mouse, and receives user operations. For example, the operation input unit 130 may receive a user operation that instructs estimation of correlations between items.
[0025] The storage unit 180 stores various data. The storage unit 180 is configured using a storage device included in the estimation device 100. The control unit 190 performs various processes by controlling each unit of the estimation device 100. The functions of the control unit 190 may be performed by a CPU (Central Processing Unit) included in the estimation device 100 reading and executing a program from the storage unit 180.
[0026] The model processing unit 191 performs various processes related to the prediction model. The prediction model here is a model that receives past values of multiple items or past values of some of those items as input and outputs predicted values of one or more items. The output items may be the same as the input items or may be different. The model processing unit 191, for example, inputs past values to a prediction model to calculate a predicted value. The model processing unit 191 also controls learning of the prediction model. The model processing unit 191 corresponds to an example of a model processing means. The estimation unit 192 estimates correlations between items using a prediction model. The estimation unit 192 corresponds to an example of an estimation means.
[0027] The estimation unit 192 may estimate the correlation between items based on the prediction accuracy of a prediction model. Fig. 3 is a diagram showing a first example of a prediction model. Fig. 3 shows an example in which a prediction model receives past values of three items, X, Y, and Z, and outputs predicted values of the items X, Y, and Z. The values of the items X, Y, and Z may be sensor measurement values obtained by sensors X, Y, and Z provided on the estimation target 900.
[0028] However, the number of input items to the prediction model is not limited to a specific number as long as it is two or more. In addition, the prediction model may output predicted values for only some of the input items. In addition, the input items to the prediction model are not limited to items from sensors.
[0029] In the example in Figure 3, the prediction model is based on the time series data X of the past values of item X. t (k) and time series data Y of past values of item Yt (l) and time series data Z of past values of item Z t (m) The input is received as: X t (k) represents the time series data of the past values of item X from time t to k steps (k is a natural number) before that. X t (k) =X t-k , X t Similarly, Y t (l) =Y t-l , Y t (l is a natural number). Z t (m) =Z t-m , Z t (m is a natural number). k, l, and m may be the same value or may be different values. The prediction model is X t (k) , Y t (l) and Z t (m) For the input, the predicted value X of item X at time t+1 is t+1 and the predicted value Y of item Y at time t+1 t+1 and the predicted value Z of item Z at time t+1 t+1 and output.
[0030] The estimation unit 192 estimates correlations between items based on differences in prediction accuracy due to differences in input items to the prediction model. For example, the model processing unit 191 receives time series data X t (k) and Y t (l) The predicted value X when both are input into the prediction model t+1 And, X t (k) and Y t (l) X of t (k) Predicted value X when only t+1 The estimation unit 192 calculates the time series data X t(k) and Y t (l) When both are input into the prediction model, and when X t (k) and Y t (l) X of t (k) Based on the difference in prediction accuracy when only X and Y are input into the prediction model, the presence or absence of a correlation between items X and Y is estimated. Time series data X t (k) and Y t (l) The predicted value X when both are input into the prediction model t+1 is expressed as equation (1).
[0031]
number
[0032] Time series data X t (k) and Y t (l) X of t (k) Predicted value X when only t+1 is expressed as equation (2).
[0033]
number
[0034] Time series data X t (k) and Y t (l) In the case of inputting both into the forecast model, and in the case of inputting the time series data X t (k) and Y t (l) X of t (k) Since the same prediction model is used in both cases where only the α-value is input to the prediction model, the prediction model is expressed by the function f in both equations (1) and (2). The predicted value X is smaller in the case of formula (1) than in the case of formula (2). t+1 If the prediction accuracy of is high, it can be assumed that there is a correlation between items X and Y. On the other hand, if the predicted value X is different between the cases of formula (1) and formula (2), t+1 If the predictive accuracy of items X and Y is the same, we can assume that items X and Y are uncorrelated. Uncorrelated here means that there is no correlation.
[0035] For example, the estimation unit 192 estimates the time series data X t (k) and Y t (l) The predicted value X when both are input into the prediction model t+1 The prediction accuracy of X t (k) and Y t (l) X of t (k) Predicted value X when only t+1 The estimation unit 192 then compares the calculated difference with a predetermined threshold, and when it is determined that the difference is equal to or greater than the threshold, it estimates that there is a correlation between the items X and Y. On the other hand, when it is determined that the difference is smaller than the threshold, the estimation unit 192 estimates that there is no correlation between the items X and Y. In this case, the predetermined threshold represents a criterion for determining whether or not there is a correlation between the items.
[0036] The prediction accuracy here is not limited to a specific type of prediction accuracy as long as it is comparable. For example, the estimation unit 192 may calculate, as the prediction accuracy, the magnitude of the error between the predicted value calculated by the model processing unit 191 and the predicted value presented as the correct answer in the training data. In this case, the estimation unit 192 evaluates that the smaller the magnitude of the error, the higher the prediction accuracy.
[0037] Alternatively, the model processing unit 191 may calculate predicted values at multiple times for each of multiple combinations of input items to the prediction model. Then, the estimation unit 192 may compare the magnitude of the error between the predicted value and the correct answer with a predetermined threshold for each combination of input items and for each time to determine whether the prediction is correct or incorrect. Then, the estimation unit 192 may calculate the accuracy rate for each combination of input items and use it as the prediction accuracy.
[0038] Fig. 4 is a diagram showing an example of a processing procedure in which the estimation device 100 estimates a correlation between items based on the prediction accuracy of a prediction model. Fig. 4 shows an example in which the model processing unit 191 calculates a predicted value of a first item, and the estimation unit 192 estimates the correlation between the first item and a second item. In the process of FIG. 4, the model processing unit 191 inputs past values for a combination of items including a first item and a second item into a prediction model, and calculates a predicted value of the first item (step S11). The estimation unit 192 compares the predicted value of the first item calculated by the model processing unit 191 in step S11 with the correct answer data, and calculates the prediction accuracy of the first item (step S12).
[0039] Furthermore, the model processing unit 191 inputs the past values of the items excluding the second item from the combination of items in step S11 into the prediction model, and calculates a predicted value of the first item (step S13). The estimation unit 192 calculates the prediction accuracy of the first item calculated by the model processing unit 191 in step S13 (step S14).
[0040] Then, the estimation unit 192 compares the prediction accuracy calculated in step S12 with the prediction accuracy calculated in step S14, and estimates the correlation between the first item and the second item (step S15). For example, the estimation unit 192 calculates a difference by subtracting the prediction accuracy calculated in step S14 from the prediction accuracy calculated in step S12. Then, the estimation unit 192 compares the calculated difference with a predetermined threshold value to determine whether the prediction accuracy calculated in step S12 is higher than the threshold value. If it is determined that the accuracy is higher than the threshold value, the estimation unit 192 estimates that there is a correlation between the first item and the second item. On the other hand, if it is determined that the accuracy is not higher than the threshold value, the estimation unit 192 estimates that there is no correlation between the first item and the second item. After step S15, the estimation device 100 ends the process in FIG.
[0041] The estimation unit 192 may estimate the strength of the correlation in addition to or instead of the presence or absence of a correlation between items. For example, in step S15, the estimation unit 192 may output the difference between the prediction accuracy calculated in step S12 and the prediction accuracy calculated in step S14 as an index value indicating the strength of correlation.
[0042] If the prediction accuracy of the value of the first item is higher when the second item is added to the input items than when it is not added, the estimation unit 192 may infer that there is a causal relationship in which the second item is a cause and the first item is an effect. For example, in step S15, if the estimation unit 192 determines that the prediction accuracy calculated in step S12 is higher than the prediction accuracy calculated in step S14 by a threshold or more, it may estimate that there is a causal relationship in which the second item is the cause and the first item is the effect.
[0043] The estimation unit 192 may estimate the correlation between items based on weighting of the input items in the prediction model. Fig. 5 is a diagram showing a second example of a prediction model. Fig. 5 shows an example of a prediction model including a weighting node for an input item. Fig. 5 also shows an example in which the prediction model receives input of past values of two items, items X and Y, and outputs a predicted value of item X.
[0044] However, as described with reference to Fig. 3, the number of input items to the prediction model is not limited to a specific number as long as it is two or more. In addition, the prediction model may output predicted values for only some of the input items, or may output predicted values for all of the input items. In addition, the input items to the prediction model are not limited to a specific type of item, such as an item from a sensor.
[0045] In the example of Figure 5, nodes N11 and N12 are input nodes. Nodes N21 and N22 are weight nodes. Nodes N31 and N32 are weighting nodes. Node N41 is an output node. In the layer between nodes N31 and N32 and node N41, processing may be performed on the output of node N31 and the output of node N32.
[0046] Node N11 is the time series data X of the past values of item X. t (k) The node N11 may output the past value of the item X as is, or the past value of the item X may be processed in the node N11. Node N12 is the time series data Y of the past values of item Y. t (l) The node N12 may output the past value of the item Y as is, or the past value of the item Y may be processed in the node N12.
[0047] Node N21 stores the weight for the past value of item X. In the example of FIG. Node N22 stores the weight for the past value of item Y. In the example of FIG. Node N31 weights the past values of item X by multiplying the value of node N11 by the value of node N21. Node N32 weights the past values of item X by multiplying the value of node N12 by the value of node N22.
[0048] As in the example of Figure 5, when the weighting node stores a weight of either 1 or 0, the weighting by the weighting node acts as a mask for the input data. In the example of Figure 5, the value of node N21 is 1, so the time series data X of the past values of item X t (k) is the predicted value X of item X t+1 In contrast, since the value of node N22 is 0, the time series data Y t (l) is the predicted value X of item X t+1 It is not used in the calculation of
[0049] However, the weights stored in a weight node are not limited to 1 or 0. For example, a weight node may store a real value greater than or equal to 0 and less than or equal to 1 as a weight. There may be some constraint on the weights stored in a weight node, such as the sum of all weights stored in the weight node being 1.
[0050] When the model processing unit 191 learns a prediction model, the weights stored in the weight nodes are also included in the learning targets. The term "learning a prediction model" as used herein means adjusting parameter values of the prediction model based on training data. For example, the model processing unit 191 may adjust the parameter values of the prediction model based on training data. t (k) and time series data Y of past values of item Y t (l) and the predicted value X of item X t+1 Then, the model processing unit 191 adjusts the parameter values of the prediction model, such as the weights stored in the weighting nodes, so that the value of the node N41 approaches the correct value.
[0051] The weights in a trained prediction model can be thought of as representing the contribution of the weighted item to the calculation of the predicted value. The greater the contribution of an item, the stronger the correlation with the output item for which the predicted value is calculated. Therefore, the estimation unit 192 estimates the correlation between items based on the weights in the trained prediction model.
[0052] As in the example of Figure 5, when the weights take values of 1 or 0, the estimation unit 192 may estimate that an input item weighted with a weight of 1 is correlated with the output item, and an input item weighted with a weight of 0 is uncorrelated with the output item. Alternatively, when the weight is a real value greater than or equal to 0 and less than or equal to 1, the estimation unit 192 may compare the weight with a predetermined threshold value. Then, the estimation unit 192 may estimate that an input item weighted with a weight equal to or greater than the threshold value is correlated with an output item, and that an input item weighted with a weight smaller than the threshold value is uncorrelated with the output item.
[0053] The estimation unit 192 may estimate the strength of the correlation in addition to or instead of the presence or absence of a correlation between items. For example, the estimation unit 192 may output the weight for each input item as an index value indicating the strength of correlation between the input item and the output item.
[0054] FIG. 6 is a diagram illustrating an example of a processing procedure in which the estimation device 100 estimates correlations between items by using a prediction model including weighting nodes for input items. 6, the model processing unit 191 learns a prediction model using training data (step S21). The model processing unit 191 learns a prediction model by including weights assigned to input items in the learning target.
[0055] Next, the estimation unit 192 estimates the correlation between items based on the weights in the trained prediction model (step S22). For example, when the weights are 1 or 0, the estimation unit 192 estimates that an input item weighted with a weight of 1 is correlated with an output item, and an input item weighted with a weight of 0 is uncorrelated with an output item. After step S22, the estimation device 100 ends the process in FIG.
[0056] Instead of weighting the input items within the prediction model, the model processing unit 191 may weight the input items to the prediction model. 7 is a diagram showing an example of a configuration in which the model processing unit 191 weights input items to a prediction model. The configuration shown in FIG.
[0057] FIG. 7 shows a case where the model processing unit 191 processes the time series data X of past values of items X, Y, and Z. t (k) , Y t (l) , Z t (m) 7 shows an example in which weighting is applied to the prediction model and input to the prediction model. In addition, in the example of FIG. 7, the prediction model uses the predicted value X t+1 , Y t+1 and Z t+1 Output.
[0058] However, as described with reference to Figures 3 and 5, the number of input items to the prediction model is not limited to a specific number as long as it is two or more. In addition, the prediction model may output predicted values for only some of the input items, or may output predicted values for all of the input items. In addition, the input items to the prediction model are not limited to a specific type of item, such as an item from a sensor.
[0059] In the example of Figure 7, the time series data X t (k) , Y t (l) , Z t (m) are input to the weight matrix via a recurrent neural network (RNN) and a normalizer, respectively. The recurrent neural network is provided to reflect data at multiple times included in the time series data in the input values to the prediction model.
[0060] The normalizer performs normalization to align the range of possible input values to the weight matrix between input items. As a method for the normalizer to perform normalization, a batch normalization method in a convolutional neural network (CNN) may be used, but is not limited to this.
[0061] The weighting matrix weights the input data to the prediction model for each input item and for each prediction model. The weighting matrix τ in the example of FIG.
[0062]
number
[0063] Each element of the matrix is used as a weight for the prediction model. The subscripts of the element names indicate the names of the input and output items in lower case. For example, weight t xx is the time series data X of the past values of item X t (k) is multiplied as a weight for the data passed through the recurrent neural network and the normalizer. xx The weighted data is the predicted value X of item X. t+1 The input is then fed into a predictive model that outputs:
[0064] Weight t yz is the time series data Y of the past values of item Y t (l) is multiplied as a weight for the data passed through the recurrent neural network and the normalizer. yz The weighted data is the predicted value Z of item Z. t+1 The input is then fed into a predictive model that outputs:
[0065] The weights, which are the elements of the weighting matrix, are set by the model processing unit 191. The model processing unit 191 may randomly set the value of each element of the weighting matrix within a real number range of 0 to 1. In this case, the range of the elements is expressed by equation (4).
[0066]
number
[0067] In expression (4), "*" denotes a wildcard that can take any of the values "x", "y", and "z".
[0068] The model processing unit 191 repeats learning of the prediction model using the same training data every time the weight matrix is set. An input item for which a large weight value is set in learning with high prediction accuracy of the predicted value is considered to have a high correlation with the output item. Therefore, the estimation unit 192 may rank the learning results in descending order of prediction accuracy of the predicted value for each output item. Then, the estimation unit 192 may estimate that the correlation between the input item, for which a large weight is set in the learning result with high prediction accuracy, and the output item is strong.
[0069] For example, the estimation unit 192 may select, for each output item, a learning result whose prediction accuracy is higher than a predetermined threshold, and calculate an average value in all learning results for each weight of input data to a prediction model that outputs a predicted value of the output item.The estimation unit 192 may then use the calculated average value as an index value indicating the strength of correlation between the output item and the input item.
[0070] Further, for example, the estimation unit 192 may estimate a predicted value X t+1 The estimation unit 192 selects a learning result in which the estimation accuracy of the predicted value X t+1 Weights t for input data to the predictive model that outputs xx , t yx , t zxThe estimation unit 192 calculates an average value in the selected learning result for each of the input items and the output items. The estimation unit 192 calculates the calculated average value as an index value indicating the strength of the correlation between the input items and the output items. Alternatively, the estimation unit 192 may compare the calculated average value with a predetermined threshold value, and if the average value is equal to or greater than the threshold value, estimate that there is a correlation between the input items and the output items, and if the average value is less than the threshold value, estimate that there is no correlation between the input items and the output items.
[0071] FIG. 8 is a diagram showing an example of a processing procedure in which the estimation device 100 estimates correlations between items when the model processing unit 191 weights input items to a prediction model. In the process of FIG. 8, the estimation device 100 starts a loop L11 for repeatedly learning a prediction model (step S31).
[0072] Next, the model processing unit 191 sets the value of each element of the weighting matrix (step S32). The elements of the weighting matrix are used as weights for input data to the prediction model. The model processing unit 191 may randomly set the value of each element of the weighting matrix within a real value range of 0 to 1.
[0073] Next, the model processing unit 191 learns the prediction model (step S33). Every time the model processing unit 191 executes the process of step S33, the model processing unit 191 may initialize parameter values of the prediction model and then learn the prediction model using the same training data. Once the recurrent neural network and the normalizer are set, they can be used in common for each learning. Therefore, there is no need to perform learning for the recurrent neural network and the normalizer for each processing of step S33.
[0074] Next, the estimation unit 192 calculates the prediction accuracy of the predicted value for each output item based on the learning result in step S33 (step S34). Then, the estimating device 100 performs a termination process of the loop L11 (step S35). The estimating device 100 repeats the loop L11 a predetermined number of times.
[0075] After the estimation device 100 finishes the process of loop L11, the estimation unit 192 estimates the correlation between items (step S36). For example, the estimation unit 192 ranks the learning results in descending order of prediction accuracy of the predicted values for each output item as described above, and estimates the correlation between items based on the weights of the learning results with high prediction accuracy. After step S36, the estimation device 100 ends the process in FIG.
[0076] As described above, the estimation unit 192 estimates relationships between items using a prediction model that receives past values of multiple items or past values of some of those items as input and outputs predicted values of one or more items. The estimation device 100 can learn this prediction model. In this respect, the estimation device 100 can estimate the relationship between multiple items without the need to construct a mathematical model that shows the relationships between these items.
[0077] In addition, the estimation unit 192 estimates the relationship between the first item and the second item based on the prediction accuracy of the predicted value of the first item output by the prediction model in response to input of past values in a combination of items including the first item and the second item, and the prediction accuracy of the predicted value of the first item output by the prediction model in response to input of past values of items excluding the second item from the combination of items.
[0078] The estimation device 100 can learn this prediction model. In this respect, the estimation device 100 can estimate the relationship between multiple items without the need to construct a mathematical model that shows the relationships between these items. Moreover, according to the estimation device 100, the relationship between items can be estimated using a trained prediction model, and since there is no need to perform learning each time to estimate the correlation, the relationship can be estimated in a relatively short time. Furthermore, the estimation device 100 can repeatedly estimate a relationship by receiving fluid data as input, and in this respect can estimate a dynamically changing relationship.
[0079] The prediction model includes a weighting node that weights past values with a weight for each item. The learning of the prediction model includes learning the weight for each item. The estimation unit 192 estimates the relationship between items based on the weight in the learned prediction model. According to the estimation device 100, the relationship between items can be estimated by learning a predictive model. In this respect, the relationship between multiple items can be estimated without the need to construct a mathematical model that shows the relationships between these items.
[0080] Furthermore, the model processing unit 191 assigns multiple weights to multiple items and learns predictions of values of the same items using the prediction model with each weighting. The estimation unit 192 estimates relationships between items based on the prediction accuracy of values of the same items using the trained prediction model with each weighting. According to the estimation device 100, the relationship between items can be estimated by learning a predictive model. In this respect, the relationship between multiple items can be estimated without the need to construct a mathematical model that shows the relationships between these items.
[0081] <Second embodiment> In the second embodiment, an example of a method of using the estimation device 100 according to the first embodiment will be described. Fig. 9 is a diagram showing an example of the configuration of an estimation device according to the second embodiment. In the configuration shown in Fig. 9, the estimation device 200 includes a communication unit 110, a display unit 120, an operation input unit 130, a storage unit 180, and a control unit 290. The control unit 290 includes a model processing unit 191, an estimation unit 192, and an explanation unit 291.
[0082] 9 that have the same configuration as the corresponding parts in FIG. 2 are given the same reference numerals (110, 120, 130, 180, 190, 191, 192) and detailed description thereof will be omitted here. The estimation device 200 differs from the estimation device 100 in that the control unit 290 further includes an explanation unit 291 in addition to the units of the control unit 190 in Fig. 2. In other respects, the estimation device 200 is similar to the estimation device 100. The estimation device 200 corresponds to an example of the estimation device 100. The estimation device 200 can be used in place of the estimation device 100 in the configuration in Fig. 1.
[0083] The explanation unit 291 explains the validity of the operation related to the item based on the estimation result of the estimation unit 192. The explanation unit 291 corresponds to an example of an explanation means. For example, consider a case where the estimation target 900 is a distillation tower, and the guide system of the distillation tower guides an operator to operate a boiler in order to make the concentration of the product obtained by distillation a predetermined concentration. In this case, when the relationship between operating the boiler and changing the concentration of the product is not clear, it is preferable to be able to explain the appropriateness of operating the boiler to the operator. If the estimation unit 192 estimates that there is a correlation between operating the boiler and changing the concentration of the product, this estimation result can be presented to the operator as an explanation of the appropriateness of operating the boiler.
[0084] The estimation device 200 may control the estimation target based on the result of estimating the correlation. For example, the display unit 120 may display an estimation result of the correlation between items and a screen inquiring whether or not to adopt the correlation of the estimation result, and the operation input unit 130 may receive a user operation to answer whether or not to adopt the correlation of the estimation result. When the operation input unit 130 receives a user operation indicating that the correlation of the estimation result is to be adopted, the control unit 290 may control the estimation target 900 based on the estimation result.
[0085] In the above example of the distillation tower, the display unit 120 may display a screen informing the user that there is a correlation between operating the boiler and changing the concentration of the product, and asking the user whether to adopt the estimated result of this correlation. When the operation input unit 130 receives a user operation indicating that the estimated result is to be adopted, the control unit 290 may automatically control the boiler.
[0086] Alternatively, the estimation device 200 may directly control the estimation target 900 based on the result of the correlation estimation, without prompting the user for confirmation. In the above distillation column example, the control unit 290 may automatically control the boiler based on an estimation that there is a correlation between operating the boiler and changing the product concentration.
[0087] As described above, the explanation unit 291 explains the validity of the operation related to the item based on the estimation result of the estimation unit 192. This allows the estimating device 200 to explain the validity of an operation without the need to qualitatively analyze the relationships between items.
[0088] <Third embodiment> In the third embodiment, another example of the method of using the estimation device 100 according to the first embodiment will be described. Fig. 10 is a diagram showing an example of the configuration of an estimation device according to the third embodiment. In the configuration shown in Fig. 10, an estimation device 300 includes a communication unit 110, a display unit 120, an operation input unit 130, a storage unit 180, and a control unit 390. The control unit 390 includes a model processing unit 191, an estimation unit 192, and a priority adjustment unit 391.
[0089] 10 that correspond to the respective parts in FIG. 2 and have the same configuration are given the same reference numerals (110, 120, 130, 180, 190, 191, 192), and detailed description thereof will be omitted here. The estimation device 300 differs from the estimation device 100 in that a control unit 390 further includes a priority adjustment unit 391 in addition to the units of the control unit 190 in Fig. 2. In other respects, the estimation device 300 is similar to the estimation device 100. The estimation device 300 corresponds to an example of the estimation device 100. The estimation device 300 can be used in place of the estimation device 100 in the configuration in Fig. 1.
[0090] The priority adjustment unit 391 adjusts the priority of the agent's behavior in reinforcement learning for the estimation target 900, which is the output source of past values, based on the estimation result of the estimation unit 192. For example, the priority adjustment unit 391 sets a high priority for an operation that the estimation unit 192 estimates to have a correlation with an item used in the reward function, thereby making it easier for the agent to select this operation as an action. The priority adjustment unit 391 corresponds to an example of a priority adjustment means.
[0091] When a learning device such as a control device for the inference target 900 learns operations for the inference target 900 by reinforcement learning, a combination of operations for the inference target 900 can be considered as the agent's actions. When there are many operable parts in the inference target 900, the actions that the agent can take increase exponentially. When the proportion of actions that can earn a reward among the many actions is small, learning may not progress.
[0092] In response to this, the priority adjustment unit 391 sets a high priority to the operation that the estimation unit 192 estimates to have a correlation with the items used in the reward function, making it easier for the agent to select this operation as an action. This increases the frequency with which the agent can obtain rewards, and is expected to help the agent progress in learning. Furthermore, by having the estimation unit 192 estimate the correlation between items online in parallel with the reinforcement learning, the priority adjustment unit 391 can update the priorities in response to dynamic changes in the correlation during learning, and it is expected that the learning device can learn efficiently.
[0093] As described above, the priority adjustment unit 391 adjusts the priority of the action of the agent in the reinforcement learning for the estimation target 900 that is the output source of the past value, based on the estimation result of the estimation unit 192. This enables the estimating device 300 to preferentially cause the agent to perform actions that are more likely to result in a reward, which is expected to promote learning.
[0094] <Fourth embodiment> 11 is a diagram illustrating an example of the configuration of an estimation device according to the fourth embodiment. In the configuration illustrated in FIG. In such a configuration, the estimation unit 611 estimates relationships between items using a prediction model that receives past values of multiple items or past values of some of those items as input and outputs predicted values of one or more items. The estimation unit 611 corresponds to an example of an estimation means.
[0095] The estimation device 610 can learn this prediction model. In this respect, the estimation device 610 can estimate the relationship between multiple items without the need to construct a mathematical model that shows the relationship between these items. The estimation unit 611 can be realized, for example, by using the functions of the estimation unit 192 shown in FIG.
[0096] <Fifth embodiment> Fig. 12 is a diagram showing an example of a processing procedure in the estimation method according to the fifth embodiment. The method shown in Fig. 12 includes performing estimation (step S611). In making the estimation (step S611), the computer estimates the relationships between items using a predictive model that receives past values of multiple items or past values of some of those items as input and outputs predicted values of one or more items. According to the estimation method shown in FIG. 12, by learning this prediction model, it is possible to estimate the relationship between multiple items without the need to construct a mathematical model that indicates the relationships between these items.
[0097] FIG. 13 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. In the configuration shown in FIG. 13, a computer 700 includes a CPU 710, a main memory device 720, an auxiliary memory device 730, an interface 740, and a non-volatile recording medium 750.
[0098] Any one or more of the above-mentioned estimation devices 100, 200, 300, and 610, or a part thereof, may be implemented in the computer 700. In this case, the operation of each of the above-mentioned processing units is stored in the auxiliary storage device 730 in the form of a program. The CPU 710 reads the program from the auxiliary storage device 730, develops it in the main storage device 720, and executes the above-mentioned processing according to the program. The CPU 710 also secures a storage area corresponding to each of the above-mentioned storage units in the main storage device 720 according to the program. Communication between each device and other devices is executed by the interface 740 having a communication function and performing communication according to the control of the CPU 710. The interface 740 also has a port for the non-volatile recording medium 750, and reads information from the non-volatile recording medium 750 and writes information to the non-volatile recording medium 750.
[0099] When the estimation device 100 is implemented in a computer 700, the operations of the control unit 190 and each of its units are stored in the form of a program in an auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing in accordance with the program.
[0100] Furthermore, CPU 710 allocates a storage area corresponding to storage unit 180 in main memory device 720 according to the program. Communication unit 110 performs communication by interface 740 having a communication function and performing communication according to control of CPU 710. Display unit 120 displays images by interface 740 having a display device and displaying images according to control of CPU 710. Reception of user operations by operation input unit 130 is performed by interface 740 having an input device and receiving user operations.
[0101] When the estimation device 200 is implemented in a computer 700, the operations of the control unit 290 and each of its units are stored in the form of a program in an auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing in accordance with the program.
[0102] Furthermore, CPU 710 allocates a storage area corresponding to storage unit 180 in main memory device 720 according to the program. Communication unit 110 performs communication by interface 740 having a communication function and performing communication according to control of CPU 710. Display unit 120 displays images by interface 740 having a display device and displaying images according to control of CPU 710. Reception of user operations by operation input unit 130 is performed by interface 740 having an input device and receiving user operations.
[0103] When the estimation device 300 is implemented in a computer 700, the operations of the control unit 390 and each of its units are stored in the form of a program in an auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it into the main storage device 720, and executes the above-mentioned processing in accordance with the program.
[0104] Furthermore, CPU 710 allocates a storage area corresponding to storage unit 180 in main memory device 720 according to the program. Communication unit 110 performs communication by interface 740 having a communication function and performing communication according to control of CPU 710. Display unit 120 displays images by interface 740 having a display device and displaying images according to control of CPU 710. Reception of user operations by operation input unit 130 is performed by interface 740 having an input device and receiving user operations.
[0105] When the estimation device 610 is implemented in the computer 700, the operation of the estimation unit 192 is stored in the form of a program in the auxiliary storage device 730. The CPU 710 reads the program from the auxiliary storage device 730, loads it in the main storage device 720, and executes the above-mentioned processing in accordance with the program.
[0106] Communication between the estimation device 610 and other devices is performed by the interface 740 having a communication function and operating under the control of the CPU 710. Interaction between the estimation device 610 and a user is performed by the interface 740 having an input device and an output device, presenting information to the user via the output device under the control of the CPU 710, and receiving user operations via the input device.
[0107] Any one or more of the above-mentioned programs may be recorded in non-volatile recording medium 750. In this case, interface 740 may read the program from non-volatile recording medium 750. Then, CPU 710 may directly execute the program read by interface 740, or may temporarily store the program in main storage device 720 or auxiliary storage device 730 and execute it.
[0108] Note that a program for executing all or part of the processing performed by the estimation devices 100, 200, 300, and 610 may be recorded in a computer-readable recording medium, and the program recorded in the recording medium may be read into a computer system and executed to perform the processing of each unit. Note that the term "computer system" here includes hardware such as an OS and peripheral devices. Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs (Read Only Memory), and CD-ROMs (Compact Disc Read Only Memory), as well as storage devices such as hard disks built into computer systems. The above-mentioned program may be for realizing part of the above-mentioned functions, or may be capable of realizing the above-mentioned functions in combination with a program already recorded in the computer system.
[0109] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and designs that do not deviate from the gist of the present invention are also included. [Industrial Applicability]
[0110] The present invention may be applied to an estimation device, an estimation method, and a recording medium. [Explanation of symbols]
[0111] 1 System 100, 200, 300, 610 estimator 110 Communications Department 120 Display section 130 Operation input section 180 Storage section 190, 290, 390 Control unit 191 Model processing section 192, 611 Estimation Department 291 Explanation Section 391 Priority adjustment section
Claims
1. an estimation means for estimating a relationship between a first item and a second item, based on a prediction accuracy of a predicted value of the first item output by the prediction model, which receives input of past values of a plurality of items or past values of some of the items and outputs a predicted value of one or more of the items, for an input of past values in a combination of items including a first item and a second item, and a prediction accuracy of a predicted value of the first item output by the prediction model for an input of past values of items excluding the second item from the combination of items; An estimation device comprising:
2. An explanation means for explaining the validity of an operation related to the item based on the estimation result of the estimation means. The estimation device of claim 1 further comprising:
3. a priority adjustment means for adjusting a priority of an action of an agent in reinforcement learning regarding an estimation target that is an output source of the past value based on an estimation result of the estimation means; The estimation device according to claim 1 or 2, further comprising:
4. The computer A prediction model that receives input of past values of a plurality of items or past values of some of the items and outputs predicted values of one or more of the items estimates a relationship between the first item and the second item based on the prediction accuracy of the predicted value of the first item that is output for an input of past values in a combination of items including a first item and a second item, and the prediction accuracy of the predicted value of the first item that the prediction model outputs for an input of past values of items excluding the second item from the combination of items. The estimation method includes:
5. On the computer, A prediction model that receives input of past values of a plurality of items or past values of some of the items and outputs predicted values of one or more of the items estimates a relationship between the first item and the second item based on the prediction accuracy of the predicted value of the first item that is output for an input of past values in a combination of items including a first item and a second item, and the prediction accuracy of the predicted value of the first item that the prediction model outputs for an input of past values of items excluding the second item from the combination of items. A program to carry out processing.
Citation Information
Patent Citations
Systems and methods for learning and predicting events
EP3564889A1
Water demand prediction method and system
JP2020112847A
Method and system for detecting correlation in data sets
US20110060703A1
Plant system information creating device
WO2016027528A1
