WO2017154040A1

WO2017154040A1 - Method for energy balancing using group formation for customers

Info

Publication number: WO2017154040A1
Application number: PCT/JP2016/001308
Authority: WO
Inventors: Shantanu Chakraborty; Toshiya Okabe
Original assignee: Nec Corporation
Priority date: 2016-03-09
Filing date: 2016-03-09
Publication date: 2017-09-14

Abstract

An energy balancing group formation method is provided. Historical demand profiles of a plurality of customers are received. A demand aggregation criterion is determined according to the received historical energy demand profiles. A demand aggregation strategy to form energy balancing groups based on the demand aggregation criterion is determined. Each energy balancing group includes a group of customers determined by using the demand aggregation criterion from the plurality of customers.

Description

[Title established by the ISA under Rule 37.2] METHOD FOR ENERGY BALANCING USING GROUP FORMATION FOR CUSTOMERS

　　The present invention relates to an energy balancing group formation method, and, for example, relates to an energy balancing group formation method to form energy balancing groups in demand side based on demands of customers.

　　Generally, an energy service provider (ESP) supplies a plurality of customers with energy such as electric power or natural gas. Thus, the ESP has to appropriately manage the energy supply to the customers. Therefore, demand (or load) aggregation is very important for the ESP in a deregulated environment. The ESP can serve multiple commercial settings (e.g. apartment buildings, commercial buildings, shopping mall, factory, etc.).

　　Various types of demand (or service) aggregations for energy supply or distribution have been already disclosed (e.g. Patent Literatures 1 to 4).

Patent Literature 1: United States Patent Publication No. 2002/0138176
Patent Literature 2: United States Patent Publication No. 2003/0040845
Patent Literature 3: United States Patent Publication No. 2011/0029341
Patent Literature 4: United States Patent No. 12/406,003

　　In order to appropriately manage energy supply, it is essential for the ESP to effectively and strategically identify the demand of customers based on grouping for energy balancing purpose (balancing group). Groupings as such are also necessary for service and price differentiation among multiple groups. Thus, it is necessary for the ESP to effectively identify clusters of similar customers (e.g. buildings) and aggregate identified similar customers for forming energy balancing groups based on a certain criterion. Hence, a generation measure of a criterion for demand aggregation (demand aggregation criterion) is needed.

　　The present invention has been made in view of the above-mentioned problem, and an object of the present invention is to form energy balancing groups each including similar customers.

　　An aspect of the present invention is an energy balancing group formation method including: receiving historical demand profiles of a plurality of customers, determining a demand aggregation criterion according to the received historical energy demand profiles, and determining a demand aggregation strategy to form energy balancing groups based on the demand aggregation criterion, each energy balancing group including a group of customers determined by using the demand aggregation criterion from the plurality of customers.

　　According to the present invention, it is possible to form energy balancing groups each including similar customers.

Fig. 1 is a block diagram schematically illustrating an outline of an energy balancing group formation method according to a first embodiment; Fig. 2 is a flow chart illustrating the procedure for determining the demand aggregation criterion (DAC); Fig. 3 is a flow chart illustrating procedure for generating the demand aggregation strategy; Fig. 4 is a diagram schematically illustrating an example of DAC sorting in ascending order; Fig. 5 is a flow chart illustrating procedure of generating a statistical model of an observation and creating samples using Bayesian MCMC in a step S24; Fig. 6 is the statistical model parameter dependency that is utilized while performing Bayesian MCMC sampling; Fig. 7 is a diagram schematically illustrating an example of sub groups; Fig. 8 is a diagram schematically illustrating an example of equivalent probability (posterior) distribution of an articulated customer; Fig. 9 is a diagram schematically illustrating an example of posterior distributions of λ₁ and λ₂ with their associated likelihoods; and Fig. 10 is a diagram schematically illustrating energy balancing groups formed by the energy balancing group formation method according to the first embodiment.

　　Exemplary embodiments of the present invention will be described below with reference to the drawings. In the drawings, the same elements are denoted by the same reference numerals, and thus a repeated description is omitted as needed.

　　Following embodiments are described to enable any person skilled in the art to make and use the disclosure, and are provided with the context of a particular application and its requirements. Embodiment of a system, a method and a computer program product (e.g. software) for demand based energy balancing group formation is described. This operation may be performed by an Energy Service Provider (ESP). In particular, the ESP uses historical demand profiles of the customers (inputs of the system) to determine a demand aggregation criterion value (a demand aggregation criterion value). The process is conducted in a group formation engine that also hosts a demand aggregator. The demand aggregator forms multiple energy balancing groups based on the criterion by utilizing a probabilistic programming algorithm.

　　The present invention identifies two influential statistical measures and designs a criterion combining these criteria for the demand aggregation. The present invention designs a demand aggregation strategy using the criterion that creates multiple energy balancing groups.

First Embodiment
　　An energy balancing group formation method according to a first embodiment will be described. Fig. 1 is a block diagram schematically illustrating an outline of the energy balancing group formation method according to the first embodiment. In the energy balancing group formation method according to the first embodiment, historical demand profiles of a plurality of customers are used as input data 101 and information representing energy balancing groups are generated as output data 102 by processing the input data 101, for example, in a group formation engine 100. In the example of Fig. 1, the group formation engine 100 executes processes for the energy balancing group formation method. Here, an engine represents a system of software or hardware to assuming or execute functions for processing data. In Fig. 1, the group formation engine 100 executes a demand aggregation criterion (DAC) determination 103 and a demand aggregation 104.

　　First, the DAC determination 103 will be described. In the present embodiment, statistical measures play roles in analyzing demand signal as well as to providing several insights related to the empirical load behavior. Therefore, these measures are incredibly important for deciding the demand aggregation criterion. The criteria defined for the demand aggregations are designed considering two statistical measures. They are listed as follows:
1. Medians of load factor (MLF) of daily demand
2. Maximum of Standard Deviation in periodic demands (MSD)

　　An operation of the energy balancing group formation method for determining the demand aggregation criterion (DAC) according to the first embodiment will be descried in detail. Hereinafter, a customer identification parameter c represents each of the customers. NC represents number of the customers and the customer identification parameter c can vary from 1 to NC, where

.

Fig. 2 is a diagram schematically illustrating procedure for determining the demand aggregation criterion (DAC).

Step S11
　　The customer identification parameter c is initially set to one (c=1).

Step S12
　　A load factor LF_i is determined from a daily demand profile of the customer corresponding to the customer identification parameter c for each day. Here, i (i is an integer equal to or more than two) represents a day in the predetermined period (also referred to as a sampling period) (e.g. between a week, a month, a year, etc.), and a day indicated by i is referred to as a Day(i). In the present embodiment, the load factor LF_i is calculated as a ratio of an average value of demand in the Day(i) (e.g. daily demand curve of an apartment building) and peak demand in the Day(i). Thus, the load factor represents a stability of demand profile (e.g. demand curve or demand pattern) over the sampling period. Specifically, the load factor LF_i is expressed by a following expression, where D_i(t) represents the demand in a particular period (also referred to as a unit period) in the Day(i) and N represent number of unit periods in the Day(i) (e.g. N=24 (a single unit period is one hour), N=48 (a single unit period is 30 minutes)).

Thus, ND daily load factors LF₁ to LF_ND are calculated, where ND represents number of days included in the sampling period.

Step S13
　　A median MLF_c of the load factors LF_i (LF₁ to LF_ND) of the customer indicated by the customer identification parameter c is calculated as follows.

Thus, the calculated medians for NC customers can be expressed by a vector VMLF.

Step S14
　　A maximum valued of standard deviations of demands (MV_t) at the samplings times 1 to N over the sampling period is calculated. First, the standard deviations σ₁ to σ_N of the demands in the unit periods t over the sampling period are calculated, where

.

Next, the maximum values of standard deviations MV_c (for a customer, c) is calculated using the calculated standard deviations σ₁ to σ_N as follows.

For example, in a case of considering the demand in sampled on January, the standard deviation of 10 A.M. is the statistical standard deviation over all the demands at 10 A.M. accumulated in January. Thus, the calculated maximum values of standard deviations for a customer c MV_c for N sampling times can be expressed by a vector VMV.

Step S15
　　Whether the customer identification parameter c is equal to NC (c=NC) is determined.

Step S16
　　When the customer identification parameter c is smaller than NC (c<NC), the customer identification parameter c is incremented by one (c=c+1). After that, the procedure returns back to the step S12.

Step S17
　　When the customer identification parameter c is equal to NC (c=NC), each of the vectors VMLF and the vector VMV is normalized. These two measures conceptually bring contradictory valuations; i.e. while higher value of VMLF is considered better from ESP's perspective (as far as predictability is concerned), lower value of VMV is considered better from the same perspective. Therefore, the normalization process should be conducted in opposite direction for each of the vectors. Thus, the maximum value of elements of the vector VMLF is converted into and the maximum value of elements of the vector VMV is converted into one (while the minimum value is converted into zero). Thus, a normalized vector NMLF is generated by normalizing the vector MLF and the normalized vector NMV is generated by normalizing the vector VMV.

Note that, NMLF_c and NMV_c represent the associated normalized values for the customer c.

Step S18
　　A demand aggregation criterion DAC_c of demand aggregation for each customer indicated by the customer identification parameter c is calculated.

Where w_MLF and w_MV are coefficients for weighting the normalizing values NMLF_c and NMV_c. w_MLF and w_MV can be appropriately defined depending on a purpose of group formation. For example, w_MLF and w_MV can be defined such that w_MLF =0.5 and w_MV =0.5.

　　Next, the demand aggregation 104 will be described. Here, an operation of the energy balancing group formation method for generating the demand aggregation strategy according to the first embodiment will be descried in detail. The demand aggregation strategy performs a demand grouping scheme that will provide several effective balancing groups of customers. The strategy first determines the DAC_c for all customers using their historical demand profiles. The strategy then utilizes a probabilistic programming approach to perform a grouping scheme on the DAC_c. The demand aggregation strategy applies a probabilistic programming approach utilizing Bayesian Markov Chain Monte Carlo (MCMC) method on the DAC values.

　　Fig. 3 is a flow chart illustrating the procedure for generating the demand aggregation strategy.

Step S21
　　The demand aggregation criteria DAC_c (DAC₁ to DAC_NC) are sorted in ascending order (i.e. from a lower value to a higher value, or from the minimum value to the maximum value). Fig. 4 is a diagram schematically illustrating an example of DAC sorting in ascending order. In the example of Fig. 4, NC=104.

Step S22
　　A first observation is performed on the all customers since the customers are not divided as of this moment. Therefore, a parameter S representing a customer of a start point of the observation is set to one. Then, a parameter E representing a customer of an end point of the observation is set to NC.

Step S23
　　The observation is represented by a OB [S:E]. OB[S:E] means an observation in a range from S-th customer to E-th customer.

Step S24
　　A statistical model of the observation is built and samples are created using Bayesian Markov Chain Monte Carlo (MCMC). Here, procedure of the step S24 will be described in detail. Fig. 5 is a flow chart illustrating the procedure of generating the statistical model of the observation and creating the samples using Bayesian MCMC in the step S24. This procedure of the step S24 described below includes steps S241 to S246. In this procedure, the model of to be fitted on the demand aggregation criteria DAC_c is generated using model parameters λ₁ and λ₂, and a model parameter τ representing identification number of an articulated (border) customer. The model parameter λ₁ is a model parameter for a range of the identification numbers equal to or smaller than the articulated customer, the model parameter λ₂ is a model parameter for a range of the identification numbers larger than the articulated customer. Note that the term "articulated customer" is defined as a customer the DAC of which differs significantly with the same of preceding customers when the customers are ordered in an ascending order. In other words, the articulated customers define boundaries of potential energy balancing groups. Every customer is equally likely to be an articulated customer as a prior belief (before an observation of DAC is made). Therefore, the probability (the parameter τ) of a particular customer to be the articulated customer is uniformly distributed over the number of customers.

Step S241
　　A hyper-parameter α for controlling other parameters is determined. In the present embodiment, the hyper-parameter α is used to parameterized the model parameters λ₁ and λ₂. The hyper-parameter α is set as an inverse of expected observations and is expressed by a following expression.

Where M represents number of customers in the observations expressed and expressed by M=E-S+1.

Step S242
　　The statistical model of the present embodiment is generated. Specifically, the prior distributions of the model parameters λ₁ and λ₂ are generated as exponential distributions as follows:

Subsequently, each costumer has the same probability of being the articulated customer in the initial state, so that the prior distribution of the model parameter τ is generated as a discrete uniform distribution as follows:

Step S243
　　A deterministic distribution λ is generated by combining with the model parameters λ₁ and λ₂. The model parameters λ₁is arranged in a range of the identification numbers equal to or smaller than the articulated customer, the model parameter λ₂ is arranged in a range of the identification numbers larger than the articulated customer as follows:

Step S244
　　A hypothetical Poisson distribution, which is parameterized by the model parameter λ being an expected value, is fitted with the observation by sampling through Bayesian MCMC.

Step S245
　　Posterior distributions of the model parameters τ, λ₁ and λ₂ are obtained from the sampling. Details thereof will be described in an example described below.

Step S246
　　An expected DAC_c (EDAC_c) for all observations is calculated using posterior distributions of the model parameters τ, λ₁ and λ₂.

Step S25
　　Referring back to Fig. 3, the steps S24 to S30 will be described. In the step S25, Δ_dac for observation is calculated as follows.

Where β is a regulatory parameter usually set within [0, 1].

Step S26
　　Whether Δ_dac is smaller than a predetermined threshold value T is determined. T can be appropriately set depending on request of the ESP, etc.

Step S27
　　When Δ_dac is smaller than the predetermined threshold value T, an articulated customer AC is calculated using the posterior distribution of the model parameter τ as follows:

Step S28
　　The present observation is divided into two distinct observations (sub groups). One of the observations is set as follows:

Step S29
　　The other of the observations is set as follows:

　　After the steps S28 and S29, the procedure is returned to the step S24 to execute recursive process. Thus, the steps S24 to S28 are executed using each of the divided observations.

Step S30
　　When Δ_dac is equal to or larger than the predetermined threshold value T, a group is formed by using S and E. That is, the (E-S+1) customers (from S-th customer to E-th customer in the DAC value distribution) belong to the formed group.

　　As described above, the articulated customer divides the observation into two distributions: one is DAC observations (customers) before the articulated customer and the other is DAC observations (customers) after the articulated customer. Note that the articulated customer can belong to either of these distributions as appropriate. Both of these distributions are modeled as exponential distributions and controlled by the hyper parameter α. Fig. 6 shows the model parameter dependency that is utilized by the Bayesian MCMC method.

Fig. 7 is a diagram schematically illustrating an example of sub groups. In Fig. 7, the observation is divided into two sub groups. There is a significant change in the DAC around the customer No. 87.

Fig. 8 is a diagram schematically illustrating an example of equivalent probability (posterior) distribution of the articulated customer. This distribution serves as the basis of group boundary determination. It can be understood that the probability of the customer No.87 is maximum and thus is selected as the articulated customer.

Fig. 9 is a diagram schematically illustrating an example of posterior distributions of λ₁ and λ₂ with their associated likelihoods. For instance, the distribution of λ₁ is squeezed around DAC value of 19.17 with a smaller standard deviation (0.5) whereas the distribution of λ₂ is around 48. That basically states that, the DAC values of the customers before the customer No. 87 do not change as significantly as they do after the customer No. 87. Since there is a significant change in the marginal expected DAC_c (i.e. Δ_dac is higher), this group will be further go through the recursive processes (one for the customers No. 1 to 87 while another for the customers No. 88 to 104) in order to form groups of similar customers.

Fig. 10 is a diagram schematically illustrating energy balancing groups formed by the energy balancing group formation method as a result of the step S30 according to the first embodiment. In this example, four energy balancing groups G1 to G4 are formed over the DAC value of the customers.

　　According to the energy balancing group formation method according to the present embodiment, each customer belongs to the group consisting of similar customers, so that the ESP can recognize the trend of customer's demand group by group. Therefore, the ESP can easily make a reasonable plan for supplying energy to the customers. Additionally, since energy balancing group includes the similar customers and energy imbalance therein can be expected to be relatively small value, the ESP can control and reduce the energy imbalance in each energy balancing group in real time based on monitoring the real-time demand of the customers.

　　Further, since the ESP can recognize the trend of customer's demand group by group, the ESP can make a reasonable price plan (price menu) for each energy balancing group based on the DAC value.

Other embodiment
　　Note that the present invention is not limited to the above exemplary embodiments and can be modified as appropriate without departing from the scope of the invention. In the first embodiment, the example where the demand aggregation criteria DAC_c (DAC₁ to DAC_NC) are sorted in ascending order is described. However, it is merely exemplary. Therefore, it should be appreciated that the demand aggregation criteria DAC_c (DAC₁ to DAC_NC) are sorted in descending order (i.e. from a higher value to a lower value, or from the maximum value to the minimum value).

　　In the above exemplary embodiments, the present invention is described as a method configuration, but the present invention is not limited to this. According to the present invention, any processing can be implemented by causing a CPU (Central Processing Unit) to execute a computer program. The program can be stored and provided to a computer using any type of non-transitory computer readable media. Non-transitory computer readable media include any type of tangible storage media. Examples of non-transitory computer readable media include magnetic storage media (such as floppy disks, magnetic tapes, hard disk drives, etc.), optical magnetic storage media (e.g. magneto-optical disks), CD-ROM (Read Only Memory), CD-R, CD-R/W, and semiconductor memories (such as mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory), etc.). The program may be provided to a computer using any type of transitory computer readable media. Examples of transitory computer readable media include electric signals, optical signals, and electromagnetic waves. Transitory computer readable media can provide the program to a computer via a wired communication line, such as electric wires and optical fibers, or a wireless communication line.

　　While the present invention has been described above with reference to exemplary embodiments, the present invention is not limited to the above exemplary embodiments. The configuration and details of the present invention can be modified in various ways which can be understood by those skilled in the art within the scope of the invention.

100 GROUP FORMATION ENGINE
101 INPUT DATA
102 OUTPUT DATA
103 DEMAND AGGREGATION CRITERION (DAC) DETERMINATION
104 DEMAND AGGREGATION

Claims

　　An energy balancing group formation method comprising:
　　receiving historical demand profiles of a plurality of customers,
　　determining a demand aggregation criterion according to the received historical energy demand profiles,　　and
　　determining a demand aggregation strategy to form energy balancing groups based on the demand aggregation criterion, each energy balancing group including a group of customers determined by using the demand aggregation criterion from the plurality of customers.
　　The energy balancing group formation method according to Claim 1, wherein
　　the demand aggregation criterion of each customer is defined by using a first value expressed by using daily load factors of the historical demand over the predetermined period, and a second value expressed by a maximum value of standard deviations of periodic demands, and
　　each of the standard deviations of periodic demands is a standard deviation of demands for a particular period.
　　The energy balancing group formation method according to Claim 2, wherein
　　the demand aggregation criterion of each customer comprises a weighted combination of the a first and second value,
　　the first value is calculated by normalizing medians of the load factors in the predetermined period of the plurality of the customers, and
　　the second value is calculated by normalizing the maximum values of standard deviations of periodic demands of the plurality of the customers.
　　The energy balancing group formation method according to Claim 3, wherein　　the daily load factor is calculated by dividing daily average demand by daily peak demand.
　　The energy balancing group formation method according to Claim 2, wherein
　　the demand aggregation strategy is determined by using the demand aggregation criterion of each customer as a statistical observation, and
　　a statistical model including model parameters based on the observation is built by defining the model parameters with associated prior distributions.
　　The energy balancing group formation method according to Claim 5, wherein
　　the statistical model built based on the demand aggregation criterion values is sampled using Bayesian Markov Chain Monte Carlo sampling method to generate the posterior distributions of the model parameters and expected demand aggregation criterion values.
　　　　The energy balancing group formation method according to Claim 6, wherein
　　　　　　a single group of the customers is divided into two distinct groups of the customers using the posterior distributions of the statistical model parameters of the demand aggregation criterion values,
　　　　　　one of two distinct groups of the customers is a group including lower expected demand aggregation criterion values, and
　　　　　　the other of two distinct groups of the customers is a group including higher expected demand aggregation criterion values.
　　The energy balancing group formation method according to Claim 7, wherein
　　group partitioning based on the posterior distribution is iteratively conducted until a ratio of expected values of demand aggregation criterion between two divided distinct groups is below a predetermined threshold.