Home energy management system and bidirectional real-time pricing prediction method for demand response thereof

Abstract

Disclosed are a home energy management system and a bidirectional real-time pricing prediction method for a demand response thereof. The bidirectional real-time pricing prediction method for a demand response may comprise: generating future power pricing prediction information by applying user-related information and real-time power pricing to a deep learning-based prediction model; generating and then analyzing a power consumption scheduling result for smart home devices on the basis of same to calculate the amount of power moved per hour and a flexible device consumption ratio; calculating a movement adjustment value and a consumption adjustment value, respectively, by using the amount of power moved per hour and the flexible device consumption ratio; calculating an incentive-penalty weight value according to a power consumption type by using the movement adjustment value and the consumption adjustment value; and deriving bidirectional real-time power pricing by reflecting same in the power pricing prediction information.

Description

Household energy management system and bidirectional real-time price forecasting method for its demand response

[0001] The present invention relates to a household energy management system and a bidirectional real-time price prediction method for the demand response thereof.

[0002]

[0003] Recent advancements in Internet of Things (IoT) technology and smart metering infrastructure have enabled smart home users to schedule real-time (RT) power consumption through home energy management systems (HEMS). This power consumption scheduling via HEMS is known as residential demand response (DR), which is an effective method for changing power demand by adjusting mobile and controllable loads. Price-based DR has been regarded as a promising means of shifting users' peak loads, and reasonable electricity pricing mechanisms can directly influence customers' ability to participate in DR. To implement price-based DR, real-time pricing (RTP) has proven to be an effective pricing mechanism that reduces users' electricity bills and mitigates grid peaks through load shifting.

[0004] While conventional technology by Wang and Paranjape et al. demonstrated that RTP could reduce loads and electricity rates during peak periods, it was found that another peak rebound could occur at other times, potentially increasing the PAR ratio. To address this, Anees and Chen reduced both electricity rates and PAR (peak-to-average ratio, hereinafter referred to as PAR) by integrating sloped block rates with RTP to set a user consumption threshold. Sloped block rates were also used to manage PAR and flatten power consumption while improving the social well-being of customers. However, these conventional technologies cannot easily accommodate the diverse electricity demands of end users due to the complexity of the residential sector and the presence of various appliances operating at different times of the day. Furthermore, because they are not directly related to actual prices, they can lead to inefficient behavioral changes and provide less flexible services.

[0005] Considering these limitations, recently proposed pricing mechanisms for virtual power plants based on customized rebate packages, decentralized RTP frameworks based on compensation fairness, and hybrid pricing mechanisms that consider both RTP and RT incentives have been suggested.

[0006] While there have been significant efforts to develop new pricing mechanisms that provide financial rewards to incentivize customer participation in DR, existing pricing mechanisms tend to offer the same energy price levels to specific user groups or regions. Furthermore, although these mechanisms may influence consumers' energy consumption patterns to some extent, they fall short in guiding them toward long-term, sustainable energy usage habits. This is because, even if consumers are influenced by RTP, their energy consumption behavior is not successfully reflected in energy prices. Consequently, the unidirectional nature of current electricity pricing mechanisms can dampen consumer motivation to participate in residential DR. Therefore, to encourage active consumer participation in residential DR, it is necessary to develop a new, customized bidirectional pricing mechanism that allows consumers to set their own electricity rates through bidirectional pricing.

[0007]

[0008] The present invention is intended to provide a household energy management system and a bidirectional real-time price forecasting method for the demand response thereof.

[0009] In addition, the present invention aims to provide a home energy management system capable of autonomously determining hourly power prices using the user's hourly mobile power and flexible device consumption ratio, and a bidirectional real-time price prediction method for the demand response thereof.

[0010] In addition, the present invention aims to provide a home energy management system capable of optimizing real-time spatial-temporal power prices through a deep learning-based prediction model and scheduling the operation of home appliances based on the predicted prices, and a bidirectional real-time price prediction method for the same.

[0011] Furthermore, the present invention aims to provide a home energy management system capable of contributing to inducing user participation in demand response by providing incentives or penalties according to the consumer's power consumption patterns, and a bidirectional real-time price prediction method for the same.

[0012]

[0013] According to one aspect of the present invention, a bidirectional real-time price prediction method for demand response is provided.

[0014] According to one embodiment of the present invention, a bidirectional real-time price prediction method for demand response may be provided, comprising: (a) obtaining user-related information and real-time power prices for a smart home user; (b) applying the user-related information and real-time power prices to a deep learning-based prediction model to generate future power price prediction information; (c) generating power consumption scheduling results for smart home devices based on the power price prediction information; (d) analyzing the power consumption scheduling results to calculate the hourly shifted power amount and flexible device consumption ratio, and calculating a shift adjustment value and a consumption adjustment value, respectively, using the hourly shifted power amount and flexible device consumption ratio; (e) calculating an incentive-penalty weight value according to the power consumption pattern using the shift adjustment value and the consumption adjustment value; and (f) deriving a bidirectional real-time power price by reflecting the incentive-penalty weight value in the power price prediction information.

[0015] The above deep learning-based prediction model can input time series data regarding user-related information and real-time power prices into a first deep learning model to extract spatial features, and apply the spatial features to a second deep learning model to generate power price prediction information that reflects temporal patterns.

[0016] The first deep learning model is a USCNN (unsupervised shallow convolutional neural network) based model, and the second deep learning model is a nested LSTM based model.

[0017] The above movement adjustment value is calculated using the following mathematical formula, but,

[0018]

[0019] Here, λ RTP (t) represents the actual real-time power price, and represents the minimum and maximum values ​​of bidirectional real-time power prices, and represents the amount of power transferred over time, and and represents the maximum credit score values ​​in the positive and negative directions.

[0020] The above consumption adjustment value is calculated using the following mathematical formula, but,

[0021]

[0022] Here, represents the actual real-time power price, and represents the minimum value of bidirectional real-time power prices, and It represents a credit score for the consumer's flexible device consumption ratio, and represents the default score, represents the maximum credit score for the flexible device consumption ratio.

[0023]

[0024] The above user-related information may include load, device information, previous consumption patterns, and whether a flexible device is used.

[0025] The above incentive-penalty weight value is calculated using the following mathematical formula, but,

[0026]

[0027] Here, represents the movement adjustment value, represents the consumption adjustment value of the flexible device, and represents the weighting of changes in real-time power prices.

[0028]

[0029] According to another aspect of the present invention, a system for performing a bidirectional real-time price prediction method for demand response is provided.

[0030] According to one embodiment of the present invention, a home energy management system may be provided, comprising: a memory for storing at least one instruction; and a processor for the instruction stored in the memory, wherein the instruction executed by the processor comprises: (a) obtaining user-related information and real-time power price for a smart home user; (b) applying the user-related information and real-time power price to a deep learning-based prediction model to generate future power price prediction information; (c) generating power consumption scheduling results for smart home devices based on the power price prediction information; (d) analyzing the power consumption scheduling results to calculate the hourly shifted power amount and flexible device consumption ratio, and calculating a shift adjustment value and a consumption adjustment value, respectively, using the hourly shifted power amount and flexible device consumption ratio; (e) calculating an incentive-penalty weight value according to the power consumption pattern using the shift adjustment value and the consumption adjustment value; and (f) reflecting the incentive-penalty weight value in the power price prediction information to derive a bidirectional real-time power price.

[0031]

[0032] By providing a home energy management system and a bidirectional real-time price prediction method for demand response thereto according to one embodiment of the present invention, it is possible to autonomously determine hourly power prices using the user's hourly mobile power and the ratio of flexible device consumption.

[0033] In addition, the present invention can optimize real-time spatial-temporal power prices through a deep learning-based prediction model and schedule the operation of home appliances based on the predicted prices.

[0034] In addition, the present invention has the advantage of contributing to inducing user participation in demand response by providing incentives or penalties according to the consumer's power consumption patterns.

[0035]

[0036] FIG. 1 is a flowchart illustrating a bidirectional real-time price prediction method for demand response in PHEMS according to one embodiment of the present invention.

[0037] FIG. 2 is a drawing illustrating the overall architecture of PHEMS according to one embodiment of the present invention.

[0038] FIG. 3 is a diagram illustrating the HSP and HFA segmentation section according to one embodiment of the present invention.

[0039] FIG. 4 is a drawing illustrating the difference between a conventional model and a first deep learning model according to an embodiment of the present invention.

[0040] FIG. 5 is a diagram illustrating an nLSTM architecture according to an embodiment of the present invention.

[0041] FIG. 6 is a diagram illustrating pseudocode of a bidirectional real-time price prediction method for demand response in PHEMS according to an embodiment of the present invention.

[0042] FIG. 7 is a diagram illustrating a power consumption determination timeline according to an embodiment of the present invention.

[0043] FIG. 8 is a block diagram schematically illustrating the internal configuration of a household energy management system according to one embodiment of the present invention.

[0044]

[0045] As used in this specification, singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as "composed" or "comprising" should not be interpreted as necessarily including all of the various components or steps described in the specification, and should be interpreted as meaning that some of the components or steps may be excluded, or that additional components or steps may be included. Furthermore, terms such as "...part," "module," etc., as used in the specification refer to a unit that processes at least one function or operation, which may be implemented in hardware or software, or a combination of hardware and software.

[0046] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0047]

[0048] FIG. 1 is a flowchart illustrating a bidirectional real-time price prediction method for demand response in PHEMS according to an embodiment of the present invention, FIG. 2 is a diagram illustrating the overall architecture of PHEMS according to an embodiment of the present invention, FIG. 3 is a diagram illustrating the HSP and HFA segmentation sections according to an embodiment of the present invention, FIG. 4 is a diagram illustrating the difference between a conventional and a first deep learning model according to an embodiment of the present invention, FIG. 5 is a diagram illustrating an nLSTM architecture according to an embodiment of the present invention, FIG. 6 is a diagram illustrating pseudocode of a bidirectional real-time price prediction method for demand response in PHEMS according to an embodiment of the present invention, and FIG. 7 is a diagram illustrating a power consumption determination timeline according to an embodiment of the present invention.

[0049] In step 110, PHEMS obtains real-time power prices every hour.

[0050] For example, PHEMS can obtain real-time pricing (RTP) every hour from the utility company server.

[0051] In step 115, PHEMS obtains user-related information. Here, user-related information may include load, device information, previous consumption patterns and whether a flexible device is used, etc.

[0052] The entire architecture of PHEMS is illustrated in Fig. 2. It will be briefly explained with reference to Fig. 2.

[0053] According to one embodiment of the present invention, PHEMS is described assuming that a smart meter and an IoT hub are installed in a smart home (household) so that a user can record data and control the power of home appliances, as shown in FIG. 2.

[0054] According to one embodiment of the present invention, PHEMS can be used to predict future real-time power prices (RTP) by analyzing user-related information such as the load of each smart home, hourly shifted power (HSP), and household flexible appliance consumption ratio (HFA). In addition, weather information, etc., may be further included as user-related information.

[0055] To this end, PHEMS is integrated with the IoT Hub and can acquire user-related information (e.g., load information for devices) updated through the IoT Hub in real time, and use this to reflect changes in power consumption and contributions to load peak reduction in future RTP forecasts. This will be more clearly understood through the explanation below.

[0056] According to one embodiment of the present invention, PHEMS analyzes the amount of power moved by time and the ratio of flexible device consumption, respectively, to determine the degree and capability of response to the demand response of smart home (household) users.

[0057] I will explain this in more detail.

[0058] Household appliances may include a shifted load (SL), a non-shiftable load (NSL), and a control load (CL) as illustrated in FIG. 2.

[0059] Hourly shifted power (HSP) represents the difference in power consumption before and after a demand response. If the load exceeds the average hourly power demand at a given time t, this is referred to as peak time (t p It can be considered as ). However, if the load at a given time t is below the average hourly power demand, it is decided to set this as valley time. Here, valley time is the opposite concept of peak time and represents a time when power consumption is reduced and electricity rates are relatively low.

[0060] Therefore, the HSP of consumer i for time t on day k can be expressed as Equation 1.

[0061]

[0062] Here, D BDR (t) and D t ADR (t) represents consumer demand before and after the demand response. Consumers are encouraged to shift energy consumption from peak hours to low-load hours, but in reality, energy consumption may increase unexpectedly.

[0063] thus, can be expressed as a positive or negative number. Using the energy consumption for user i over k days A discrete ordered list set of values ​​can be obtained, arranged from the smallest value to the largest value, and a credit score can be calculated by dividing it into specific intervals as shown in FIG. 3.

[0064] Generally, flexible household appliances can be divided into movable load appliances (SL), non-movable loads (NSL), and controllable loads (CL). SL appliances include dishwashers and washing machines, while NSL appliances may include refrigerators and lighting fixtures. Electric vehicles and home energy storage devices can be considered flexible appliances.

[0065] It can be defined as the ratio of energy consumed by flexible devices to the total power of household appliances. As this ratio increases, the structure of the user's device can be improved. Just like It can also be divided into specific sections, as shown in Fig. 3.

[0066] The weight of evidence method (WOE: hereinafter referred to as WOE) is used to analyze end-user responsiveness and ability to respond to in-home demand response (DR). Each user's credit score is calculated based on the WOE, which can have a direct impact on the user's real-time price as a positive or negative signal.

[0067] The binary variable φ indicates whether the selected evaluation index affects the household's power consumption behavior. If φ=1, it means that it affects the power consumption behavior.

[0068] Logistic regression, which represents the relationship between p{φ=1|X}, can be expressed as Equation 2. Here, X i={X HSP , X HFA} am.

[0069]

[0070]

[0071]

[0072] Here, α, ρ1, and ρ2 represent the regression model parameters. Additionally, p(φ=1) represents the evaluation index X for changes in consumer energy behavior. i ={X HSP , X HFA It represents the probability of being correlated with}. Mathematical Equation 2 can be transformed into Mathematical Equation 3. To improve the standardization of different types of indicators, Mathematical Equation 4 can be used to calculate the impact of each value of a specific attribute variable on the classification result.

[0073] p(φ=1|X p =X pq ) and p(φ=0|X p =X pq ) is attribute variable X p =X pq Given , they represent the probabilities that the sample categories are φ=1 and φ=0, respectively. As WOE increases, the probability and weight of φ=1 may increase. Evaluation index X i ={X HSP , X HFA Based on the optimal partial result of} X i The WOE transformation of can be expressed as Equation 5.

[0074]

[0075] Here, μ m and γ n represents a virtual binary variable, and X i If the value of belongs to the m-th or n-th category, μ respectively m =1 or γ n =1. Therefore, after WOE conversion, Equation 3 can be expressed as Equation 6.

[0076]

[0077] To enhance the ability of Equation 6 to intuitively predict the performance of the end user's demand response (DR), it can be converted into a linear expression of the ratio logarithm as in Equation 7.

[0078]

[0079] Here, F base R represents a constant calculated using known score values ​​by utilizing a method that assigns the exact score expected at a specific ratio or adopts a score obtained by doubling the ratio. These methods are used to improve the practicality and applicability of credit score models [X m-1 , X m The average credit score can be used within the range. Here, the specific interval to which the evaluation index value belongs can be identified to determine the score corresponding to that interval.

[0080] In step 120, PHEMS applies user-related information and real-time power prices to a pre-trained deep learning-based prediction model to generate future power price prediction information for the remaining scheduling period.

[0081] In step 125, PHEMS can generate scheduling results for smart home devices based on power price forecast information.

[0082] In step 130, PHEMS calculates the hourly shifted power amount and flexible device consumption ratio based on the scheduling results, and calculates the shift adjustment value and the consumption adjustment value, respectively, using the hourly shifted power amount and flexible device consumption ratio.

[0083] In step 135, PHEMS calculates an incentive-penalty weight value according to the power consumption pattern using the movement adjustment value and the consumption adjustment value.

[0084] In step 140, PHEMS derives bidirectional real-time power prices by reflecting incentive-penalty weighting values ​​in the power price prediction information.

[0085] As described above, PHEMS can generate future power price prediction information for the remaining scheduling period using user-related information updated every hour and real-time power prices, and solve optimization problems based on this to execute optimal scheduling of home appliances.

[0086] For the sake of ease of understanding and explanation, we will first explain the formulation of the optimization problem.

[0087] I will explain this in more detail.

[0088] Two-way real-time power pricing (hereinafter referred to as CBi-RTP) provides positive incentives for smart home users to use power efficiently, while imposing penalties for excessive consumption. As a result, the higher the smart home user's HFA, the lower the user's actual electricity bill, and this can induce the user to actively improve the configuration of their home appliances.

[0089] The CBi-RTP based on the evaluation model can be expressed as Equation 8.

[0090]

[0091]

[0092] Here, λ i (t) represents CBi-RTP after PHEMS optimization, and λ RTP (t) represents the actual RTP. and are respectively λ i Represents the minimum and maximum values ​​of (t). Also, Δλ i (t) represents the incentive-penalty weight based on the user's power consumption behavior. For example, Δλ iIf (t) is positive, it acts as an incentive, whereas if it is negative, it can act as a penalty. In the following, incentive or penalty should be understood to refer to the incentive-penalty weight. The discounted or increased RTP of user i at time t is defined as in Equation 10 and can be reconstructed as in Equation 11.

[0093]

[0094]

[0095] Here, and represents the RTP change coefficient for user i's HSP and HFA. and represent user i's credit score for HSP and HFA, respectively. δ represents the RTP change weight, and F i,0 represents the base score considering user i's initial score. X HSP In this case, positive incentives or negative penalties can change consumption behavior to reduce the peak period of the load, and as a result, reduce the CBi-RTP for power consumption. This can serve to induce end users to regulate consumption.

[0096]

[0097] Here, and represent the maximum credit score values ​​in the positive and negative directions, respectively. If X HSP If >0, the acquired represents a positive value, and as a result, (t) may decrease.

[0098] Likewise, X HFA The CBi-RTP change caused by can be expressed as Equation 13. Here, represents the maximum credit score for HFA.

[0099]

[0100] The goal of smart home applications is to minimize the electricity consumption costs of smart homes by considering SL, NSL, and CL devices throughout the entire time while ensuring user convenience.

[0101] As mentioned above, home appliances in a smart home can be classified into three categories: SLs, NSLs, and CLs.

[0102] SLs are appliances such as dishwashers, washing machines, and ovens that can be scheduled to operate during non-peak hours, which can reduce hourly electricity costs and lower overall electricity costs. These SL appliances can have binary operating states of “on” and “off”.

[0103] NSLs include refrigerators and lighting, and the demand for them cannot be adjusted and must be met.

[0104] CLs are devices such as air conditioners (AC), electric water heaters (EWH), and plug-in electric vehicles (PEVs) that can operate flexibly within a predefined range of energy consumption and have continuous variables.

[0105] The goal of residential home applications is to minimize the electricity consumption costs of a smart home by considering SL, NSL, and CL devices throughout the entire time while ensuring user comfort. Therefore, the objective function of PHEMS can be expressed as Equation 14.

[0106]

[0107] Here, and represents the power demand of the SL, NSL, and CL devices, respectively, which means the values ​​at time t for the a-th, b-th, and c-th devices, respectively. The first part of Equation 14 represents the cost given according to CBi-RTP λ(t) for the current slot, and the second part represents the set cost of future slots considering future price predictions based on the proposed USCNN-nLSTM model.

[0108] The constraints on these home appliances are as follows.

[0109] SL device constraints: Constraints of SL, such as washing machines, dishwashers, and electric ovens, can be expressed as Equations 15 to 17.

[0110]

[0111]

[0112]

[0113] Here, a represents a general scheduleable machine, and a=1, 2, ⪋, N a am. and and represent the deadline and the time required for the SL to complete the task, respectively. If a scheduleable machine is planned to start operation at time t, the task is k of the operation cycle. a It continues without interruption until the stage ends (represented by Equation 15). The constraints in Equations 16 and 17 limit the start time and the required time, respectively. Finally, the total demand for SL can be expressed as in Equation 18.

[0114]

[0115] Here, , and 1 indicates the operating state at time t, otherwise it means that it is not operating. represents the power of household appliances.

[0116] CL device constraints: Compared to SL, CL devices such as AC, EWH, and PEV can flexibly use energy between the lowest and highest demand, which can be expressed as Equation 19.

[0117]

[0118] Here, represents the minimum energy consumption of the b-th CL device. In particular, in one embodiment of the present invention, indoor temperature and hot water temperature are taken into account, which are two important factors affecting home user comfort through the adjustment of power usage between AC and EWH.

[0119] Regulating indoor temperature using AC is one of the most critical loads in a home, as it directly affects the comfort of residents. Variable operation of AC can be applied to reduce operating costs by utilizing the physical properties of thermal inertia in smart home structures, which can be represented as shown in Equation 20.

[0120]

[0121] Here, and and represent the indoor and outdoor temperatures at time t, respectively, and X AC (t) indicates the operating state of the AC, where 1 means the AC is operating and 0 means it is not operating. M a , c a , R er represents the mass of indoor air, the specific heat of air, and the equivalent thermal resistance of the house, respectively. In addition, P AC represents AC power, and the indoor temperature at each time point can be calculated using Equation 20. This is for cooling and can be easily modified to model AC heating operation.

[0122] Also, AC operating status X AC(t) is affected by the current indoor temperature relative to the defined allowable indoor temperature, which can be expressed as Equation 21.

[0123]

[0124]

[0125] Here, and and represent the lower and upper limits of the indoor temperature, respectively. Therefore, Equation 18 can be expressed as Equation 22.

[0126] During WEH operation, the principle of energy conservation is applied, and the water tank temperature at time t after using hot water can be expressed as Equation 23.

[0127]

[0128] Here, T inlet represents the temperature after hot water use, the water temperature at time t, the volume of the water tank, the amount of hot water used, and the inlet water temperature, respectively. In Equation 23, the temperature of the cold water is used as a reference point, and it is assumed that the energy is zero (0).

[0129] When the EWH operates at rated power, the water tank temperature at time t after heating can be expressed as Equation 24.

[0130]

[0131] Here X EWH (t) represents the operating state of the EWH at time t, where 1 means operating and 0 means not operating. C w represents the specific heat of water, and M w represents the mass of water when the EWH is full. Finally, the operating state of the EWH can be affected by the current temperature and the upper and lower limits of the EWH temperature, which can be expressed as in Equation 25.

[0132]

[0133]

[0134] Here, and and represent the lower and upper limits of the allowable water temperature, respectively, and based on this, Equation 18 for EWH can be expressed as Equation 26.

[0135] PEVs are considered as CL devices in PHEMS and can contribute to improving the scheduling performance of smart homes and reducing electricity costs. To predict the plug-in power of a PEV, the influence of driving range can be expressed as in Equation 27, and battery constraints can be expressed as in Equation 28.

[0136]

[0137]

[0138] Here, represents the PEV energy when plugged in, the minimum energy of the PEV, and the PEV energy when plugged out, respectively. E CC represents the battery energy consumption per kilometer and is set to 0.159 kWh / km. and represent the minimum and maximum states of PEV energy, respectively, and produce. Given For wa d The amount of can be calculated using mathematical formula 27. However, in reality The value of is known, but Since and d are unknown, the given for The conditional probability of can be calculated as in Equation 29.

[0139]

[0140] Here, M A2B2 PEV battery energy when plugged in When plugging out It represents the conditional probability that becomes. Therefore, the kinetics of the PEV over a day can be expressed as Equation 30.

[0141]

[0142] Here, and ε₀ represents the power of the PEV, the maximum and minimum charge / discharge power of the PEV, and the efficiency of the PEV, respectively. P PEV (t) can have a positive or negative value indicating charging or discharging. Finally, the daily power consumption of a CL appliance considering AC, EWH, and PEV can be expressed as in Equation 31.

[0143]

[0144] In addition, according to one embodiment of the present invention, since the CBi-RTP price is based on RTP, it must be within a reasonable price range to increase applicability. Accordingly, the CBi-RTP constraint can be expressed as Equation 32.

[0145]

[0146] In a smart home, power balance constraints must be guaranteed. These power balance constraints can be expressed as Equation 33.

[0147]

[0148] Here, P Grid (t) represents the power purchased or sold from the grid at time t to ensure power consumption.

[0149] According to one embodiment of the present invention, energy costs can be provided to consumers one hour in advance to enable the management of power consumption of various home appliances in a smart home with various characteristics. To this end, a deep learning model can be used to predict future RTPs whenever the current real-time power price (RTP) is provided. Accordingly, future RTPs can be repeatedly performed every hour when the real-time power price (RTP) is provided.

[0150] To this end, we will explain deep learning models.

[0151] Convolutional Neural Networks (hereinafter referred to as CNNs) are widely adopted for price forecasting because they can fully utilize the interrelationships between RTPs and their specific structures to extract more important features. However, conventional CNNs can easily face problems such as overfitting and vanishing gradients due to the large number of parameters. Furthermore, while CNNs assume that the spatial characteristics of input data are invariant, RTPs actually exhibit spatial variability.

[0152] Therefore, in one embodiment of the present invention, to address this, a first deep learning model based on a USCNN (unsupervised shallow convolutional neural network) is utilized to improve data feature quality and increase RTP prediction accuracy. The USCNN consists of unsupervised convolution (USC), pooling, and fully connected layers, among which the USC layer is a core component. Input and output It is assumed that... Here, W, K, and F represent the window length, kernel size, and number of features, respectively. Thus, the USCNN model can be expressed as Equation 34.

[0153]

[0154] Here, y i,j wa b i,j , xi,a , h i,a and represent the elements and bias matrix of row i and column b, and the elements and weights of row i and column a, respectively. The weights of h are the matrix h depending on the changes in i and j. i,a Since they differ, they cannot be shared with each other. Figure 4 shows the difference between a conventional CNN and a USCNN. The large box boundary represents a feature map, and the small square represents a convolutional kernel sliding on the map to extract features. The black arrows indicate the convolutional steps of the kernels sliding at different locations. In Figure 4 (a), kernels marked with the same color share the same weight parameters, whereas in Figure 4 (b), USCNN kernels are marked with different patterns to indicate that their weight parameters are different. Since the USCNN model cannot extract time-series features, the deep learning model according to one embodiment of the present invention may learn temporal features using a second deep learning model (nLSTM).

[0155] According to one embodiment of the present invention, a first deep learning model (USCNN) can extract spatial features by receiving user-related information and real-time power prices as time series data.

[0156] The second deep learning model (nLSTM) can analyze spatial features extracted from the first deep learning model and analyze periodic changes and RTP rules in past data.

[0157] The LSTM model is an improved network of the RNN that relies on additional memory units to overcome the vanishing gradient or runaway problems, and is widely used for time-series forecasting such as load, solar, and wind power price prediction. Since power price values ​​possess distinct time-series features, strong temporal evolution patterns, and long-term dependencies, it is possible to explore the rules of time-series fluctuations in datasets obtained using LSTMs. In LSTMs, memory and gate cells can effectively recognize and capture the long-term dependencies of the target sequence. Generally, an LSTM network consists of a series of LSTMs, and the output of each layer can be processed as the input to the next layer.

[0158] In one embodiment of the present invention, a nested LSTM (nLSTM) model is used to effectively solve the problem of temporal variability and to allow selective access to internal memory when generating the time hierarchy of RTP. In nLSTM, memory functions are nested instead of stacking, allowing for easy expansion into a deep architecture.

[0159] FIG. 5 is a diagram illustrating the structure of a second deep learning model (nLSTM) according to an embodiment of the present invention. As shown in FIG. 5, the memory cell value of the nLSTM is calculated using an LSTM architecture, which acts as an internal unit for its own memory cell. Subsequently, long-term information trained by the internal unit can be selectively learned and transmitted using traditional LSTM gates. Thus, this process enables the internal memory to learn and process events such as price spikes over the long term, which is particularly useful when such events are irrelevant to the immediate present. The internal update formulas of the nLSTM are given by Equations 35 to 41.

[0160]

[0161]

[0162]

[0163]

[0164]

[0165]

[0166]

[0167] Here, And h t represents the input gate state, forget gate state, memory cell, output, and hidden state, respectively. x t , σ, W, and b represent the input, sigmoid, weight matrix, and bias vector, respectively. The temporal features of RTP can be iteratively computed based on nLSTM.

[0168] To summarize, PHEMS receives the Real-Time Price (RTP) every hour t, updates user-related information (necessary information such as power demand) as input to the deep learning model, and generates future RTPs (power price forecasts) for the remaining scheduling period through the previously trained deep learning model. Subsequently, PHEMS can generate scheduling results by utilizing the predicted RTPs to make optimal decisions that satisfy the objective function for the home appliances. Scheduling Results It is the same as. Here, is a discrete variable, and is a continuous variable that can operate within a predefined range. The HSP and HFA values ​​of a smart home user are divided into specific intervals, and credit scores corresponding to the HSP and HFA can be provided. Subsequently, a discounted / increased RTP is calculated, and a customized CBi-RTP for a specific user can be derived.

[0169] According to one embodiment of the present invention, PHEMS can obtain power consumption decisions for the current and remaining time slots by optimizing every hour t, as shown in FIG. 7. However, regarding the power demand of a household, only the decision for the current time is implemented, thereby providing optimal energy management guidelines for the smart home. Through this iterative process, PHEMS is made robust against price uncertainty and can ensure dynamic adjustment of control policies and self-correction of the model when new information is received and changes occur in the operating environment. Furthermore, by enabling smart home users to participate in electricity pricing, there is an advantage in increasing the enthusiasm of household users to participate in Residential Demand Response (DR) and significantly enhancing economic benefits. The complete pseudocode for this is shown in FIG. 6.

[0170]

[0171] FIG. 8 is a block diagram schematically illustrating the internal configuration of a household energy management system according to one embodiment of the present invention.

[0172] Referring to FIG. 8, a household energy management system according to one embodiment of the present invention is configured to include a memory (810) and a processor (820).

[0173] The memory (810) stores at least one instruction for performing a bidirectional real-time price prediction method for demand response according to one embodiment of the present invention.

[0174] The processor (820) can execute instructions stored in memory (810). The instructions executed by the processor (820) can each perform a series of processes to receive real-time power price (RTP) at every hour t, update user-related information (necessary information such as power demand) as input to a deep learning model, generate future RTP (power price prediction information) for the remaining scheduling period through a previously trained deep learning model, generate scheduling results through an optimal decision satisfying an objective function for home appliances using the predicted RTP, calculate the amount of power shifted by hour and the consumption ratio of flexible devices based on this, calculate a shift adjustment value and a consumption adjustment value using the said amount of power shifted by hour and the consumption ratio of flexible devices, calculate an incentive-penalty weight value according to the power consumption pattern using the shift adjustment value and the said consumption adjustment value, and reflect this in the power price prediction information to derive a bidirectional real-time power price.

[0175] As this is the same as explained with reference to FIGS. 1 to 7, a redundant explanation will be omitted.

[0176]

[0177] An apparatus and method according to an embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the computer-readable medium may be those specifically designed and configured for the present invention, or they may be those known and available to a person skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

[0178] The hardware device described above may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

[0179] The present invention has been described above with reference to its embodiments. Those skilled in the art will understand that the present invention may be implemented in modified forms without departing from the essential characteristics of the invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the invention is defined by the claims, not by the foregoing description, and all variations within the scope of the claims should be interpreted as being included in the invention.

Claims

1. (a) A step of obtaining user-related information and real-time power prices for smart home users; (b) a step of generating future power price prediction information by applying the above user-related information and real-time power prices to a deep learning-based prediction model; (c) a step of generating power consumption scheduling results for smart home devices based on the above power price prediction information; (d) a step of analyzing the power consumption scheduling results to calculate the hourly shifted power amount and the flexible device consumption ratio, and using the hourly shifted power amount and the flexible device consumption ratio to calculate the shift adjustment value and the consumption adjustment value, respectively; (e) a step of calculating an incentive-penalty weighting value according to the power consumption pattern using the above movement adjustment value and the above consumption adjustment value; and (f) A bidirectional real-time price forecasting method for demand response comprising the step of deriving a bidirectional real-time power price by reflecting the above incentive-penalty weighting values ​​in the above power price forecasting information.

2. In Paragraph 1, A bidirectional real-time price prediction method for demand response, characterized in that the deep learning-based prediction model inputs time series data regarding user-related information and real-time power prices into a first deep learning model to extract spatial features, and applies the spatial features to a second deep learning model to generate power price prediction information reflecting temporal patterns.

3. In Paragraph 2, The above-mentioned first deep learning model is a USCNN (unsupervised shallow convolutional neural network) based model, and A bidirectional real-time price prediction method for demand response, characterized in that the second deep learning model is a nested LSTM-based model.

4. In Paragraph 1, A bidirectional real-time price forecasting method for demand response, characterized in that the above-mentioned movement adjustment value is calculated using the following mathematical formula. Here, represents the actual real-time power price, and and represents the minimum and maximum values ​​of bidirectional real-time power prices, and represents the amount of power transferred over time, and and represents the maximum credit score values ​​in the positive and negative directions.

5. In Paragraph 1, A bidirectional real-time price forecasting method for demand response, characterized in that the above consumption adjustment value is calculated using the following mathematical formula. Here, represents the actual real-time power price, and represents the minimum value of bidirectional real-time power prices, and It represents a credit score for the consumer's flexible device consumption ratio, and represents the default score, represents the maximum credit score for the flexible device consumption ratio.

6. In Paragraph 1, A bidirectional real-time price prediction method for demand response, characterized in that the above user-related information includes load, device information, previous consumption patterns, and whether a flexible device is used.

7. In Paragraph 1, A bidirectional real-time price forecasting method for demand response, characterized in that the above incentive-penalty weighting values ​​are calculated using the following mathematical formula. Here, represents the movement adjustment value, represents the consumption adjustment value of the flexible device, and represents the weighting of changes in real-time power prices.

8. A computer-readable recording medium having program code for performing the method according to claim 1.

9. In household energy management systems, Memory for storing at least one instruction; and The instructions stored in the above memory include a processor, The instructions executed by the above processor are, respectively, (a) A step of obtaining user-related information and real-time power prices for smart home users; (b) a step of generating future power price prediction information by applying the above user-related information and real-time power prices to a deep learning-based prediction model; (c) a step of generating power consumption scheduling results for smart home devices based on the above power price prediction information; (d) a step of analyzing the power consumption scheduling results to calculate the hourly shifted power amount and the flexible device consumption ratio, and using the hourly shifted power amount and the flexible device consumption ratio to calculate the shift adjustment value and the consumption adjustment value, respectively; (e) a step of calculating an incentive-penalty weighting value according to the power consumption pattern using the above movement adjustment value and the above consumption adjustment value; and (f) A household energy management system characterized by performing a step of deriving a bidirectional real-time power price by reflecting the above incentive-penalty weighting value in the above power price prediction information.

10. In Paragraph 9, A home energy management system characterized by the above-described deep learning-based prediction model inputting time series data regarding user-related information and real-time electricity prices into a first deep learning model to extract spatial features, and applying the spatial features to a second deep learning model to generate electricity price prediction information reflecting temporal patterns.

11. In Paragraph 10, The above-mentioned first deep learning model is a USCNN (unsupervised shallow convolutional neural network) based model, and A home energy management system characterized in that the second deep learning model is a nested LSTM-based model.

12. In Paragraph 9, A home energy management system characterized by the above-mentioned movement adjustment value being calculated using the following mathematical formula. Here, represents the actual real-time power price, and and represents the minimum and maximum values ​​of bidirectional real-time power prices, and represents the amount of power transferred over time, and and represents the maximum credit score values ​​in the positive and negative directions.

13. In Paragraph 9, A household energy management system characterized by the above consumption adjustment value being calculated using the following mathematical formula. Here, represents the actual real-time power price, and represents the minimum value of bidirectional real-time power prices, and It represents a credit score for the consumer's flexible device consumption ratio, and represents the default score, represents the maximum credit score for the flexible device consumption ratio.

14. In Paragraph 9, A household energy management system characterized by the above incentive-penalty weighting value being calculated using the following mathematical formula. Here, represents the movement adjustment value, represents the consumption adjustment value of the flexible device, and represents the weighting of changes in real-time power prices.

Images