Artificial intelligence device and operation method thereof

Abstract

An artificial intelligence device according to an embodiment of the present disclosure may comprise: a memory for storing an energy prediction model trained through deep learning or a machine learning algorithm; and one or more processors for receiving an energy prediction request including a prediction request time point, and in response to the receiving of the energy prediction request, acquiring, through the energy prediction model, an energy prediction result from the prediction request time point to a time point after a preset time interval.

Description

Artificial intelligence device and its method of operation

[0001] The present invention relates to an artificial intelligence device, and more specifically, to an artificial intelligence device that provides a demand response service.

[0002] Demand Response Service (DRS) is a service that regulates users' power consumption in response to fluctuations in power demand within a power system. It plays a crucial role in maintaining power supply stability and reducing costs, particularly during periods of power shortages or peak demand.

[0003] Demand response services may include detailed services such as demand management services, incentive provision services, and real-time response services.

[0004] Demand management service is a service in which power companies request customers to reduce power consumption or postpone use for a certain period of time when power supply is insufficient or the power grid is overloaded.

[0005] The incentive service is a service that provides financial rewards to consumers who reduce their electricity usage during high-demand hours or switch to lower-demand hours.

[0006] Real-time response service is a service that sends signals to consumers in real time based on the status of the power grid or fluctuations in market prices, and enables immediate adjustment of power consumption accordingly.

[0007] However, the existing real-time response service, which is a type of demand response service, had a problem in that it could not respond at the time the user wanted, as it was a service that could only respond at specific times, such as when a request for a surplus power notification was received, when a request for a demand response from a power company was received, or when a network failure occurred.

[0008] The purpose of the present disclosure may be to provide a demand response service at a time desired by the user.

[0009] The purpose of the present disclosure may be to provide prediction results for the amount of energy from a desired point in time to a specific point in the future.

[0010] The purpose of the present disclosure may be to provide energy forecasts and energy schedules in response to requests from customers or power utility companies.

[0011] An artificial intelligence device according to one embodiment of the present disclosure may include: a memory for storing an energy prediction model learned through a deep learning or machine learning algorithm; and one or more processors that receive an energy prediction request including a prediction request time point, and in response to receiving the energy prediction request, obtain an energy prediction result from the prediction request time point to a time point after a preset time interval from the prediction request time point through the energy prediction model.

[0012] A method of operation of an artificial intelligence device according to one embodiment of the present disclosure may include the step of receiving an energy prediction request including a prediction request time point, and the step of obtaining an energy prediction result up to a time point after a preset time interval from the prediction request time point through an energy prediction model learned through a deep learning or machine learning algorithm in response to receiving the energy prediction request.

[0013] In a computer-readable recording medium having a program recorded thereon for performing a method of operation of an artificial intelligence device according to one embodiment of the present disclosure, the method of operation

[0014] The method may include the step of receiving an energy prediction request including a time point for the prediction request; and the step of obtaining an energy prediction result up to a time point after a preset time interval from the time point for the prediction request through an energy prediction model learned through a deep learning or machine learning algorithm in response to receiving the energy prediction request.

[0015] According to an embodiment of the present disclosure, position encoding can help an energy prediction model learn temporal patterns more effectively through sine and cosine functions that reflect periodicity. Through this, the energy prediction model can understand the temporal relationships of time series data and accurately predict daily recurring energy usage patterns or fluctuations in power generation.

[0016] In addition, according to an embodiment of the present disclosure, using location encoding information allows an energy prediction model to more accurately identify energy changes over time, thereby significantly improving prediction accuracy.

[0017] Furthermore, the energy prediction model according to the embodiment of the present disclosure can predict energy quantity patterns more accurately by reflecting the periodicity of time compared to existing models. Accordingly, demand response services such as on-demand calls can be provided efficiently.

[0018] FIG. 1 is a block diagram for explaining the components of an artificial intelligence device according to one embodiment of the present disclosure.

[0019] FIG. 2 is a drawing for explaining the configuration of an artificial intelligence server according to one embodiment of the present disclosure.

[0020] FIG. 3 is a flowchart for explaining the operation method of an artificial intelligence device according to an embodiment of the present disclosure.

[0021] FIG. 4 is a diagram illustrating the learning process of an energy prediction model according to one embodiment of the present disclosure.

[0022] FIG. 5 is a diagram illustrating a training data set used for training an energy prediction model according to one embodiment of the present disclosure.

[0023] FIG. 6 is a diagram illustrating a code that encodes time information through a position encoding method according to one embodiment of the present disclosure.

[0024] FIG. 7 is a diagram illustrating the process of outputting energy prediction results through an energy prediction model in response to an energy prediction request according to one embodiment of the present disclosure.

[0025] FIGS. 8 and 9 are drawings illustrating the predicted energy distribution output by an energy prediction model when a sine function is used as a periodic function according to an embodiment of the present disclosure.

[0026] FIG. 10 is a diagram for comparing the prediction error resulting from the application of a conventional energy prediction model and an energy prediction model according to an embodiment of the present disclosure.

[0027] Artificial intelligence refers to the field of researching artificial intelligence or the methodologies to create it, while machine learning refers to the field of researching methodologies to define the various problems addressed in the field of artificial intelligence and to solve them.

[0028] Machine learning is also defined as an algorithm that improves the performance of a task through consistent experience.

[0029] An Artificial Neural Network (ANN) is a model used in machine learning that can refer to a model capable of problem-solving, composed of artificial neurons (nodes) that form a network through the connection of synapses.

[0030] An artificial neural network can be defined by connection patterns between neurons in different layers, a learning process that updates model parameters, and an activation function that generates output values.

[0031] An artificial neural network may include an input layer, an output layer, and optionally one or more hidden layers. Each layer may include one or more neurons, and the artificial neural network may include synapses connecting the neurons. In an artificial neural network, each neuron may output a function value of an activation function for input signals, weights, and biases input through the synapses.

[0032] Model parameters refer to parameters determined through learning, including synaptic connection weights and neuron biases. Hyperparameters refer to parameters that must be set before training in a machine learning algorithm, including the learning rate, number of iterations, mini-batch size, and initialization function.

[0033] The objective of training an artificial neural network can be viewed as determining model parameters that minimize the loss function. The loss function can be used as an indicator to determine optimal model parameters during the training process of an artificial neural network.

[0034] Machine learning can be classified into supervised learning, unsupervised learning, and reinforcement learning depending on the learning method.

[0035] Supervised learning refers to a method of training an artificial neural network with labels provided for the training data; a label can refer to the correct answer (or result value) that the artificial neural network must infer when training data is input into it.

[0036] Unsupervised learning can refer to a method of training an artificial neural network without being given labels for the training data.

[0037] Reinforcement learning can refer to a learning method that trains an agent defined within an environment to select an action or sequence of actions that maximizes the cumulative reward in each state.

[0038] Machine learning implemented as a deep neural network (DNN) containing multiple hidden layers among artificial neural networks is also called deep learning, and deep learning is a part of machine learning.

[0039] In the following, machine learning is used to include deep learning.

[0040] FIG. 1 is a block diagram for explaining the components of an artificial intelligence device according to one embodiment of the present disclosure.

[0041] The artificial intelligence device (100) can be implemented as a fixed device or a mobile device, such as a TV, projector, mobile phone, smartphone, desktop computer, laptop, digital broadcasting terminal, PDA (personal digital assistants), PMP (portable multimedia player), navigation, tablet PC, wearable device, set-top box (STB), DMB receiver, radio, washing machine, refrigerator, desktop computer, digital signage, robot, vehicle, etc.

[0042] Referring to FIG. 1, the artificial intelligence device (100) may include a communication interface (110), an input interface (120), a learning processor (130), a sensor (140), an output interface (150), a memory (170), and a processor (180).

[0043] The communication interface (110) can transmit and receive data with external devices, such as other artificial intelligence devices or AI servers (200), using wired or wireless communication technology. For example, the communication interface (110) can transmit and receive sensor information, user input, learning models, control signals, etc., with external devices.

[0044] The communication technologies used by the communication interface (110) include GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), LTE (Long Term Evolution), 5G, WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Bluetooth (Bluetooth), RFID (Radio Frequency Identification), Infrared Data Association (IrDA), ZigBee, NFC (Near Field Communication), etc.

[0045] The input interface (120) can acquire various types of data.

[0046] The input interface (120) may include a camera (121) for capturing images, a microphone (122) for receiving audio signals, and a user input interface (123) for receiving information from a user.

[0047] A camera (121) or a microphone (122) can be treated as a sensor, and a signal obtained from the camera (121) or the microphone (122) can be named as sensing data or sensor information.

[0048] The input interface (120) can obtain input data to be used when obtaining an output using training data and a training model for model training. The input interface (120) may also obtain unprocessed input data, in which case the processor (180) or the learning processor (130) can extract input features as a preprocessing step for the input data.

[0049] The camera (121) processes image frames, such as still images or video, obtained by an image sensor in video call mode or shooting mode. The processed image frames may be displayed on a display (151) or stored in memory (170).

[0050] The microphone (122) processes external acoustic signals into electrical voice data. The processed voice data can be utilized in various ways depending on the function (or application running) being performed by the artificial intelligence device (100). Meanwhile, various noise removal algorithms can be applied to the microphone (122) to remove noise generated during the process of receiving external acoustic signals.

[0051] The user input interface (123) is for receiving information from a user, and when information is input through the user input interface (123), the processor (180) can control the operation of the artificial intelligence device (100) to correspond to the input information.

[0052] The user input interface (123) may include mechanical input means (or mechanical keys, such as buttons, dome switches, jog wheels, jog switches, etc. located on the front / rear or side of the artificial intelligence device (100)) and touch input means.

[0053] As an example, a touch input means may consist of a virtual key, soft key, or visual key displayed on a touchscreen through software processing, or a touch key placed on a part other than the touchscreen.

[0054] The learning processor (130) can train a model composed of an artificial neural network using training data. The trained artificial neural network can be called a learning model. The learning model can be used to infer a result value for new input data other than the training data, and the inferred value can be used as a basis for judgment to perform some action.

[0055] The learning processor (130) can perform AI processing together with the learning processor (240) of the AI ​​server (200).

[0056] The learning processor (130) may include memory integrated into or implemented in the artificial intelligence device (100). The learning processor (130) may also be implemented using memory (170), external memory directly coupled to the artificial intelligence device (100), or memory maintained in an external device.

[0057] The sensor (140) can acquire at least one of internal information of the artificial intelligence device (100), surrounding environment information of the artificial intelligence device (100), and user information using various sensors.

[0058] The sensor (140) may include one or more of a proximity sensor, an illuminance sensor, an accelerometer, a magnetic sensor, a gyroscope, an inertial sensor, an RGB sensor, an IR sensor, a fingerprint recognition sensor, an ultrasonic sensor, a light sensor, a microphone, a lidar sensor, and a radar sensor.

[0059] The output interface (150) can generate output related to visual, auditory, or tactile senses.

[0060] The output interface (150) may include a display (151) for outputting an image, an audio output interface (152) for outputting audio, a haptic device (153) for outputting tactile information, and a light output interface (154) for outputting light.

[0061] The display (151) displays (outputs) information processed by the artificial intelligence device (100). For example, the display (151) can display information on the execution screen of an application running on the artificial intelligence device (100), or UI (User Interface) and GUI (Graphic User Interface) information based on such execution screen information.

[0062] The display (151) can be implemented as a touch screen by forming a layered structure with the touch sensor or by being formed as an integral unit. The touch screen functions as a user input interface (123) that provides an input interface between the artificial intelligence device (100) and the user, and at the same time, can provide an output interface between the artificial intelligence device (100) and the user.

[0063] The audio output interface (152) can output audio data received from the communication interface (110) or stored in the memory (170) in call signal reception, call mode or recording mode, voice recognition mode, broadcast reception mode, etc.

[0064] The audio output interface (152) may include at least one of a receiver, a speaker, and a buzzer.

[0065] The haptic device (153) generates various tactile effects that the user can feel. A typical example of the tactile effect generated by the haptic device (153) can be vibration.

[0066] The light output interface (154) outputs a signal to indicate the occurrence of an event using the light of the light source of the artificial intelligence device (100). Examples of events occurring in the artificial intelligence device (100) may include receiving a message, receiving a call signal, a missed call, an alarm, a schedule notification, receiving an email, receiving information through an application, etc.

[0067] The memory (170) can store data that supports various functions of the artificial intelligence device (100). For example, the memory (170) can store input data, training data, training models, training history, etc. obtained from the input interface (120).

[0068] The processor (180) can determine at least one executable operation of the artificial intelligence device (100) based on information determined or generated using a data analysis algorithm or a machine learning algorithm.

[0069] The processor (180) can control the components of the artificial intelligence device (100) to perform a determined operation.

[0070] To this end, the processor (180) can request, search, receive, or utilize data from the learning processor (130) or memory (170), and can control the components of the artificial intelligence device (100) to execute a predicted operation or a preferred operation among the at least one executable operation.

[0071] If the processor (180) requires the connection of an external device to perform a determined operation, it can generate a control signal to control the external device and transmit the generated control signal to the external device.

[0072] The processor (180) can obtain intent information regarding user input and determine the user's requirements based on the obtained intent information.

[0073] The processor (180) can obtain intent information corresponding to user input by using at least one of a Speech To Text (STT) engine for converting voice input into a string or a Natural Language Processing (NLP) engine for obtaining intent information of natural language.

[0074] At least one of the STT engine or NLP engine may be composed of an artificial neural network, at least a portion of which is trained according to a machine learning algorithm. Additionally, at least one of the STT engine or NLP engine may be trained by a learning processor (130), trained by a learning processor (240) of an AI server (200), or trained by distributed processing thereof.

[0075] The processor (180) can collect history information, including the operation details of the artificial intelligence device (100) or user feedback regarding the operation, and store it in memory (170) or a learning processor (130), or transmit it to an external device such as an AI server (200). The collected history information can be used to update the learning model.

[0076] The processor (180) can control at least some of the components of the artificial intelligence device (100) to run an application stored in memory (170).

[0077] The processor (180) can operate two or more of the components included in the artificial intelligence device (100) in combination with each other to operate the application.

[0078] FIG. 2 is a drawing for explaining the configuration of an artificial intelligence server according to one embodiment of the present disclosure.

[0079] Referring to FIG. 2, the AI ​​server (200) may refer to a device that trains an artificial neural network using a machine learning algorithm or uses a trained artificial neural network.

[0080] The AI ​​server (200) may be composed of multiple servers to perform distributed processing and may be defined as a 5G network. The AI ​​server (200) may be included as part of the configuration of the artificial intelligence device (100) to perform at least some of the AI ​​processing together.

[0081] The AI ​​server (200) may include a communication interface (210), memory (230), a learning processor (240), and a processor (260).

[0082] The communication interface (210) can transmit and receive data with an external device, such as an artificial intelligence device (100).

[0083] The memory (230) may include a model memory (231). The model memory (231) may store a model (or artificial neural network, 231a) that is being learned or has been learned through the learning processor (240).

[0084] The learning processor (240) can train the artificial neural network (231a) using training data. The training model may be used while mounted on the AI ​​server (200) of the artificial neural network, or it may be used while mounted on an external device such as an artificial intelligence device (100).

[0085] The learning model may be implemented in hardware, software, or a combination of hardware and software. If part or all of the learning model is implemented in software, one or more instructions constituting the learning model may be stored in memory (230).

[0086] The processor (260) can use a learning model to infer a result value for new input data and generate a response or control command based on the inferred result value.

[0087] FIG. 3 is a flowchart for explaining the operation method of an artificial intelligence device according to an embodiment of the present disclosure.

[0088] The processor (180) may be provided in multiple units.

[0089] Referring to FIG. 3, the processor (180) of the artificial intelligence device (100) can receive an energy prediction request including a prediction time (S301).

[0090] The processor (180) can receive an energy prediction request from an external device through the communication interface (110).

[0091] An energy forecast request may include information regarding the forecast time. The forecast time may be a future point in time regarding the energy forecast result that the user wishes to know. The energy forecast result may include at least one of the amount of energy generated or the amount of energy consumed.

[0092] The processor (180) of the artificial intelligence device (100) can obtain an energy prediction result through an energy prediction model in response to a received energy prediction request (S303).

[0093] The energy prediction model may be a model based on an artificial neural network trained with a deep learning algorithm or a machine learning algorithm. The energy prediction model may be a model that outputs energy prediction results from a prediction point in time up to a point in time after a preset time interval. The energy prediction results may include the trend of energy amounts from the prediction point in time up to a point in time corresponding to the preset time interval.

[0094] The energy prediction model can be a transLSTM model. The transLSTM model can be a hybrid model that combines a Transformer model and an LSTM (Long Short-Term Memory) model.

[0095] The transLSTM model is explained with reference to the following figure.

[0096] FIG. 4 is a diagram illustrating the learning process of an energy prediction model according to one embodiment of the present disclosure.

[0097] In Fig. 4, the energy prediction model (400) may be a tranLSTM model.

[0098] Referring to FIG. 4, input data input to the energy prediction model (400) may include energy amount information, weather information, and time information. The input data of the energy prediction model (400) may include past energy amount information, past weather information, and time information corresponding to the past energy amount information and the past weather information. The time information may include information encoded according to a location encoding method based on a periodic function.

[0099] Input data may be in the form of sequence data or time series data. Input data may be referred to as sequence data or time series data.

[0100] Energy amount information may be the amount of energy at specific time intervals during a past period. The amount of energy may include at least one of the amount of power generated or the amount of energy consumed. The period may be one week or two weeks, and the specific time interval may be one hour or one minute, but this is merely an example.

[0101] Weather information (or meteorological information) may include at least one of wind speed, temperature, humidity, cloud cover, and precipitation at specific time intervals during a certain period in the past.

[0102] Time information may include information indicating the time corresponding to energy amount information and weather information.

[0103] Input data may be a set of input unit data obtained at specific time intervals over a certain period. Input data may include input unit data.

[0104] The processor (180) can convert input unit data containing corresponding energy amounts, weather information, and time information into an embedding unit vector.

[0105] For example, if the time information is 10:00 AM, one week prior to the present, the processor (180) can convert input unit data including energy amount, weather information, and time information (10:00 AM) corresponding to that time into an embedding unit vector. The processor (180) can transmit a plurality of embedding unit vectors to an energy prediction model (400).

[0106] FIG. 5 is a diagram illustrating a training data set used for training an energy prediction model according to one embodiment of the present disclosure.

[0107] Referring to FIG. 5, the training data set (500) may include a past data set (510) and a current data set (520). The training data set (500) may also include only the past data set (510).

[0108] The historical data set (510) may include energy amounts and weather information (wind speed, temperature, humidity, cloud cover, precipitation) collected during a past week period (D-7 to D-1). A1 may include energy amount information measured on the day (D-7) corresponding to one week prior to the day (D-day) including the prediction time point. A1 may include energy amounts for each hour for 24 hours.

[0109] Likewise, B1 may include wind speed information measured on a day (D-7) corresponding to one week prior to the day (D-day) including the prediction time point. B1 may include wind speeds for each hour for 24 hours.

[0110] Time information (or a specific point in time) can be matched with energy amount information and weather information. For example, the specific point in time may be a point in time representing 10:00 AM on a day (D-7) corresponding to one week prior to the day (D-day) containing the prediction point in time. The specific point in time can be encoded through a location encoding method and used to train an energy prediction model (400) together with energy amount and weather information.

[0111] The current data set (520) may include hourly predicted energy amounts and hourly predicted weather information for the day including the current time (the day including the prediction time). For example, if the current time is 11:00 AM and the prediction time is the energy prediction amount for 24 hours from 4:00 PM on D-day, A8 may include hourly predicted energy amounts from 0:00 AM to 24:00 PM on D-day, and B8 may include hourly predicted wind speeds from 0:00 AM to 24:00 PM on D-day.

[0112] The processor (180) can convert energy amount and weather information corresponding to a first specific point in time in the past into a first embedding unit vector. The processor (180) can add first time information (or first order information) of the first input unit data to the first embedding unit vector through a positional encoding method. The processor (180) can generate first period information from the first time information through a period input model to be described later, and can reflect the generated first period information in the first embedding unit vector.

[0113] The processor (180) can input the first embedding unit vector into the energy prediction model (400).

[0114] The processor (180) can convert energy amount and weather information corresponding to a second specific point in time in the past into a second embedding vector. The second specific point in time may be a point in time having a certain time interval from the first characteristic point in time.

[0115] The processor (180) can add time information (or sequence information) of the second input unit data to the second embedding unit vector through a positional encoding method. The processor (180) can generate second period information from the second time information through a period input model and can reflect the generated second period information in the second embedding unit vector.

[0116] Period information can be referred to as position encoding information.

[0117] FIG. 6 is a diagram illustrating a code that encodes time information through a position encoding method according to one embodiment of the present disclosure.

[0118] Figure 6 may be a code corresponding to the operation performed by the transformer encoder (410).

[0119] In Fig. 6, the code can be written using the Python language.

[0120] Referring to FIG. 6, the first code line (601) may be a line that loads the NumPy library. NumPy may be used as an alias np.

[0121] The second code line (602) may be a line defining a positional_encoding_with_sine function that outputs positional encoding information using a sine function. The time variable of the positional_encoding_with_sine function may be assigned a number or a list containing numbers, and an integer between 0 and 23 may be used. The period variable of the positional_encoding_with_sine function may be assigned a period value (24).

[0122] The third code line (603) may be a line that converts the value assigned to the time variable into a real number from 0 to 1.

[0123] The fourth code line (604) may be a line that calculates position encoding information using a sine function.

[0124] The fifth code line (605) may be a line that returns a positional_encoding value. The calculated positional_encoding value may be referred to as positional encoding information.

[0125] For example, the location encoding information for 10:00 AM (sin(10)) may be 0.866, and the location encoding information for 15:00 PM (sin(15)) may be -0.707.

[0126] Again, Figure 4 is explained.

[0127] The energy prediction model (400) may include a transformer encoder (410), an LSTM model (420), and a dense layer (430). The transformer encoder (410) may be referred to as the transformer model (410).

[0128] The transformer encoder (410) can convert an embedding unit vector containing position encoding information into a time series feature vector.

[0129] The transformer encoder (410) can learn the cyclic characteristics between multiple embedding unit vectors.

[0130] The transformer encoder (410) can learn how data at each time point is related to each other using a self-attention mechanism. The transformer encoder (410) can learn the correlation between power generation, weather information, and time information.

[0131] The transformer encoder (410) can sequentially output the transformed time series feature vectors for each time point.

[0132] The LSTM model (420) can sequentially process time series feature vectors output from the transformer encoder (410). The LSTM model (420) can learn the pattern of change in power generation over time. The LSTM model (420) has strengths in learning long-term dependencies, so it can effectively analyze the long-term impact of weather or time changes on power generation.

[0133] The LSTM model (420) can sequentially process time series feature vectors output from the transformer encoder (410) to output a prediction vector or a sequence of prediction vectors corresponding to the predicted energy amount from a specific point in the future to a certain time interval.

[0134] The dense layer (430) can post-process the prediction vector or prediction vector sequence output to the LSTM model (420) to output an energy prediction result including the predicted energy amount for a certain time interval from a specific point in the future. The post-processing can perform at least one of dimensionality reduction, weighted sum calculation, and activation function application.

[0135] The quantile layer (440) can evaluate the uncertainty of the energy prediction result (post-processing result) output from the density layer (430). The quantile layer (440) can output multiple quantiles for the energy prediction result through a quantile regression method. Quantile regression is a technique for predicting a specific quantile value based on the distribution of data, and may be a technique for predicting a value in a specific range instead of a general average prediction.

[0136] The loss function used in the quantile layer (440) may be Pinball Loss. Pinball Loss measures the prediction error in a specific quantile and can impose a loss differentially for under- or over-prediction.

[0137] The quantile layer (440) may be included in the energy prediction model (400) or provided separately.

[0138] FIG. 7 is a diagram illustrating the process of outputting energy prediction results through an energy prediction model in response to an energy prediction request according to one embodiment of the present disclosure.

[0139] In FIG. 7, the artificial intelligence device (100) may receive an energy forecast request requesting an energy forecast result for a future point in time, 9:00 AM. The energy forecast request may be any one of a surplus power alarm, a request from the grid, or a call due to a network failure.

[0140] The artificial intelligence device (100) can transmit a data set (710) to an energy prediction model (400). The data set (710) may be a set including a past data set (510) and a current data set (520) as illustrated in FIG. 5.

[0141] The artificial intelligence device (100) can obtain a data set from memory (170) or an AI server (200).

[0142] The artificial intelligence device (100) may further include a period input model (720). The period input model (720) may be a model that converts a specific time point into period information. The period input model (720) may convert 9:00 AM into positional encoding information through the get_positional_encoding_with_sine function illustrated in FIG. 5.

[0143] The energy prediction model (400) can output an energy prediction result for an amount of energy from a prediction time point to a certain time interval based on location encoding information based on a data set (710) and a prediction request time point.

[0144] The artificial intelligence device (100) can control the charging or discharging of the ESS battery based on the output energy prediction result.

[0145] FIGS. 8 and 9 are drawings illustrating the predicted energy distribution output by an energy prediction model when a sine function is used as a periodic function according to an embodiment of the present disclosure.

[0146] FIG. 8 shows a first energy prediction distribution (810) by applying a position encoding method through first period information (820) having the form of a sine function when the prediction request time is 00:00. Quantiles 1 to 9 included in the first energy prediction distribution (810) can represent the minimum range to the maximum range of the energy prediction results.

[0147] The first energy prediction distribution (810) is the predicted result for 2 days, where the horizontal axis represents time and the vertical axis represents the amount of energy.

[0148] The first period information (820) can represent a period function corresponding to the amount of energy for 2 days from 00:00. The horizontal axis of the first period information (820) is time, and the vertical axis can represent the value of the period function.

[0149] Quantiles 1 to 9 included in the first energy prediction distribution (810) may have a shape similar to a graph (811) representing the actual energy amount distribution.

[0150] FIG. 9 shows a second energy prediction distribution (910) by applying a position encoding method through second period information (920) having the form of a sine function when the prediction request time is 09 o'clock. Quantiles 1 to 9 included in the second energy prediction distribution (910) can represent the minimum range to the maximum range of the energy prediction results.

[0151] The second energy prediction distribution (910) is the result of predictions over two days, where the horizontal axis represents time and the vertical axis represents the amount of energy.

[0152] The second period information (920) can represent a period function corresponding to the amount of energy for 2 days starting from 09:00. The horizontal axis of the second period information (920) is time, and the vertical axis can represent the value of the period function.

[0153] Quantiles 1 to 9 included in the second energy prediction distribution (910) may have a shape similar to a graph (911) representing the actual energy amount distribution.

[0154] That is, according to an embodiment of the present disclosure, it can be confirmed that an energy prediction model using location encoding information has an energy prediction distribution similar in form to an actual energy distribution.

[0155] In this way, location encoding can help energy forecasting models learn temporal patterns more effectively through sine and cosine functions that reflect periodicity. This enables energy forecasting models to understand the temporal relationships of time series data and accurately predict daily recurring energy usage patterns or fluctuations in power generation.

[0156] In addition, using location encoding information allows energy prediction models to more accurately identify energy changes over time, which can significantly improve prediction accuracy.

[0157] Again, Figure 3 is explained.

[0158] The memory (170) can store an energy prediction model. The processor (180) can learn the energy prediction model and store it in the memory (170). In another embodiment, the processor (180) can receive an energy prediction model learned by the artificial intelligence server (200) from the artificial intelligence server (200).

[0159] One or more processors (180) of the artificial intelligence device (100) can control power based on the acquired energy prediction results (S305).

[0160] In one embodiment, one or more processors (180) can control the charging and discharging of an Energy Storage System (ESS) battery based on the energy prediction results from a prediction point in time to a point in time after a certain time interval.

[0161] One or more processors (180) can control to charge the ESS battery when the predicted amount of power generation exceeds the demand amount, and can control to discharge the ESS battery when the predicted amount of power generation is less than the demand amount.

[0162] One or more processors (180) can transmit energy prediction results obtained through the communication interface (110) to a user's terminal.

[0163] FIG. 10 is a diagram for comparing the prediction error resulting from the application of a conventional energy prediction model and an energy prediction model according to an embodiment of the present disclosure.

[0164] Referring to FIG. 10, the hourly prediction error rates are shown when a conventional energy prediction model is used and when an energy prediction model (400) according to an embodiment of the present disclosure is used.

[0165] Existing energy forecasting models may be models that predict the energy consumption or power generation of the following day based on midnight (00:00). Existing energy forecasting models typically use sequential time indices for each hour (e.g., 0:00, 1:00, 2:00, ..., 23:00) as input, or they may use simple time labels. While existing energy forecasting models input time information sequentially, they do not explicitly provide information that time repeats periodically.

[0166] The energy prediction model (400) according to an embodiment of the present disclosure may be a model that predicts energy consumption and power generation by reflecting the periodicity of time.

[0167] When averaging the hourly prediction error rates, it can be seen that the error rate has improved by 11.2% (16.7%-5.5%).

[0168] As such, the energy prediction model according to the embodiment of the present disclosure can predict energy quantity patterns more accurately by reflecting the periodicity of time compared to existing models. Accordingly, demand response services such as on-demand calls can be provided efficiently.

[0169] An artificial intelligence device (100) according to one embodiment of the present disclosure may include: a memory (170) that stores an energy prediction model (400) learned through a deep learning or machine learning algorithm; and one or more processors (180) that receive an energy prediction request including a prediction request time and, in response to receiving the energy prediction request, obtain an energy prediction result from the prediction request time to a time after a preset time interval from the prediction request time through the energy prediction model.

[0170] The above energy prediction model (400) can obtain the energy prediction result from the time of the prediction request, the time information corresponding to each of the past energy amount information and the past weather information, and the past energy amount information.

[0171] Each of the above time information and the above prediction request time point can be input into the energy prediction model (400) as information encoded according to a position encoding method based on a periodic function.

[0172] The above energy prediction model (400) may include a tranLSTM model that combines a transformer model (410) and an LSTM (Long Short-Term Memory) model (420).

[0173] One or more processors (180) can convert the past energy amount information, the past weather information, and the time information into embedding unit vectors, the transformer model (410) can learn the cyclic characteristics between a plurality of embedding unit vectors to output time series feature vectors, and the LSTM model (420) can sequentially process the output time series feature vectors to output a prediction vector or a sequence of prediction vectors corresponding to the energy prediction result.

[0174] The above energy prediction model (400) may further include a dense layer (430) that post-processes the output prediction vector or prediction vector sequence and a quantile layer (440) that obtains the energy prediction result based on the post-processing result obtained from the dense layer through a quantile regression method.

[0175] The above one or more processors (180) can control the charging or discharging of an energy storage system (ESS) based on the acquired energy prediction results.

[0176] The above-described disclosure can be implemented as computer-readable code on a medium on which a program is recorded. A computer-readable medium includes all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable media include a Hard Disk Drive (HDD), a Solid State Disk (SSD), a Silicon Disk Drive (SDD), ROM, RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc. Additionally, the computer may include a processor (180) of an artificial intelligence device.

Claims

In artificial intelligence devices, Memory for storing an energy prediction model learned through a deep learning or machine learning algorithm; and One or more processors comprising receiving an energy prediction request including a prediction request time point, and in response to receiving said energy prediction request, obtaining an energy prediction result from said prediction request time point to a time point after a preset time interval from said prediction request time point through said energy prediction model. Artificial intelligence device. In paragraph 1, The above energy prediction model Obtaining past energy amount information, past weather information, time information corresponding to each of the past energy amount information and past weather information, and obtaining the energy prediction result from the time of the prediction request Artificial intelligence device. In paragraph 2, Each of the above time information and the above prediction request time point is Information encoded according to a position encoding method based on a periodic function, which is input into the above energy prediction model Artificial intelligence device. In paragraph 3, The above energy prediction model including a tranLSTM model that combines a Transformer model and an LSTM (Long Short-Term Memory) model Artificial intelligence device. In paragraph 4, The above one or more processors Convert the above past energy amount information, the above past weather information, and the above time information into embedding unit vectors, and The above Transformer model output time-series feature vectors by learning the cyclic characteristics between multiple embedding unit vectors, and The above LSTM model Processing the output time series feature vectors sequentially to output a prediction vector or a sequence of prediction vectors corresponding to the energy prediction result. Artificial intelligence device. In paragraph 5, The above energy prediction model A dense layer for post-processing the above-mentioned output prediction vector or prediction vector sequence and A quantile layer further comprising obtaining the energy prediction result based on the post-processing result post-processed in the dense layer through a quantile regression method Artificial intelligence device. In paragraph 1, The above one or more processors Controlling the charging or discharging of an Energy Storage System (ESS) based on acquired energy prediction results Artificial intelligence device. In the method of operating an artificial intelligence device, A step of receiving an energy forecast request including the time of the forecast request; and A step comprising: in response to receiving the above energy prediction request, obtaining an energy prediction result up to a point in time after a preset time interval from the time of the prediction request through an energy prediction model learned via a deep learning or machine learning algorithm. Method of operation of an artificial intelligence device. In paragraph 8, The above energy prediction model A model that obtains past energy amount information, past weather information, time information corresponding to each of the past energy amount information and past weather information, and the energy prediction result from the time of the prediction request. Method of operation of an artificial intelligence device. In Paragraph 9, Each of the above time information and the above prediction request time point is Information encoded according to a position encoding method based on a periodic function, which is input into the above energy prediction model Method of operation of an artificial intelligence device. In Paragraph 10, The above energy prediction model including a tranLSTM model that combines a Transformer model and an LSTM (Long Short-Term Memory) model Method of operation of an artificial intelligence device. In Paragraph 11, The method further includes the step of converting the past energy amount information, the past weather information, and the time information into embedding unit vectors. The above Transformer model output time-series feature vectors by learning the cyclic characteristics between multiple embedding unit vectors, and The above LSTM model Processing the output time series feature vectors sequentially to output a prediction vector or a sequence of prediction vectors corresponding to the energy prediction result. Method of operation of an artificial intelligence device. In Paragraph 12, The above energy prediction model A dense layer for post-processing the above-mentioned output prediction vector or prediction vector sequence and A quantile layer further comprising obtaining the energy prediction result based on the post-processing result post-processed in the dense layer through a quantile regression method Method of operation of an artificial intelligence device. In Paragraph 13, A step of further controlling the charging or discharging of an Energy Storage System (ESS) based on the acquired energy prediction result Method of operation of an artificial intelligence device. In a computer-readable recording medium on which a program for performing a method of operation of an artificial intelligence device is recorded, The above method of operation A step of receiving an energy forecast request including the time of the forecast request; and A step comprising: in response to receiving the above energy prediction request, obtaining an energy prediction result up to a point in time after a preset time interval from the time of the prediction request through an energy prediction model learned via a deep learning or machine learning algorithm. Recording media. In paragraph 15, The above energy prediction model A model that obtains past energy amount information, past weather information, time information corresponding to each of the past energy amount information and past weather information, and the energy prediction result from the time of the prediction request. Recording media. In Paragraph 16, Each of the above time information and the above prediction request time point is Information encoded according to a position encoding method based on a periodic function, which is input into the above energy prediction model Recording media. In Paragraph 17, The above energy prediction model including a tranLSTM model that combines a Transformer model and an LSTM (Long Short-Term Memory) model Recording media. In Paragraph 18, The method further includes the step of converting the past energy amount information, the past weather information, and the time information into embedding unit vectors. The above Transformer model output time-series feature vectors by learning the cyclic characteristics between multiple embedding unit vectors, and The above LSTM model Processing the output time series feature vectors sequentially to output a prediction vector or a sequence of prediction vectors corresponding to the energy prediction result. Recording media. In Paragraph 19, The above energy prediction model A dense layer for post-processing the above-mentioned output prediction vector or prediction vector sequence and A quantile layer further comprising obtaining the energy prediction result based on the post-processing result post-processed in the dense layer through a quantile regression method Recording media.

Images