Data collection method and apparatus for improving confidence of virtual sensor

Abstract

A data collection method and apparatus capable of improving the reliability of a virtual sensor are disclosed. The disclosed data collection method for improving the reliability of a virtual sensor comprises: a step of estimating the reliability of a pre-trained virtual sensor model included in a first sensor system; a step of calculating a similarity value between a measurement value of a first low-cost sensor included in the first sensor system and a measurement value of a second low-cost sensor included in the second sensor system when the reliability is determined to be low; and a step of collecting the measurement value of the second low-cost sensor and the measurement value of the high-cost sensor included in the second sensor system when the similarity value is greater than or equal to a preset threshold.

Description

Technology Field

[0001] The present invention relates to a data collection method and apparatus for a virtual sensor, and more specifically, to a data collection method and apparatus capable of improving the reliability of a virtual sensor. Background Technology

[0003] Physical sensors are hardware devices designed to be installed in real-world environments to detect and measure physical phenomena, and they are widely used in various industrial fields for the purpose of data collection. The data collected through physical sensors is utilized to increase production output or to identify and resolve issues in the production process.

[0004] However, physical sensors incur high installation and maintenance costs, and depending on the installation environment, various limitations may arise, such as difficulties in access due to strong radiation. Due to these limitations of physical sensors, it is difficult to efficiently collect data at a low cost using only physical sensors.

[0005] As data collection through sensors is a major task across various industries, virtual sensor technology combining physical sensor data with artificial intelligence (AI) has emerged to complement the limitations of physical sensors and collect cost-effective, reliable data.

[0006] Virtual sensors, which emerged to overcome the various limitations of physical sensors, are a technology that replaces existing physical sensors by receiving measurement values ​​from other physical sensors and indirectly estimating the measurement value of the target sensor to be predicted. In particular, virtual sensors are also being utilized recently in metaverse environments based on sensor data.

[0007] Since virtual sensors replace physical sensors with predictions from artificial intelligence models trained on measurements from physical sensors, they eliminate not only the purchase costs of physical sensors but also the need to build infrastructure related to physical sensors, thereby significantly increasing cost efficiency. Furthermore, while physical sensors require replacement with technically improved devices as technology advances, virtual sensors only require software updates, offering advantages in terms of time and resource consumption.

[0008] As mentioned above, while virtual sensors offer various advantages over physical sensors, there are also drawbacks due to the reliance on artificial intelligence (AI) models for prediction. If the AI ​​model used in a virtual sensor has a capacity too small to accommodate the complexity of the training data, prediction accuracy may decrease. Furthermore, because AI models rely heavily on training data, there is a high probability of inaccurate predictions if the training data contains anomalies or if there is a significant discrepancy between the training data and the actual environment being predicted. Consequently, if the accuracy of the virtual sensor is low, its credibility also declines.

[0009] Much research is underway to improve the reliability of such virtual sensors. For example, algorithms are being developed to build thousands of virtual sensor models simultaneously to identify the most accurate model for a specific use case or to enhance prediction performance. However, since the prediction accuracy of deep learning models fundamentally improves with more training data, methods to effectively collect a sufficient amount of training data are required to increase the reliability of virtual sensors.

[0010] Relevant prior art includes Korean published patents No. 2023-0121504, No. 2009-0001148, and No. 2024-0108627. The problem to be solved

[0012] The present invention is intended to provide a data collection method and apparatus capable of increasing the reliability of a virtual sensor.

[0013] In particular, the present invention is intended to provide a method and apparatus for collecting training data of a virtual sensor in order to increase the reliability of the virtual sensor. means of solving the problem

[0015] According to one embodiment of the present invention for achieving the above-mentioned purpose, a data collection method for improving the reliability of a virtual sensor is provided, comprising: a step of estimating the reliability of a pre-learned virtual sensor model included in a first sensor system; a step of calculating a similarity value between a measurement value of a first low-cost sensor included in the first sensor system and a measurement value of a second low-cost sensor included in the second sensor system when the reliability is determined to be low; and a step of collecting the measurement value of the second low-cost sensor and the measurement value of the high-cost sensor included in the second sensor system when the similarity value is greater than or equal to a preset threshold.

[0016] In addition, according to another embodiment of the present invention for achieving the above-mentioned purpose, a data collection method for improving the reliability of a virtual sensor is provided, comprising: a step of estimating the reliability of a pre-learned virtual sensor model included in a first sensor system; a step of searching for a second sensor system including a second low-cost sensor that exhibits a measurement value similar to that of a first low-cost sensor included in the first sensor system when the reliability is determined to be low; and a step of collecting the measurement value of the second low-cost sensor and the measurement value of a high-cost sensor included in the second sensor system.

[0017] Additionally, according to another embodiment of the present invention for achieving the above-mentioned purpose, a data collection system for improving the reliability of a virtual sensor is provided, comprising: a first sensor system including a first low-cost sensor, a first high-cost sensor, and a pre-learned virtual sensor model; a second sensor system including a second low-cost sensor and a second high-cost sensor; and a data collection device for collecting measurement values ​​of the second low-cost sensor and the second high-cost sensor, wherein the data collection device estimates the reliability of the virtual sensor model, calculates a similarity value between the measurement value of the first low-cost sensor and the measurement value of the second low-cost sensor according to the reliability, and collects the measurement values ​​of the second low-cost sensor and the second high-cost sensor according to the similarity value. Effects of the invention

[0019] According to one embodiment of the present invention, data for retraining a virtual sensor is collected under conditions where the reliability of the virtual sensor is low, thereby reducing unnecessary data collection for retraining the virtual sensor. Consequently, maintenance costs for operating the virtual sensor can be reduced, and efficient operation of the virtual sensor is possible. Brief explanation of the drawing

[0021] FIG. 1 is a diagram illustrating a data collection system for improving the reliability of a virtual sensor according to an embodiment of the present invention. FIG. 2 is a diagram illustrating a data collection method for improving the reliability of a virtual sensor according to an embodiment of the present invention. FIG. 3 is a diagram illustrating a data collection method for improving the reliability of a virtual sensor according to another embodiment of the present invention. Specific details for implementing the invention

[0022] The present invention is susceptible to various modifications and may have various embodiments; specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the invention to specific embodiments, and it should be understood that the invention includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. Similar reference numerals have been used for similar components in the description of each drawing.

[0024] The present invention relates to a data collection method for improving the reliability of a virtual sensor. The collected data is used for retraining the virtual sensor, and through retraining, the reliability of the virtual sensor can be improved.

[0025] Although it is important to periodically retrain the virtual sensor to increase its reliability, continuously collecting data for the purpose of improving the reliability of the virtual sensor is inefficient; therefore, one embodiment of the present invention collects data for the retraining of the virtual sensor according to the reliability of the virtual sensor. The reliability of the virtual sensor can be estimated from the data output by the virtual sensor.

[0026] Since retraining of the virtual sensor is unnecessary when the reliability of the virtual sensor is high, according to one embodiment of the present invention, unnecessary data collection for retraining the virtual sensor can be reduced, and thus maintenance costs for virtual sensor operation can be reduced and efficient operation of the virtual sensor is possible.

[0027] Hereinafter, embodiments according to the present invention will be described in detail with reference to the attached drawings.

[0029] FIG. 1 is a diagram illustrating a data collection system for improving the reliability of a virtual sensor according to an embodiment of the present invention.

[0030] Referring to FIG. 1, a data collection system according to one embodiment of the present invention includes a first sensor system (110), a second sensor system (120), and a data collection device (130).

[0031] The first sensor system (110) includes a first low-cost sensor (111), a first high-cost sensor (112), and a pre-trained virtual sensor model (113). Here, the low-cost sensor (111) refers to a low-cost sensor, and the high-cost sensor (112) refers to a high-cost sensor. The criteria for distinguishing between the low-cost sensor and the high-cost sensor can be set in various ways depending on the embodiment. For example, a high-cost sensor whose performance, such as reliability, stability, and resolution, is above a preset threshold performance may be the high-cost sensor.

[0032] And the virtual sensor model (123) is a model trained to receive the measurement value of a low-cost sensor and predict the measurement value of a high-cost sensor, and the training data for the virtual sensor model (123) may include the measurement value of a first low-cost sensor (111) and the measurement value of a first high-cost sensor (112) measured in the same environment. The measurement value of the first high-cost sensor (112) may be labeled with the measurement value of the first low-cost sensor (111).

[0033] The second sensor system (120) includes a second low-cost sensor (121) and a second high-cost sensor (122).

[0034] The data collection device (130) collects the measurement values ​​of the second low-cost sensor (121) and the second high-cost sensor (122), and the collected measurement values ​​are used for retraining the virtual sensor model (120).

[0035] At this time, the data collection device (130) estimates the reliability of the virtual sensor model (113) for data collection, and calculates the similarity value between the measurement value of the first low-cost sensor (111) and the measurement value of the second low-cost sensor (121) according to the estimated reliability. Then, according to the calculated similarity value, it collects the measurement values ​​of the second low-cost sensor (121) and the second high-cost sensor (122).

[0036] In one embodiment, the data collection device (130) can calculate the cosine similarity value between the measurement value of the first low-cost sensor (111) and the measurement value of the second low-cost sensor (121) when the estimated reliability is low. And when the cosine similarity value is greater than or equal to a preset threshold, the measurement values ​​of the second low-cost sensor (121) and the second high-cost sensor (122) can be collected.

[0037] If the reliability of the virtual sensor model (113) is low, retraining of the virtual sensor model is required to increase reliability. In one embodiment of the present invention, by selectively collecting data for retraining when retraining is required, the maintenance cost of virtual sensor operation can be reduced and the virtual sensor can be operated efficiently.

[0038] In addition, for the retraining of the virtual sensor model, data collected in an environment similar to the first sensor system (110) is required, and since the measurement value of the first low-cost sensor (111) and the measurement value of the second low-cost sensor (121) are similar, it means that the environments of the first and second sensor systems (110, 120) are similar to each other, the data collection device (130) collects data from the second sensor system (120) by using the similarity of the measurement values ​​of the first and second low-cost sensors (111, 121).

[0040] FIG. 2 is a diagram illustrating a data collection method for improving the reliability of a virtual sensor according to an embodiment of the present invention.

[0041] A data collection method according to one embodiment of the present invention may be performed in a computing device including a memory and a processor, and the aforementioned data collection device may be an example of such a computing device.

[0042] Referring to FIG. 2, a computing device according to an embodiment of the present invention estimates the reliability (S210) of a pre-trained virtual sensor model included in a first sensor system. Here, the virtual sensor model may be a deep learning model to which weights of a plurality of artificial neural networks are applied, which is trained to receive a measurement value of a first low-cost sensor included in the first sensor system and output a predicted value for a measurement value of a first high-cost sensor included in the first sensor system. That is, the virtual sensor model may be a hybrid model derived from a plurality of artificial neural networks, which is a deep learning model with the same structure as the artificial neural network, but in which the weights of the artificial neural network are reflected according to a pre-set method. As an example, the virtual sensor model may be a model in which the average value of the weights of the artificial neural network is set as the weight.

[0043] In step S210, the computing device, as an example, can estimate the confidence level by comparing the confidence interval for the predicted value output by the virtual sensor model with a preset threshold interval. To this end, the computing device estimates the average value for the predicted values ​​of the artificial neural networks output by the artificial neural networks, and the computing device uses [Equation 1] to estimate the average value ( ) can be calculated.

[0044]

[0045] Here, M is the number of artificial neural networks, m is the index for the artificial neural network, represents the average value of the predicted values ​​output by each artificial neural network.

[0046] And the computing device estimates the variance of the prediction values ​​of the artificial neural networks using the average value, and, for example, using [Equation 2], the variance ( ) can be calculated.

[0047]

[0048] Here, represents the variance of the predicted values ​​output by each artificial neural network.

[0049] Then, the computing device calculates the confidence interval for the virtual sensor model's prediction using the estimated variance value.

[0050] [Mathematical Equation 2] can be expressed as [Mathematical Equation 3], where, is the aleatoric variance value, represents the epistemic variance value.

[0051]

[0052] Since the aleatoric variance is the variance due to uncertainty inherent in the data and the epistemic variance is the variance due to uncertainty of the model itself, the computing device can calculate the confidence interval using the epistemic variance. The computing device can calculate the confidence interval as shown in [Equation 4].

[0053]

[0054] Here, is the predicted value of the virtual sensor model, represents the epistemic variance, and Z represents the Z-score.

[0055] The computing device can estimate that the reliability of the virtual sensor model's prediction is low when the confidence interval length is greater than the critical interval length, and that the reliability of the virtual sensor model's prediction is high when the confidence interval length is less than or equal to the critical interval length.

[0056] If the computing device determines that the reliability estimated in step S210 is low, it calculates (S220) the similarity value between the measurement value of the first low-cost sensor included in the first sensor system and the measurement value of the second low-cost sensor included in the second sensor system.

[0057] And if the similarity value calculated in step S220 is greater than or equal to a preset threshold, the computing device collects the measurement value of the second low-cost sensor and the measurement value of the second high-cost sensor included in the second sensor system (S230).

[0058] As an example, the computing device periodically calculates a similarity value, and if the state in which the periodically calculated similarity value is above a threshold is maintained for a preset threshold time or longer, it collects a measurement value, and if the similarity value becomes smaller than the threshold while the measurement value is being collected, it may stop collecting.

[0059] When the computing device determines that the amount of collected data is sufficient for retraining the virtual sensor model, it can receive the measurement value of the second low-cost sensor as input and perform training on the virtual sensor model to predict the measurement value of the high-cost sensor included in the second sensor system.

[0061] FIG. 3 is a diagram illustrating a data collection method for improving the reliability of a virtual sensor according to another embodiment of the present invention.

[0062] Referring to FIG. 3, a computing device according to an embodiment of the present invention estimates the reliability of a pre-learned virtual sensor model included in a first sensor system (S310). The computing device can estimate the reliability in the same manner as the estimation method described above.

[0063] And if the computing device determines that the estimated reliability is low, it searches for a second sensor system (S320) that includes a second low-cost sensor which exhibits a measurement value similar to that of a first low-cost sensor included in a first sensor system. Among a plurality of sensor systems including low-cost sensors and high-cost sensors, the computing device may search for a second sensor system that includes a second low-cost sensor which exhibits a measurement value where the cosine similarity value with respect to the measurement value of the first low-cost sensor is greater than or equal to a preset threshold.

[0064] Then, the computing device collects the measurement values ​​of the second low-cost sensor and the measurement values ​​of the high-cost sensor included in the second sensor system (S330). The collected measurement values ​​can be used for retraining the virtual sensor model. If multiple second sensor systems are found, the computing device may, as an example, collect the measurement values ​​of the low-cost and high-cost sensors of the second sensor system in which the cosine similarity value is maximum.

[0066] The technical details described above may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiments, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. Hardware devices may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

[0068] As described above, the present invention has been explained by specific details such as specific components, limited embodiments, and drawings; however, this is provided merely to aid in a more comprehensive understanding of the invention, and the invention is not limited to the above embodiments. A person skilled in the art to which the invention pertains can make various modifications and variations from this description. Therefore, the scope of the invention should not be limited to the described embodiments, and all things equivalent to or having equivalent variations to the claims set forth below, as well as the claims themselves, shall be considered to fall within the scope of the concept of the invention.

Claims

Claim 1 A method for collecting data to improve the reliability of a virtual sensor, comprising: a step of estimating the reliability of a pre-trained virtual sensor model included in a first sensor system; a step of calculating a similarity value between a measurement value of a first low-cost sensor included in the first sensor system and a measurement value of a second low-cost sensor included in the second sensor system when the reliability is determined to be low; and a step of collecting the measurement value of the second low-cost sensor and the measurement value of the high-cost sensor included in the second sensor system when the similarity value is greater than or equal to a preset threshold, wherein the virtual sensor model is a model to which weights of a plurality of artificial neural networks are applied, which are trained to receive the measurement value of the first low-cost sensor and output a predicted value for the measurement value of the high-cost sensor included in the first sensor system. Claim 2 delete Claim 3 A data collection method for improving the reliability of a virtual sensor according to claim 1, wherein the step of estimating the reliability of the virtual sensor model comprises: a step of estimating an average value for the predicted values ​​of the artificial neural networks; a step of estimating a variance value for the predicted values ​​of the artificial neural networks using the average value; a step of calculating a confidence interval for the predicted values ​​of the virtual sensor model using the variance value; and a step of estimating the reliability by comparing the confidence interval with a preset threshold interval. Claim 4 In claim 3, the step of estimating the reliability is a data collection method for improving the reliability of a virtual sensor, wherein the reliability is estimated as low when the length of the reliability interval is greater than the length of the critical interval, and as high when the length of the reliability interval is less than or equal to the length of the critical interval. Claim 5 In claim 3, the variance value includes an aleatoric variance value and an epistemic variance value, and the confidence interval is a confidence interval calculated from the epistemic variance value. A data collection method for improving the reliability of a virtual sensor. Claim 6 A method for collecting data to improve the reliability of a virtual sensor, comprising: a step of estimating the reliability of a pre-trained virtual sensor model included in a first sensor system; a step of calculating a similarity value between a measurement value of a first low-cost sensor included in the first sensor system and a measurement value of a second low-cost sensor included in the second sensor system when the reliability is determined to be low; and a step of collecting the measurement value of the second low-cost sensor and the measurement value of the high-cost sensor included in the second sensor system when the similarity value is greater than or equal to a preset threshold value, wherein the step of collecting the measurement value of the high-cost sensor involves collecting the measurement value when the state in which the similarity value calculated periodically is greater than or equal to the threshold value is maintained for a preset threshold time or longer, and stopping the collection when the similarity value becomes smaller than the threshold value while the measurement value is being collected. Claim 7 A method for collecting data to improve the reliability of a virtual sensor comprises: a step of estimating the reliability of a pre-trained virtual sensor model included in a first sensor system; a step of calculating a similarity value between a measurement value of a first low-cost sensor included in the first sensor system and a measurement value of a second low-cost sensor included in the second sensor system when the reliability is determined to be low; and a step of collecting the measurement value of the second low-cost sensor and the measurement value of the high-cost sensor included in the second sensor system when the similarity value is greater than or equal to a preset threshold value, wherein the method further comprises a step of receiving the measurement value of the second low-cost sensor and performing training on the virtual sensor model to predict the measurement value of the high-cost sensor included in the second sensor system. Claim 8 A method for collecting data to improve the reliability of a virtual sensor, comprising: a step of estimating the reliability of a pre-trained virtual sensor model included in a first sensor system; a step of searching for a second sensor system including a second low-cost sensor that exhibits a measurement value similar to that of a first low-cost sensor included in the first sensor system when the reliability is determined to be low; and a step of collecting the measurement value of the second low-cost sensor and the measurement value of a high-cost sensor included in the second sensor system, wherein the step of searching for the second sensor system includes searching for a second sensor system including a second low-cost sensor that exhibits a measurement value where the cosine similarity value with respect to the measurement value of the first low-cost sensor is greater than or equal to a preset threshold. Claim 9 delete Claim 10 A data collection system for improving the reliability of virtual sensors, comprising: a first sensor system including a first low-cost sensor, a first high-cost sensor, and a pre-trained virtual sensor model; a second sensor system including a second low-cost sensor and a second high-cost sensor; and a data collection device for collecting measurements of the second low-cost sensor and the second high-cost sensor, wherein the data collection device estimates the reliability of the virtual sensor model, calculates a similarity value between the measurements of the first low-cost sensor and the measurements of the second low-cost sensor according to the reliability, and collects the measurements of the second low-cost sensor and the second high-cost sensor according to the similarity value.

Images