Method of calculating power saving amount by using deep learning model and device therefor

Abstract

According to a method of calculating a power saving amount by a device comprising a processor and a memory by using a deep learning model according to an embodiment of the present invention, predicted power usage assuming that a power saver is not installed in the current state can be calculated simply by inputting environmental data for a factory where the power saver is installed into a power usage prediction model which has been trained with a training dataset for an arbitrary factory where the power saver is not installed. Therefore, the power saving amount can be accurately calculated without the need to specify a time point for comparison with the current state.

Description

Method for calculating power savings using a deep learning model and device therefor

[0001] The present invention relates to a method and device for calculating power savings using a deep learning model. More specifically, the present invention relates to a method and device for easily and accurately calculating power savings before and after the installation of a power-saving device by predicting power usage prior to the installation of the device in a specific location.

[0002] As the need for carbon neutrality grows increasingly important and ESG (Environmental, Social, Governance) management emerges as a core topic in corporate operations, many businesses are continuously trying to reduce electricity consumption in all locations where they conduct business, such as factories and buildings, and the installation of separate power-saving devices is a representative example of such efforts.

[0003] Meanwhile, various national support measures are being developed to encourage reduced electricity usage, with carbon points being a prime example. Carbon points, also known as carbon neutrality points, are a system aimed at reducing energy consumption, such as electricity, water, and city gas, used by individuals, commercial users, and apartment complexes to address climate change. Points are typically awarded in exchange for reduced carbon emissions. Since carbon emissions are directly related to electricity usage, reducing electricity usage earns carbon points commensurate with the amount of electricity saved.

[0004] This carbon point system must be operated transparently above all else, as carbon points themselves are goods with monetary value. More specifically, the amount of electricity saved, which is converted into carbon points, must be accurately calculated. Failure to accurately calculate the amount of electricity saved will result in more or less carbon points being paid, which may result in unintended preferential treatment or damage.

[0005] However, in the case of installing a power-saving device, since the installation itself is already saving power, it is difficult to specify a point in time to compare with the current state for calculating the amount of power saved. For example, it is possible to specify a point in time (month, day, etc.) in the previous year when the power-saving device was not installed as the point in time to compare. However, since internal / external environments that affect power usage may be different in the current and previous years, there is a problem in that a simple comparison may have a negative effect on the accurate calculation of the amount of power saved.

[0006] Nonetheless, in order to calculate the amount of power savings, it is necessary to specify a point in time that is comparable to the present time with the power saving device installed, and above all, to assume a case where the power saving device is not installed, and the present invention relates to this.

[0007] The technical problem to be solved by the present invention is to provide a method for calculating power savings by utilizing a deep learning model capable of calculating an accurate power savings amount assuming a case where a power savings device is not installed even though a power savings device is installed in a specific location for power savings, and a device therefor.

[0008] Another technical problem that the present invention seeks to solve is to provide a method and device for calculating the amount of power saved by utilizing a deep learning model that can independently evaluate how faithfully power is being saved when installing a power-saving device to save power, and promote more effective power saving based on the evaluation results.

[0009] Another technical challenge to be solved by the present invention is to provide a method for calculating power savings using a deep learning model, which can contribute to establishing a power usage operation plan by visually providing users with power savings and expected carbon points, and a device therefor.

[0010] Another technical problem that the present invention seeks to solve is to provide a method for calculating power savings by utilizing a deep learning model that can ensure reliability in converting the calculated power savings into carbon points, and a device therefor.

[0011] The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0012] In order to achieve the above technical problem, a method for calculating power savings by utilizing a deep learning model in a device including a processor and a memory according to an embodiment of the present invention comprises: (a) a first step of receiving a learning data set for learning a power usage prediction model, (b) a second step of learning the power usage prediction model using the received learning data set, (c) a third step of receiving environmental data affecting power usage from a factory in which a power saver is installed and in which power savings are to be calculated, (d) a fourth step of inputting the received environmental data into the learned power usage prediction model and outputting an estimated power usage, and (e) a fifth step of calculating power savings by utilizing actual power usage of the factory in which the power saver is installed and in which power savings are to be calculated and the output estimated power usage, wherein the learning data set includes at least one learning data corresponding to environmental data for at least one arbitrary factory in which the power saver is not installed and power usage data in this case.

[0013] According to one embodiment, the environmental data may include one or more of external temperature data, internal temperature data, external humidity data, internal humidity data, data on the number of equipment placed in the factory, status data of the equipment, operating time data of the equipment, and power consumption specification data of the equipment for the arbitrary factory for a predetermined period of time.

[0014] According to one embodiment, the environmental data may further include one or more of the production volume of a product produced at the arbitrary factory during the predetermined period of time, characteristics of the product, status data of a lighting system installed at the arbitrary factory, operating time data, and power consumption specification data of the lighting system.

[0015] According to one embodiment, the environmental data received in the third step is:

[0016] The above power saving device may be installed, and the environmental data may be collected for a certain period of time for a factory that wants to calculate the amount of power savings.

[0017] According to one embodiment, the second step may be to set environmental data as an independent variable for each learning data included in the learning data set, and power usage data in this case as a dependent variable, and learn the relationship between the independent variable and the dependent variable for all learning data, while learning in a way that minimizes the difference between the predicted power usage output by the power usage prediction model using the environmental data included in each learning data and the power usage data included in each learning data.

[0018] According to one embodiment, after the fifth step, a sixth step of retraining the power usage prediction model may be further included, wherein the sixth step may include a sixth step of calculating MSE of input and output to the power usage prediction model with respect to environmental data received from a factory in which the power saving device is installed and the amount of power savings is to be calculated, and a sixth step of retraining the power usage prediction model in a direction of maximizing the calculated MSE with respect to environmental data received from a factory in which the power saving device is installed and the amount of power savings is to be calculated.

[0019] According to one embodiment, after the sixth step, the seventh step may further include evaluating the power saving fidelity of a factory in which the power saving device is installed and the power saving amount is to be calculated, wherein the eighth step may include a seventh step of inputting another environmental data received after the time point of receiving the environmental data from the factory in which the power saving device is installed and the power saving amount is to be calculated into the re-learned power usage prediction model to output an estimated power usage, and a seventh step of evaluating the power saving fidelity by comparing the estimated power usage output in the seventh step with an actual power usage corresponding to another environmental data received after the time point of receiving the environmental data from the factory in which the power saving device is installed and the power saving amount is to be calculated.

[0020] According to one embodiment, after the fifth step, an eighth step of converting the calculated power savings into carbon points by applying blockchain technology may be further included.

[0021] According to one embodiment, after the fifth step, a ninth step of visualizing the calculated power savings and converted carbon points and outputting them to a user terminal may be further included.

[0022] In order to achieve the above technical problem, an apparatus for calculating power savings by utilizing a deep learning model according to an embodiment of the present invention comprises one or more processors, a network interface, a memory for loading a computer program to be executed by the processor, and a storage for storing large-capacity network data and the computer program, wherein the computer program comprises: (A) a first step of receiving a learning data set for learning a power usage prediction model, (B) a second step of learning the power usage prediction model using the received learning data set, (C) a third step of receiving environmental data affecting power usage from a factory in which a power saver is installed and in which power savings are to be calculated, (D) a fourth step of inputting the received environmental data into the learned power usage prediction model and outputting the predicted power usage, and (E) a fifth step of calculating power savings by utilizing the actual power usage of the factory in which the power saver is installed and in which power savings are to be calculated, and the output predicted power usage, wherein the learning data set comprises environmental data for one or more arbitrary factories in which the power saver is not installed, and this environmental data. The power usage data of the case includes one or more corresponding learning data.

[0023] According to an embodiment of the present invention for achieving the above technical task, a computer program stored in a computer-readable medium includes: (AA) a first step of receiving a learning data set for learning a power usage prediction model; (BB) a second step of learning the power usage prediction model using the received learning data set; (CC) a third step of receiving environmental data affecting power usage from a factory in which a power saving device is installed and in which power savings are to be calculated; (DD) a fourth step of inputting the received environmental data into the learned power usage prediction model and outputting the predicted power usage; and (EE) a fifth step of calculating power savings using the actual power usage of the factory in which the power saving device is installed and in which power savings are to be calculated, and the output predicted power usage, wherein the learning data set includes at least one learning data corresponding to environmental data for at least one arbitrary factory in which the power saving device is not installed and power usage data in this case.

[0024] According to the present invention as described above, by simply inputting environmental data for a factory in which a power-saving device is installed into a power usage prediction model trained with a learning data set for an arbitrary factory in which a power-saving device is not installed, the predicted power usage can be calculated assuming that a power-saving device is not installed in the current state, thereby enabling accurate power-saving calculation without having to specify a point in time to be compared with the current state.

[0025] In addition, when receiving environmental data for a factory with a power-saving device installed, the power usage prediction model learned with the learning data set for an arbitrary factory without a power-saving device installed calculates the expected power usage for the factory and continuously retrains it to perform fine tuning to ensure the reliability of the expected power usage calculation for the factory with a power-saving device installed, and compares it with the actual power usage to evaluate how faithfully power is being saved. This has the effect of contributing to going beyond the passive attitude of checking the numerical power-saving amount and thinking about and implementing more active power-saving measures.

[0026] In addition, it has the effect of providing objectivity as a basis for subsequent control actions by users for power saving by evaluating the power saving fidelity by comparing it with the recommended expected power usage in the current state in which the power saving device is installed.

[0027] Additionally, blockchain technology, specialized for security, is applied to convert the generated power savings into carbon points, which has the effect of providing reliability in the process of issuing carbon points, which are a means of material value.

[0028] Additionally, the calculated power savings and converted carbon points are visualized and output to the user terminal, which improves user convenience and allows the user to utilize the data in establishing their own power usage plan.

[0029] The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

[0030] FIG. 1 is a drawing showing the overall configuration of a device for calculating power savings using a deep learning model according to a first embodiment of the present invention.

[0031] FIG. 2 is a diagram illustrating the overall system structure that provides a method for calculating power savings by utilizing a deep learning model according to a second embodiment of the present invention.

[0032] FIG. 3 is a flowchart showing representative steps of a method for calculating power savings using a deep learning model according to a second embodiment of the present invention.

[0033] FIG. 4 is a diagram exemplarily illustrating the structure of a power usage prediction model in a method for calculating power savings using a deep learning model according to a second embodiment of the present invention.

[0034] FIG. 5 is a flowchart showing representative steps of a method for calculating power savings using a deep learning model according to a third embodiment of the present invention.

[0035] FIG. 6 is a flowchart detailing the sixth step of retraining a power usage prediction model in a method for calculating power savings using a deep learning model according to a third embodiment of the present invention.

[0036] FIG. 7 is a flowchart detailing the seventh step of evaluating power saving fidelity in a method for calculating power saving amount using a deep learning model according to a third embodiment of the present invention.

[0037] FIG. 8 is a flowchart showing representative steps of a method for calculating power savings using a deep learning model according to a fourth embodiment of the present invention.

[0038] Figure 9 is an exemplary diagram illustrating how power savings and carbon points are visualized and output on a user terminal.

[0039] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. The advantages and features of the present invention, and methods for achieving them, will become clear with reference to the embodiments described in detail below together with the attached drawings. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various different forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

[0040] Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in their common sense to those of ordinary skill in the art to which the present invention pertains. Furthermore, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0041] Meanwhile, the terms used in this specification are intended to describe embodiments and are not intended to limit the present invention. In this specification, singular forms also include plural forms, unless specifically stated otherwise.

[0042] As used herein, the terms “comprises” and / or “comprising” do not exclude the presence or addition of one or more other components, steps, operations and / or elements.

[0043] FIG. 1 is a drawing showing the overall configuration of a device (100) that calculates power savings by utilizing a deep learning model according to a first embodiment of the present invention.

[0044] However, this is only a preferred embodiment for achieving the purpose of the present invention, and some components may be added or deleted as needed, and the role performed by one component may be performed by another component as well.

[0045] A device (100) for calculating power savings by utilizing a deep learning model according to a first embodiment of the present invention may include a processor (10), a network interface (20), a memory (30), a storage (40), and a data bus (50) connecting them, and of course, may further include additional components required to achieve the purpose of the present invention.

[0046] The processor (10) controls the overall operation of each component. The processor (10) may be any of a CPU (Central Processing Unit), an MPU (Micro Processor Unit), an MCU (Micro Controller Unit), or a processor of a type widely known in the technical field to which the present invention pertains, and may be implemented as an artificial intelligence model processor, such as a machine learning model processor or a deep learning model processor. In addition, the processor (10) may perform operations for at least one application or program for performing a method for calculating power savings by utilizing a deep learning model according to the second embodiment of the present invention.

[0047] The network interface (20) supports wired and wireless Internet communication of the device (100) that calculates power savings using the deep learning model according to the first embodiment of the present invention, and may also support other known communication methods. Accordingly, the network interface (20) may be configured to include a corresponding communication module.

[0048] The memory (30) stores various types of information, commands, and / or information, and can load one or more computer programs (41) from the storage (40) to perform a method of calculating power savings by utilizing a deep learning model according to the second embodiment of the present invention. In Fig. 1, RAM is illustrated as one of the memories (30), but it goes without saying that various storage media can be used as the memory (30).

[0049] Storage (40) can non-temporarily store one or more computer programs (41) and large-capacity network information (42). This storage (40) can be any one of non-volatile memory such as Read Only Memory (ROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), flash memory, a hard disk, a removable disk, or any type of computer-readable recording medium widely known in the art to which the present invention pertains.

[0050] A computer program (41) is loaded into a memory (30), and one or more processors (10) can execute (A) a first step of receiving a learning data set for learning a power usage prediction model, (B) a second step of learning the power usage prediction model using the received learning data set, (C) a third step of receiving environmental data that affects power usage from a factory in which a power saving device is installed and in which power saving is to be calculated, (D) a fourth step of inputting the received environmental data into the learned power usage prediction model and outputting the predicted power usage, and (E) a fifth step of calculating the power saving amount using the actual power usage of the factory in which the power saving device is installed and in which power saving is to be calculated and the output predicted power usage.

[0051] The operations performed by the computer program (41) briefly mentioned above can be viewed as a function of the computer program (41), and a more detailed description will be provided later in the description of a method for calculating power savings by utilizing a deep learning model according to the second embodiment of the present invention.

[0052] The data bus (50) serves as a path for transferring commands and / or information between the processor (10), network interface (20), memory (30), and storage (40) described above.

[0053] The device (100) for calculating power savings by utilizing the deep learning model according to the first embodiment of the present invention, which has been briefly described above, may be in the form of an independent device, for example, an electronic device or a server (including a cloud), and the electronic device may be not only a desktop PC or server device that is fixedly installed and used in one place, but also a portable device that is easy to carry, such as a smart phone, tablet PC, notebook PC, PDA, PMP, etc. Any electronic device that has a CPU corresponding to a processor (10) installed therein and only a network function may be used.

[0054] Hereinafter, assuming that the device (100) for calculating power savings by utilizing a deep learning model according to a first embodiment of the present invention is in the form of a "server" among the electronic devices that are independent devices, a process for providing a method for calculating power savings by utilizing a deep learning model according to a second embodiment of the present invention through a dedicated application installed on a user terminal (300) of a user who wishes to receive the power savings calculation result described below will be described with reference to FIGS. 2 to 9.

[0055] FIG. 2 is a diagram illustrating the overall system structure that provides a method for calculating power savings by utilizing a deep learning model according to a second embodiment of the present invention.

[0056] Referring to FIG. 2, the entire system for servicing the method for calculating power savings by utilizing a deep learning model according to the second embodiment of the present invention may include a device (100) for calculating power savings by utilizing a deep learning model according to the first embodiment of the present invention, which corresponds to a server, one or more factory energy management systems (FEMS, Factory Energy Management Systems, each of which includes one or more power savers, one or more energy information measuring / collecting / control devices, etc., 200), a user terminal (300) that may include all electronic devices having network and display functions, such as smartphones, smart watches, smart glasses, desktop computers, laptop computers, tablet computers, PDAs, PDPs, PMPs, etc., and in some cases, a government agency server (server of a government agency / local government, etc. operating a carbon point system, 400), and the method for calculating power savings by utilizing a deep learning model according to the second embodiment of the present invention may be provided by the device (100) for calculating power savings by utilizing a deep learning model according to the first embodiment of the present invention as the subject of the service provision. It is about the service provided to the user based on the system shown in 2.

[0057] FIG. 3 is a flowchart showing representative steps of a method for calculating power savings using a deep learning model according to a second embodiment of the present invention.

[0058] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0059] In addition, each step will be explained on the premise that the device (100) that calculates the amount of power savings by utilizing the deep learning model according to the first embodiment of the present invention is implemented in the form of a server.

[0060] In addition, the method of calculating power savings using a deep learning model according to the second embodiment of the present invention can be applied to any place where power is used, but the following description assumes a "factory", and it should be understood that the word "factory" used below can be replaced with words such as "building" or "specific place."

[0061] First, a learning data set (Set) for learning a power usage prediction model is received (S310), which is called the first step.

[0062] Here, the power usage prediction model is an artificial intelligence deep learning model based on LSTM (Long Short-Term Memory) installed in the device (100), more specifically, in the processor (10) included in the device (100). Details will be described later, and the first step can be viewed as a step of collecting learning data for learning the power usage prediction model, which is an artificial intelligence deep learning model.

[0063] Here, the training data set may be a data set including one or more training data corresponding to power usage data in a factory, more specifically, environmental data for one or more random factories where power saving devices are not installed, and power usage data in this case.

[0064] In this case, environmental data for one or more random factories where power saving devices are not installed, and power usage data in this case, are learning data, so the more the better. However, due to the nature of learning data that is difficult to collect itself, it is possible to prepare a learning data set by augmenting the collected learning data using a known data augmentation technique.

[0065] Meanwhile, since the data is about an arbitrary factory, there are no restrictions on the type of factory, such as an agricultural / fishery product processing factory, a clothing and textile factory, an automobile manufacturing factory, a semiconductor manufacturing factory, etc. However, due to the characteristics of the factory location, the deviation in electricity usage depending on the type of goods handled or produced may be significantly different, so the data about an arbitrary factory can be limited to data about factories that handle or produce the same goods, in which case it will be possible to implement a more reliable and specialized electricity usage prediction model.

[0066] For example, if the data for any factory is limited to data for clothing and textile factories, the power usage prediction model may be specialized for clothing and textile factories because it was trained only with the data set for clothing and textile factories. In this case, the use of the power usage prediction model may not be appropriate for factories that handle or produce other goods, not just clothing.

[0067] Conversely, if we do not limit the data on factories that handle or produce electricity, it could be a more universally applicable and general-purpose power consumption prediction model. However, the actual prediction results may be somewhat less reliable, as data on factories that handle or produce different goods from those handled or produced by the factories whose electricity consumption we are trying to predict are also used for learning.

[0068] As such, the reliability and versatility of the power usage prediction model to be built can vary depending on the type of factory the training dataset is from. Therefore, it's best to adopt an appropriate approach based on the business's circumstances. For example, businesses in the early stages will likely struggle with operating costs, so they might choose a general-purpose training dataset that applies to all factories, rather than limiting it to factories handling or producing specific products. Other businesses, however, with ample operating funds and aiming to enhance their services, might choose a training dataset that focuses on specific products and factories handling or producing specific products, thereby ensuring reliability and providing accurate prediction results for specific customer segments.

[0069] Meanwhile, the learning data set includes not only power usage data for a specific factory but also environmental data for the factory. Here, environmental data means all data that can directly or indirectly affect the power usage data, and for example, it may include one or more of external temperature data, internal temperature data, external humidity data, internal humidity data, data on the number of equipment deployed in the factory, status data of the equipment, operating time data of the equipment, and power consumption specification data of the equipment for the arbitrary factory for a given period of time, and it goes without saying that it may further include other data.

[0070] Here, environmental data including one or more of external temperature data, internal temperature data, external humidity data, internal humidity data, data on the number of equipment deployed in the factory, equipment status data, equipment operating time data, and equipment power consumption specification data for any factory for a given period of time is called first environmental data, and is relatively universal data that is expressed in objective numbers with little deviation depending on the type of product being handled or produced, and if a power usage prediction model is trained using a learning data set consisting only of such first environmental data and power usage data for it, a prediction model that has selected the versatility described above can be implemented.

[0071] Furthermore, environmental data may further include one or more of the production volume of products produced in any factory for a given period of time, the characteristics of the products, the status data of the lighting system installed in any factory, the operating time data, and the power consumption specification data of the lighting system. These environmental data are called secondary environmental data to distinguish them from the primary environmental data, and are data that may have a large deviation depending on the type of product handled or produced and may differ from factory to factory. If a power usage prediction model is trained using a learning data set consisting only of such secondary environmental data and the corresponding power usage data, a prediction model that has selected the reliability described above can be implemented.

[0072] Meanwhile, if the learning data set is composed of first environment data and its corresponding power usage data, as well as second environment data and its corresponding power usage data, then either a prediction model biased toward generality or a prediction model biased toward reliability can be implemented depending on the amount or ratio of each of the first environment data and the second environment data.

[0073] If a learning data set is received, the device (100) trains a power usage prediction model using the received learning data set (S320), which is called the second step.

[0074] Before explaining the second step, let's explain the power usage prediction model that was previously withheld.

[0075] FIG. 4 is a diagram exemplarily illustrating the structure of a power usage prediction model in a method for calculating power savings using a deep learning model according to a second embodiment of the present invention.

[0076] The power usage prediction model can be implemented with an LSTM network structure. LSTM is a type of recurrent neural network (RNN) designed to overcome the limitations of existing RNNs in capturing and learning long-term dependencies in sequential data, and is a network structure that can model and learn the relationship between independent variables, which are input data, and dependent variables, which are output data.

[0077] In a method for calculating power savings by utilizing a deep learning model according to a second embodiment of the present invention, an independent variable may be first environmental data including at least one of external temperature data, internal temperature data, external humidity data, internal humidity data, data on the number of equipment deployed in the factory, status data of the equipment, operating time data of the equipment, and power consumption specification data of the equipment for an arbitrary factory for a predetermined period of time, second environmental data including at least one of the production volume of products produced in the arbitrary factory for a predetermined period of time, characteristics of the products, status data of a lighting system installed in the arbitrary factory, operating time data, and power consumption specification data of the lighting system, and any one of all of the first environmental data and the second environmental data, and a dependent variable may be an expected power usage amount influenced by at least one of the first environmental data and the second environmental data.

[0078] In this way, when a power usage prediction model is implemented using an LSTM network, if only the environmental data (input data) corresponding to the independent variable is input into the power usage prediction model, the predicted power usage (output data) corresponding to the dependent variable can be automatically calculated and output. However, there will inevitably be a difference between the input data and the calculated output data during the learning process. This is called the Reconstruction Error, and the loss function of the power usage prediction model is trained to minimize the Mean Squared Error (MSE, the average of the squares of the differences between the input data and the output data), that is, to minimize the reconstruction error. This is to ensure the reliability of the power usage prediction model.

[0079] In summary, the power usage prediction model implemented with the LSTM network structure sets environmental data as an independent variable for each learning data included in the learning data set, and power usage data in this case as a dependent variable, and learns the relationship between the independent and dependent variables for all learning data, but proceeds with learning in a way that minimizes the difference between the predicted power usage output by the power usage prediction model using the environmental data included in each learning data and the power usage data included in each learning data.

[0080] Meanwhile, the power usage prediction model implemented with the LSTM network structure is only one example of a prediction model implementation, and it is entirely possible to implement a prediction model with a different structure.

[0081] If the first and second steps described above have been performed, the learning of the power usage prediction model is primarily completed, and we return to the description of Fig. 3.

[0082] If the power usage prediction model has been trained, the device (100) receives environmental data affecting power usage from a factory that has a power saving device installed and wants to calculate the amount of power saved (S330), which is referred to as the third step.

[0083] The third step corresponds to the step where the method of calculating power savings by utilizing the deep learning model according to the second embodiment of the present invention is actually provided to the user. The environmental data received in the third step is environmental data received from a user (factory) who wants to calculate power savings based on the current standard, assuming that a power saving device is installed but the power saving device is not installed.

[0084] Meanwhile, the environmental data here may be at least one of the first environmental data and the second environmental data, but it is desirable to receive the corresponding environmental data depending on which environmental data the power usage prediction model used for learning, in order to ensure maximum reliability in calculating the expected power usage.

[0085] In addition, the environmental data received in the third stage may be received in real time from the factory, but it may be environmental data collected over a certain period of time rather than in real time. This is because there are cases where it is preferable to treat environmental data as time-series data and collect it over a certain period of time and then calculate it, rather than calculating the expected power usage in real time by receiving environmental data according to independent variables.

[0086] For example, in the case of internal temperature data, the internal temperature can change at any time during the day while the factory is operating, such as rising or falling. If such data is received in real time and the expected power usage is calculated in real time, the expected power usage will also change in real time whenever the internal temperature rises or falls, which may cause confusion to the user. However, if the internal temperature data is collected and received on a daily basis, a single expected power usage can be calculated by reflecting all cases where the temperature rises or falls within the day. Therefore, not only can the load on the computational aspect of the device (100) be reduced, but also convenience can be increased from the user's perspective because the expected power usage can be confirmed as a single number.

[0087] However, receiving environmental data as environmental data collected over a predetermined period of time is only a preferred embodiment, and receiving it in real time, and further calculating the expected power usage in real time, is not excluded, and can be freely selected according to the settings of the operator of the device (100).

[0088] Meanwhile, the device (100) can receive environmental data from the factory and also receive power usage data for the factory.

[0089] Here, the power usage data may be real-time power usage data when environmental data is received in real time, or may be power usage data for a predetermined period when environmental data collected over a predetermined period is received, and is for calculating the power savings amount described later.

[0090] If environmental data is received, the device (100) inputs the received environmental data into a power usage prediction model that has completed learning to output the expected power usage (S340), which is called the fourth step, and the power saving device is installed, and the actual power usage of the factory where the power saving amount is to be calculated and the output expected power usage amount are used to calculate the power saving amount (S350), which is called the fifth step.

[0091] The expected power usage output in the fourth step is the expected power usage calculated by assuming that the power saver is not installed in the factory where the power saver is installed. This is because the power usage prediction model was trained using a learning data set for factories where the power saver is not installed, so even if the power saver is currently installed, it can calculate the expected power usage in the case where the power saver is not installed using only environmental data.

[0092] If the expected power usage is calculated, the device (100) can calculate the power saving amount by using the actual power usage of the factory received in the fourth step and the calculated expected power usage, more specifically by performing a minus operation on the two usages.

[0093] As a result of calculating the power savings, if power savings are being achieved, the user can maintain the current state or perform additional control actions to achieve more efficient power savings. If power savings are not being achieved, the user can perform more aggressive control actions.

[0094] So far, a method for calculating power savings using a deep learning model according to the second embodiment of the present invention has been described. According to the present invention, by simply inputting environmental data about a factory where a power saver is installed into a power usage prediction model trained using a training data set for an arbitrary factory where a power saver is not installed, the predicted power usage can be calculated assuming that a power saver is not installed in the current state. Therefore, accurate power savings can be calculated without having to specify a point in time to be compared with the current state.

[0095] Meanwhile, as briefly explained above, by checking the power savings calculation results, the user can maintain the current status or perform more active control actions. However, since the user's subjective judgment as to whether or not appropriate power savings are being achieved by numerically checking the power savings may lack objectivity as a basis for subsequent control actions. Below, a method for calculating power savings using a deep learning model according to a third embodiment of the present invention that complements this will be described.

[0096] FIG. 5 is a flowchart showing representative steps of a method for calculating power savings using a deep learning model according to a third embodiment of the present invention.

[0097] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0098] In addition, each step will be explained on the premise that the device (100) that calculates the amount of power savings by utilizing the deep learning model according to the first embodiment of the present invention is implemented in the form of a server.

[0099] The description of steps 1 to 5 is identical to the description of the method for calculating power savings using a deep learning model according to the second embodiment of the present invention, so detailed descriptions will be omitted to avoid redundant descriptions, and only the differences will be described.

[0100] If the device (100) has calculated the amount of power savings, the power usage prediction model is retrained (S360), which is called the 6th step. After the 6th step, the power saving fidelity of the factory where the power saving device is installed and the power saving amount is to be calculated is evaluated (S370), which is called the 7th step.

[0101] As explained above, the power usage prediction model is trained with a learning data set for an arbitrary factory without a power saving device installed, so even if it receives environmental data for any factory as input data, it tends to restore the predicted power usage for a factory without a power saving device installed as much as possible. However, from the third stage onwards, environmental data for factories with power saving devices installed will be continuously received, so fine-tuning of the power usage prediction model is required, and this corresponds to the retraining discussed in the sixth stage.

[0102] In addition, as mentioned above, the subjective judgment of whether or not appropriate power saving is being achieved by numerically confirming the amount of power saving may lack objectivity as a basis for the user's subsequent control actions. The power saving sincerity discussed in step 7 can supplement this, and will be described in detail below with reference to FIGS. 6 and 7.

[0103] FIG. 6 is a flowchart detailing a sixth step of retraining a power usage prediction model in a method for calculating power savings using a deep learning model according to a third embodiment of the present invention, and FIG. 7 is a flowchart detailing a seventh step of evaluating power savings fidelity.

[0104] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0105] First, referring to FIG. 6, the device (100) calculates the MSE of input and output to the power usage prediction model for environmental data received from a factory where a power saving device is installed and the power saving amount is to be calculated (S360-1), which is referred to as step 6-1.

[0106] In the explanation of the power usage prediction model above, it is inevitable that there will be a difference between the input data and the output data produced during the learning process, which is called the reconstruction error, and the loss function of the power usage prediction model is said to proceed with learning in the direction of minimizing the MSE (Mean Squared Error, the average of the squares of the differences between the input data and the output data), that is, minimizing the reconstruction error. This was to calculate the predicted power usage assuming that no power saving device is installed.

[0107] In step 6-1, MSE is calculated in the same way, but the difference is that the input data is not for a factory without a power-saving device installed, but rather environmental data received from a factory with a power-saving device installed but for which the power-saving amount is to be calculated.

[0108] If the MSE is calculated, the device (100) retrains the power usage prediction model in a direction that maximizes the calculated MSE for environmental data received from a factory where a power saving device is installed and the power saving amount is to be calculated (S360-2), and this is referred to as step 6-2.

[0109] As mentioned earlier, the power usage prediction model is trained only with a learning data set for an arbitrary factory where power saving devices are not installed, a kind of voice learning data set, so it tends to restore as much as possible to a voice pattern even when receiving any environmental data. A large calculated MSE means that the difference between the input data and the output data is large, which means that restoration to a voice pattern is not done properly. In other words, it means that the predicted power usage, which is the output data reconstructed from the input data, the environmental data, may be suitable for a factory where power saving devices are installed rather than a factory where they are not installed.

[0110] To put it another way, the power usage prediction model is suitable for calculating the expected power usage in a factory where a power saver is installed assuming that the power saver is not installed, but it is not suitable for calculating the expected power usage in a factory where a power saver is installed when environmental data about the factory itself is input (i.e., if the environmental data is at this level, the factory where the power saver is installed will use at this level of electricity). This problem can be solved through retraining, and while most learning has the problem of being difficult to collect learning data to utilize for learning, the present invention has the advantage that retraining data is automatically collected simply by continuously using the service.

[0111] Meanwhile, the retraining of the power usage prediction model according to steps 6-1 and 6-2 described above may be performed each time environmental data is received in step 3 and the predicted power usage is calculated in step 4, but retraining may be performed all at once after a certain level of environmental data received in step 3 has been collected, or retraining may be performed at regular intervals.

[0112] Now, referring to FIG. 7 for evaluating the power saving fidelity, the device (100) has a power saving device installed, and inputs another environmental data received after the point in time when the environmental data is received from a factory from which the power saving amount is to be calculated into a relearned power usage prediction model to output the predicted power usage (S370-1), which is referred to as step 7-1.

[0113] In evaluating the sincerity of power saving, it is a prerequisite to first receive environmental data from the factory where the power saving device is installed, and then receive other environmental data again to calculate the expected power usage using the power usage prediction model and relearn it. This is because only when there is a history of at least two or more expected power usage calculations and relearning, such as receiving environmental data about the factory where the power saving device is installed and first calculating the expected power usage based on it, and then receiving environmental data about the factory again and calculating the expected power usage based on it, can it become a standard or target for evaluating whether power saving is being properly carried out.

[0114] In this case, although it is said to be at least twice, it is desirable to receive environmental data from the factory where the power saving device is installed as many times as possible to calculate the expected power consumption and retrain the power consumption prediction model before evaluating the power consumption, because only then can the power consumption prediction model be fine-tuned for the factory where the power saving device is installed.

[0115] After step 7-1, the device (100) evaluates the power saving fidelity by comparing the expected power usage output in step 7-1 with the actual power usage corresponding to another environmental data received after the point in time when the environmental data was received from a factory where a power saving device is installed and from which the power saving amount is to be calculated (S370-2), which is referred to as step 7-2.

[0116] The expected power consumption calculated in Step 7-1 is the expected power consumption that a factory with power-saving devices installed would need to consume if it had this level of environmental data while the power consumption prediction model was fine-tuned. Comparing this to the actual power consumption allows for an evaluation of how faithfully the factory can comply with power-saving measures.

[0117] Here, the evaluation can be done by simply comparing the expected power usage and the actual power usage. If the expected power usage is greater than the actual power usage, it can be evaluated as “good,” if it is the same, it can be evaluated as “average,” and if it is smaller, it can be evaluated as “poor.” Each evaluation result can be given a numerical range to provide detailed evaluation content.

[0118] So far, a method for calculating power savings using a deep learning model according to the third embodiment of the present invention has been described. According to the present invention, when a power usage prediction model trained with a learning data set for an arbitrary factory without installed power saving devices receives environmental data for a factory with installed power saving devices, the model calculates the expected power usage for this and continuously retrains the model to perform fine tuning to ensure the reliability of the expected power usage calculation for the factory with installed power saving devices, and compares this with the actual power usage to evaluate how faithfully power is being saved. This can contribute to moving beyond the passive attitude of checking numerical power savings to considering and implementing more proactive power saving measures. In addition, since the faithfulness of power savings is evaluated by comparing it with the recommended expected power usage in the current state where the power saving devices are installed, it can provide an objectivity as a basis for the user's subsequent control actions for power savings.

[0119] This time, we will explain a method for calculating power savings by utilizing a deep learning model according to the fourth embodiment of the present invention related to carbon point conversion.

[0120] FIG. 8 is a flowchart showing representative steps of a method for calculating power savings using a deep learning model according to a fourth embodiment of the present invention.

[0121] However, this is only a preferred embodiment for achieving the purpose of the present invention, and it is obvious that some steps may be added or deleted as needed, and one step may be included in another step and performed.

[0122] In addition, each step is assumed to be performed by a device (100) that calculates power savings by utilizing a deep learning model according to the first embodiment of the present invention, and the implementation form is assumed to be a server, and the explanation will continue on the assumption that the user terminal (300) is a “smartphone.”

[0123] The description of steps 1 to 5 is identical to the description of the method for calculating power savings using a deep learning model according to the second embodiment of the present invention, so detailed descriptions will be omitted to avoid redundant descriptions, and only the differences will be described.

[0124] If the power saving amount is calculated in step 5, the device (100) can convert the calculated power saving amount into carbon points by applying blockchain technology (S380), which is called step 8.

[0125] Since carbon points are a means of payment with monetary value, reliability must be ensured during the payment process to users. To this end, the present invention seeks to apply blockchain technology, which focuses on security.

[0126] The specific conversion process will be as follows: the device (100) notifies the government agency server (400) of the amount of power savings for the user, the government agency server (400) confirms this and approves the conversion of carbon points, and the device (100) or a separate carbon point operation server (not shown) converts and grants carbon points equivalent to the amount of power savings to the user.

[0127] Furthermore, the device (100) can visualize the calculated power savings and converted carbon points and output them to the user terminal (300) (S390), which is referred to as step 9 and is exemplarily attached to FIG. 9.

[0128] Here, visualization can be done in a variety of ways, tailored to the desired output, in addition to the amount of power saved and carbon points, thereby improving user convenience and enabling users to plan their own power usage.

[0129] Finally, the device (100) for calculating power savings using a deep learning model according to the first embodiment of the present invention and the method for calculating power savings using a deep learning model according to the second to fourth embodiments of the present invention can be implemented as a computer program stored on a medium according to the fifth embodiment of the present invention, which includes all the same technical features, in this case, in combination with a computer device, the first step of receiving (AA) a learning data set (Set) for learning a power usage prediction model, (BB) a second step of learning the power usage prediction model using the received learning data set, (CC) a third step of receiving environmental data affecting power usage from a factory in which a power saving device is installed and in which power savings are to be calculated, (DD) a fourth step of inputting the received environmental data into the learned power usage prediction model and outputting the predicted power usage, and (EE) a fifth step of calculating power savings using the actual power usage of the factory in which the power saving device is installed and in which power savings are to be calculated and the output predicted power usage, will be executed, and will not be described in detail to avoid redundant description. It goes without saying that all of the technical features applied to the device (100) for calculating power savings using a deep learning model according to the first embodiment of the present invention described above and the method for calculating power savings using a deep learning model according to the second to fourth embodiments of the present invention can be equally applied to a computer program stored in a medium according to the fifth embodiment of the present invention.

[0130] Although embodiments of the present invention have been described with reference to the attached drawings, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without altering the technical concept or essential features thereof. Therefore, the embodiments described above should be understood to be illustrative in all respects and not restrictive.

Claims

1. A method for calculating power savings of a device including a processor and memory, (a) A first step of receiving a learning data set for learning a power usage prediction model; (b) a second step of training the power usage prediction model using the received learning data set; (c) a third step of receiving environmental data affecting power usage from a factory where a power saving device is installed and the amount of power saving is to be calculated; and (d) a fourth step of inputting the received environmental data into the power usage prediction model for which learning has been completed and outputting the predicted power usage; and (e) a fifth step of calculating the amount of power saved by using the actual power usage of the factory where the power saving device is installed and the output expected power usage for calculating the amount of power saved; Including, The above training data set is, One or more learning data corresponding to environmental data for one or more random factories where the power saving device is not installed and power usage data in this case, A method to calculate power savings using a deep learning model.

2. In paragraph 1, The above environmental data is, Including at least one of external temperature data, internal temperature data, external humidity data, internal humidity data, data on the number of equipment placed in the factory, status data of the equipment, operating time data of the equipment, and power consumption specification data of the equipment for the arbitrary factory for a given period of time. A method to calculate power savings using a deep learning model.

3. In paragraph 2, The above environmental data is, It further includes at least one of the production volume of the product produced at the arbitrary factory during the above-mentioned predetermined period of time, the characteristics of the product, the status data of the lighting system installed at the arbitrary factory, the operating time data, and the power consumption specification data of the lighting system. A method to calculate power savings using a deep learning model.

4. In paragraph 1, The environmental data received in the third step above is: Environmental data collected over a specified period of time for a factory where the above power saving device is installed and the amount of power saving is to be calculated. A method to calculate power savings using a deep learning model.

5. In paragraph 1, The second step above is, For each learning data included in the above learning data set, environmental data is set as an independent variable, and in this case, power usage data is set as a dependent variable, and the relationship between the independent and dependent variables is learned for all learning data. The power usage prediction model is trained in a way that minimizes the difference between the predicted power usage output by using the environmental data included in each learning data and the power usage data included in each learning data. A method to calculate power savings using a deep learning model.

6. In paragraph 1, After the above 5th step, Step 6 of retraining the above power usage prediction model; Including more than, The sixth step above is, Step 6-1 of calculating the MSE of input and output to the power usage prediction model for environmental data received from a factory where the power saving device is installed and the power saving amount is to be calculated; and Step 6-2 of retraining the power usage prediction model in a direction that maximizes the calculated MSE for environmental data received from a factory where the power saving device is installed and from which the amount of power saving is to be calculated; A method for calculating power savings by utilizing a deep learning model including:

7. In paragraph 6, After the above step 6, Step 7: Evaluating the power saving efficiency of a factory where the above power saving device is installed and the amount of power saving is to be calculated; Including more than, The above 7th step is, Step 7-1 of inputting another environmental data received after the time point of receiving the environmental data from the factory where the power saving device is installed and from which the power saving amount is to be calculated into the re-learned power usage prediction model to output the predicted power usage; and Step 7-2 for evaluating the power saving fidelity by comparing the expected power usage output in the above step 7-1 with the actual power usage corresponding to another environmental data received after the time point of receiving the environmental data from the factory where the power saving device is installed but the power saving amount is to be calculated; A method for calculating power savings by utilizing a deep learning model including:

8. In paragraph 1, After the above 5th step, Step 8: Converting the calculated power savings into carbon points by applying blockchain technology; A method for calculating power savings by utilizing a deep learning model including more.

9. In paragraph 8, After the above 5th step, Step 9 of visualizing the calculated power savings and converted carbon points and outputting them to a user terminal; A method for calculating power savings by utilizing a deep learning model including more.

10. One or more processors; network interface; A memory that loads a computer program to be executed by the processor; and Including storage for storing large amounts of network data and the computer program, The above computer program is executed by one or more processors, (A) A first step of receiving a learning data set for learning a power usage prediction model; (B) a second step of training the power usage prediction model using the received learning data set; (C) a third step of receiving environmental data affecting power usage from a factory where a power saving device is installed and the amount of power saving is to be calculated; and (D) a fourth step of inputting the received environmental data into the power usage prediction model for which learning has been completed and outputting the predicted power usage; and (E) A fifth step of calculating power savings by using the actual power usage of the factory where the power saving device is installed and the output expected power usage for calculating the power savings; Including, The above training data set is, One or more learning data corresponding to environmental data for one or more random factories where the power saving device is not installed and power usage data in this case, A device that calculates power savings using a deep learning model.

11. In combination with a computing device, (AA) A first step of receiving a training data set for training a power usage prediction model; (BB) A second step of training the power usage prediction model using the received learning data set; (CC) A third step of receiving environmental data affecting power usage from a factory where a power saving device is installed and the amount of power saving is to be calculated; and (DD) A fourth step of inputting the received environmental data into the power usage prediction model for which learning has been completed and outputting the predicted power usage; and (EE) A fifth step of calculating power savings by using the actual power usage of the factory where the power saving device is installed and the output expected power usage for calculating the power savings; Including, The above training data set is, One or more learning data corresponding to environmental data for one or more random factories where the power saving device is not installed and power usage data in this case, A computer program stored on a computer-readable medium.

Images