WO2018131174A1

WO2018131174A1 - Microgrid power management system and method of managing

Info

Publication number: WO2018131174A1
Application number: PCT/JP2017/001833
Authority: WO
Inventors: Shantanu Chakraborty; Alexander VIEHWEIDER
Original assignee: Nec Corporation
Priority date: 2017-01-13
Filing date: 2017-01-13
Publication date: 2018-07-19

Abstract

A power management system for a micro-grid having a connection to a main power grid, the power management system including: a receiving unit configured to receive main power grid information and micro-grid resource information; a prediction unit configured to predict, for a predetermined period of time: future micro-grid power consumption, micro-grid power resource availability, and a possibility of the main power grid becoming unavailable, the predictions being based, at least in part, on information received by the receiving unit; and a planning unit configured to determine and output an efficient operation plan of using a micro-grid resource and main grid power for the predetermined period of time, the efficient operation plan being based, at least in part, on the predictions of the prediction unit.

Description

DESCRIPTION

TITLE OF INVENTION

MICROGRID POWER MANAGEMENT SYSTEM AND METHOD OF MANAGING

TECHNICAL FIELD

The present invention relates to stochastic energy resource planning for localized energy sources (also known as a microgrid) which operate in parallel with a main power grid. More specifically, the present invention relates to optimal planning and operation of controllable energy resources such as energy storage devices, diesel generators, and various distributed energy resources so that such controllable energy resources may be efficiently used in conjunction with, or independently from, the main power grid.

BACKGROUND ART

In recent years, microgrids have been used to both reduce dependence on and enhance existing energy distribution infrastructure in which power is distributed to homes, facilities, and the like within a wide area over a main power grid. Since microgrids are able to utilize one or more energy resources along with a connection to the main power grid, it is often necessary to determine which resource should be utilized throughout any given time period based on a multitude of factors in order to optimize use of available energy in terms of cost, C02 reduction, etc. Some examples of microgrid management can be found in Patent Literatures 1 to 4.

[Citation List]

[Patent Literature]

Patent Literature 1 : Optimized dispatch planning of distributed resources in electrical power systems. United States Patent Publication No. US20040044442 A1

Patent Literature 2: Planning economic energy dispatch in electrical grid under uncertainty. United States Patent Publication No. US 20140025351 A1

Patent Literature 3: Optimized Energy Management System. United States Patent Publication No. US8903560 B2

Patent Literature 4: Optimized Security Constrained Unit Commitment Dispatch Using Linear Programming for Electricity Markets. United States Patent No. US20040181420 A1 DISCLOSURE OF INVENTION

Problem to be Solved by the Invention

One major problem that isn't considered in the above-mentioned patent literature is the rampant power-grid outages that occur in emerging countries (e.g. India) due to unreliable power grids. This is considered to be one of the most influential factors leading to large productivity gaps. Critical communication infrastructures such as Mobile Base Transceiver Stations (BTS) require uninterrupted and reliable power supply for operation. Because of this, industrial customers are often required to invest heavily in backup diesel engine generators (DGs) to mitigate blackouts, and the use of DGs can cost around 40-50% of their monthly electricity bill and significantly contribute to a staggering C02 footprint. The problem has become even more severe with the inclusion of dynamic pricing where the grid energy price dynamically changes over time depending on production, environment factors, etc.

Therefore, predictive optimal scheduling of a diesel engine generator (DG), battery energy storage, and the like under uncertainty with regard to load demand, renewable power generation, and grid price coupled with an unreliable grid structure is of great importance from the perspective of optimal microgrid operation. Hence, it is important for emerging countries to reduce overwhelming DG fuel consumption, and there is a need for a power management system and method for microgrids that can more accurately predict energy resource availability and provide optimal management thereof in order to reduce costs, conserve resources, reduce pollution, and the like. Means for Solving the Problem

In order to solve the above described problems, various configurations of a power management system and a power management method are disclosed.

As a first aspect of the present invention, a power management system for a micro-grid having a connection to a main power grid is provided including: a receiving unit configured to receive main power grid information and micro-grid resource information; a prediction unit configured to predict, for a predetermined period of time: future micro-grid power consumption, micro-grid power resource availability, and a possibility of the main power grid becoming unavailable, the predictions being based, at least in part, on information received by the receiving unit; and a planning unit configured to determine and output an efficient operation plan of using a micro-grid resource and main grid power for the predetermined period of time, the efficient operation plan being based, at least in part, on the predictions of the prediction unit.

As a second aspect of the present invention, the power management system of the first aspect further includes: a micro-grid controller configured to control a micro-grid power resource in accordance with the efficient operation plan output by the planning unit, thereby improving efficiency of use of the micro-grid power resource. As a third aspect of the present invention, in the power management system of the first aspect, the main power grid information is at least one of: grid connection status and grid price; and the micro-grid resource information is at least one of: resource type, resource price, resource status, grid consumption power, renewable resource power availability, fossil fuel power availability, and battery power availability.

As a fourth aspect of the present invention, in the power management system of the first aspect, the efficient operation plan is determined in accordance with a pre-determined optimization objective.

As a fifth aspect of the present invention, in the power management system of the fourth aspect, the pre-determined optimization objective is at least one of: fossil fuel use reduction, total operation cost reduction, CO2 emission reduction, renewable resource use maximization, and micro-grid resource use maximization.

As a sixth aspect of the present invention, a non-transitory computer readable storage medium is provided which contains instructions for causing a computer to function as the power management system of any of the first to fifth aspects.

As a seventh aspect of the present invention, a computer-implemented power management method for a micro-grid having a connection to a main power grid is provided which includes steps of: receiving main power grid information and micro-grid resource information; predicting, for a predetermined period of time: future micro-grid power consumption, micro-grid power resource availability, and a possibility of the main power grid becoming unavailable, the predictions being based, at least in part, on information received in the receiving step; determining an efficient operation plan of using a micro-grid resource and main grid power for the predetermined period of time, the efficient operation plan being based, at least in part, on the predictions of the prediction unit; and outputting the efficient operation plan.

As an eighth aspect of the present invention, a non-transitory computer readable storage medium is provided which contains instructions for causing a computer to perform the computer-implemented power management method of the seventh aspect.

Advantageous Effects of the Invention

The present invention provides an adaptive energy resource planning system and power management method that operates with a microgrid which is connected to an unreliable main grid while considering possible uncertainties regarding different prediction signals. BRIEF DESCRIPTION OF DRAWINGS

Fig. 1 is a block diagram schematically illustrating an outline of an embodiment of present invention (Adaptive Resource Planning Unit, ARPU) with basic input and output blocks.

Fig. 2 is a block diagram of the prediction units of the embodiment of the power management system.

Fig. 3 is a block diagram of an uncertainty sets generator/operational setting adaptation module of the embodiment.

Fig. 4 is a flow chart of the process of calculating the uncertainty set for prediction errors of the embodiment.

Fig. 5 is flow chart of the process for determining an uncertainty set for the preceding period variability of the embodiment.

Fig. 6 is a flow chart of stochastic scenario generation for prediction signals in the embodiment.

Fig. 7 is a flow chart of stochastic scenario generation for grid status prediction in the embodiment.

Fig. 8 is a flow chart of an adaptation of the prediction window module of the embodiment.

Fig. 9 is a flow chart of an adaptation of the application window module of the embodiment.

Fig. 10 is a flow chart of the optimization engine that outputs an optimal control sequence of controllable energy resources. Fig. 11 shows three example scenarios of a continuous prediction signal (load demand prediction).

Fig. 12 shows a process of generating two scenarios from a probabilistic grid status prediction by varying a threshold value.

Fig. 13 illustrates a block schematic of a general purpose computer suitable for applying the computer implemented power management method of the present invention.

EMBODIMENTS FOR CARRYING OUT THE INVENTION

An exemplary embodiment 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 repeated descriptions are omitted as needed.

The following embodiment is described as a specific example using the concept of the present invention to enable a person skilled in the art to make and use the invention. However, the present invention should not be considered as being limited to these embodiments and various configurations and modifications are possible without deviating from the spirit or scope of the invention.

Exemplary Embodiment

As an embodiment of the present invention, a power management system (also referred to as an adaptive resource planning unit, i.e. , ARPU) will be described. Fig. 1 is a diagram that schematically illustrates an outline of the ARPU 1 01 . The ARPU 101 receives load demand data, PV power generation, and grid price. Moreover, the ARPU additionally takes, as input, controllable resources statuses. The data can be, for example, data supplied by a memory unit, a storage device, the I nternet, a measurement device or the like. The controllable resources in this embodiment are battery storage units and diesel fuel based generators; however, the controllable resources are not lim ited thereto.

I n Fig. 1 , the inputs to the power management system are shown in the receiving unit 1 02. The input to the system (at the receiving unit 1 02) includes the current measurements of the controllable resource status (e.g. , battery state of charge SOC), energy demand , energy price, renewable power generation (e.g. , PV Power), and grid status (i .e. , outage or no-outage). The ARPU 1 01 has a prediction unit 1 03 (shown in Fig. 2) which utilizes different predictive models that provide predictions of associated signals for a particular prediction window w. The predictive models of this embodiment are: load demand prediction 201 , renewable power prediction 202, grid price prediction 203, and grid status prediction 204. Among these, the grid status prediction 204 is a classifier that provides the binary grid status prediction with an associated probability for each of the binary classes. The other prediction models are regression engines that predict continuous prediction signals (e.g. , the amount of load demand for the next w periods in kilowatts (kW), the amount of power generated by renewable generator for the next w periods in kW, and the dynamic grid price for the next w periods in $/kWh). The prediction unit 1 03 initially performs cleaning (filtering) of the data by aligning all data in the same granularity where the data are collected from the measurement database 1 05 which contains the historical measurements of load demand , PV power, gird energy price and grid status. Next, an optional step of data preprocessing is performed utilizing historical training data. Finally, the prediction models are created for appropriate pred iction signals using the associated current measurement data from the measurement database 1 05.

Hereinbelow, specific steps to be performed in the load demand prediction model 201 , the renewable power prediction model 202, and the grid price prediction model 203 will be described . First, the training data (optionally pre-processed) is received, and a grid-search process and a cross-validation process are performed over a repeatedly selected random subset of a training set and testing sets by applying a regression method (e.g. , support vector regression). The grid-search process searches for the optimal set of hyper-parameters over a randomly selected subset of the training set, while the cross-validation process validates the hyper-parameters with the remainder of the training set. Next, hyper-parameters of the regression method as the outcome of the grid-search and cross-validation are selected, and predictive models are built. Thereafter, the predictive models of each of the load demand prediction model 201 , renewable power prediction model 202, grid price prediction model 203 by applying associated historical and current measurement data are applied in order to produce a prediction as output which is stored in a prediction database 104. An example of the prediction output for a 24-hour period is illustrated in Fig. 11 .

Next, a description of the steps to be performed in the grid status prediction model 204 will be given. First, training data (pre-processed by One-Hot encoding) is received, and grid-search and cross-validation over a repeatedly selected random subset of the training set and testing sets by applying a classification method (e.g., random forest classifier) is performed. The binary classes for the grid status are ON (1 ) and OFF (0) indicating whether or not a connection to the main power grid is available. Next, hyper-parameters of the classifier that minimizes the training error are selected, and the predictive models are built. Thereafter, the predictive models that provide the grid status prediction as well as the probability of each status are applied in order to produce a prediction as output, and this output is stored in the prediction database 104. An example is shown below in Table 1 . Table 1

The ARPU 101 of this embodiment further includes a first planning unit 106 (an uncertainty sets generation/operation setting adaption module) and a second planning unit 107 (an optimization engine) which together may be a single planning unit although they are represented separately in this embodiment for ease of understanding. The first planning unit 106 will now be described with reference to Fig. 3.

Fig. 3 shows a detailed block diagram of the first planning unit 106 and the continuous prediction signals block contains the uncertainty set generation and stochastic scenario generation for the continuous prediction signals, i.e. the prediction signals that are generated by the regression models (load prediction, renewable power generation prediction, and grid price prediction). On the other hand, the binary prediction signal block contains the grid status prediction signal which is a binary signal, and the adaptation module contains a prediction window module and an application window module. In block 301, the uncertainty set for prediction error with optimization mode is determined which is shown in detail in Fig. 4. For each of the continuous prediction signals, the error between prediction data and measurement data for a predefined historical horizon are calculated for a predetermined period (from t— N_s to t-1), where the current time is t, and N_s is the predefined horizon. The error deviation is calculated in step S1-301 by the equation:

E ·■= {Predi - esi)|≡¾

where Predi and Mes_t are the prediction and measurement of the i-th period, respectively.

In step S2-301, the kernel density estimation of the error

(KDE-E) is calculated, and in step S3-301, it is checked whether the KDE-E is in a list of predefined probability distribution functions (PDF). The list of PDFs contains known probability distributions such as Gaussian, Exponential, Laplace, Beta, or the like. If step S3-301 yields 'YES', the process moves to step S4-301 where the optimization mode is set to 'expectation minimization' since, the KDE-E is known. Thereafter, uncertainty sets of the error, I/f using KDE-E parameters are generated in step S5-301 . On the other hand, if step S3-301 yields 'NO', the optimization mode is set to 'worst cost minimization', since the KDE-E is not known . In such a case, uncertainty sets of the error, 1/f using a random distribution within upper/lower bounds of the error, are generated .

Here, an explanation of how 'the uncertainty set for preceding period variability' in block 302 of Fig. 3 is determined will be given with reference to the process flow of Fig. 5. For each of the continuous prediction signals, the difference between two consecutive measurement data over a predefined historical horizon are calculated for a predetermined period (from t - N_s to t - 1) to determine the preceding period variability (PPV) in step S1 -302, where the current time is t and N_s is the predefined horizon. The preceding period variability is given by the following equation .

Next, the kernel density estimation of the PPV (KDE-P) is calculated in step S2-302. In step S3-301 , it is checked whether the KDE-P is in the list of predefined probability distribution functions (PDF) containing known PDFs such as Gaussian, Exponential , Laplace, Beta, or the like, and if step S3-302 yields 'YES', uncertainty sets of the PPV, X/f using KDE-P parameters are generated . However, if S3-302 yields 'NO', uncertainty sets of the error, 11_$ using a random distribution within upper/lower bounds of the PPV, are generated.

Here, an explanation of how the 'stochastic scenarios generator for prediction signals' in block 303 of Fig. 3 are determined will be given with reference to the process flow of Fig. 6. The stochastic scenarios generator for prediction signals are determined for each of the continuous prediction signals for a particular prediction window, w. In step S1-303, the uncertainty sets Ή| and IZ| are collected, and the stochastic scenarios for each of the prediction signals are generated by applying a bivariate distribution combining the uncertainty sets 1_f and 11_$. The bivariate distribution creates a random sample space combining two probability distribution functions (in this case, the aforementioned uncertainty sets). Fig. 11 shows an example of 3 load demand prediction scenarios considering the generated uncertainty sets.

Regarding block 304 of Fig. 3, the 'stochastic scenarios generator for grid status prediction' is generated as shown in Fig. 7. A number of unique stochastic scenarios of a grid status prediction are generated in step S1-304 by varying the threshold of the probability of the grid status being ON. The grid status prediction is calculated using the equation,

where GSS^ is a grid status scenario for period i and threshold T. The probability Ρ (5ϋ = 1) represents the probability of the grid status being ON at period i. Fig . 12 shows, as an example, a process of generating two scenarios where the threshold , T ε {0.4,0.5}.

The adaptation module in Fig. 3 will now be described . The adaption module has a prediction window adaptation module 305 and an application window module 306. A flow chart for the prediction window module 305 is shown in Fig. 8.

First, in step S 1 -305 of Fig. 8, the historical prediction error for each of the continuous prediction signals is calculated , followed by step S2-305 which contains the following sub-steps (not shown):

i. Determine the periodical ( i -th) standard deviation

(EStd ) °f ^{tn e} prediction error for each of the continuous prediction signals, s E CS over a set CS.

ii. Smooth the EStdi by taking the moving average over a movement window mw using the following equation :

k=i+mw

SEStd^ =— Y EStd

k=i

iii. Determine a period, t_s for which a distance in consecutive SEStd_s ^l cross over a predefined threshold Th_s in accordance with the relation:

In step S3-305, the prediction window 305 is adapted by defining the adaptive prediction window, w as:

The other part of the adaptation module of Fig. 3 is an application window module 306. The flow chart of the application window module 306 is shown in Fig. 9. In step S1 -306, the error between the measured controllable resource status (e.g., the historical SOC of the battery) from the measurement database 105 and optimized controllable resource status (e.g. the resultant SOC from the second planning unit 107) from resource status optimization database 108 is calculated. Next, in step S2-306, the maximum k≤ kmax ^{a n cl} k≤ w for which the difference between measured and optimized status stays below a particular threshold is determined, where k_max is the upper bound of the application window.

Hereinafter, operations of the second planning unit 107 (optimization engine) will be described with reference to the flow chart of Fig. 10.

First, in step S1 -107, the scenarios of the continuous prediction signals are received from the first planning unit 106, and the scenario space is reduced. Typically, the scenario space grows exponentially with the number of prediction signals. Therefore, the scenario space needs to be reduced. The reduction process is conducted by choosing a sub-space of the scenario set where the member scenarios are closer to the original prediction signals. In the step S2-107, the scenarios of different prediction signals are merged and if the optimization mode was previously set to 'expectation minimization', the scenario probability distribution is determined in accordance with the list of PDFs in step S3-1 07 (where the scenarios of the reduced scenario space are assigned with weights proportional to their closeness to the original prediction signals) and the objective function (i.e. , the function used to achieve optimization in terms of a specific objective such as diesel fuel use reduction , total operation cost reduction , CO2 emission reduction, renewable resource use maximization , micro-grid resource use maximization, or the like) is set accordingly. The objective function of 'expectation minimization' may contain a cost function (which can be single or multi-objective function that combines multiple objectives) whose expected values needed to be minimized by deciding on the controllable resources. Otherwise, if the optimization mode was previously set to 'worst cost minimization', the process flow goes to step S5-1 07 and the objective function is set accordingly. That is, the objective function of 'worst cost minimization' performs an optimization process where at first, the worst cost scenario is determined and finally, the worst cost scenario is minimized by deciding on the controllable resources.

Next, in step S6-1 07, the objective function set in either S4-1 07 or S5-1 07 is merged with the grid status prediction scenarios with a uniform probability distribution of the prediction window using control resource statuses while satisfying control resource constraints. In step S7-107, it is determined whether or not the objective function is a convex function. If so, the objective function is transformed into a mixed integer linear programming problem (step S8-107) and the transformed objective function is solved by a branch-and-cut solver (step S9-107). Otherwise, if the objective function is not a convex function in step S7-107, a meta-heuristic algorithm is used to solve the objective function (S10-107). The output of the flow is an efficient (in terms of the objective) operation control sequence (operation plan) for next 'application window' periods (i.e. , a predetermined period of time), to be carried out by a resource control unit or the like.

Other embodiments

It is important to note that the present invention is not limited to the above exemplary embodiment and can be modified as appropriate without departing from the scope of the invention.

In the above exemplary embodiment, the present invention is described as system (i.e., an adaptive resource planning unit) configured by a number of dedicated electronic units that perform specific functions, but the present invention is not limited to this. The present invention may be implemented by causing a CPU (Central Processing Unit) to execute a computer program in which the functions of each dedicated electronic unit are performed thereby. 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 also 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 embodiment, 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 to be within the scope of the invention.

Claims

1 . A power management system for a micro-grid having a connection to a main power grid, the power management system comprising:

a receiving unit configured to receive main power grid information and micro-grid resource information;

a prediction unit configured to predict, for a predetermined period of time:

future micro-grid power consumption,

micro-grid power resource availability, and

a possibility of the main power grid becoming unavailable, the predictions being based, at least in part, on information received by the receiving unit; and

a planning unit configured to determine and output an efficient operation plan of using a micro-grid resource and main grid power for the predetermined period of time, the efficient operation plan being based, at least in part, on the predictions of the prediction unit.

2. The power management system of Claim 1 , further comprising: a micro-grid controller configured to control a micro-grid power resource in accordance with the efficient operation plan output by the planning unit, thereby improving efficiency of use of the micro-grid power resource.

3. The power management system of Claim 1 , wherein

the main power grid information is at least one of the group consisting of: grid connection status and grid price; and

the micro-grid resource information is at least one of the group consisting of: resource type, resource price, resource status, grid consumption power, renewable resource power availability, fossil fuel power availability, and battery power availability.

4. The power management system of Claim 1 , wherein

the efficient operation plan is determined in accordance with a pre-determined optimization objective.

5. The power management system of Claim 4, wherein

the pre-determined optimization objective is at least one selected from the group consisting of: fossil fuel use reduction, total operation cost reduction, C02 emission reduction, renewable resource use maximization, and micro-grid resource use maximization.

6. A non-transitory computer readable storage medium containing instructions for causing a computer to function as the power management system of any of Claims 1 to 5.

7. A computer-implemented power management method for a micro-grid having a connection to a main power grid, the power management method comprising the steps of:

receiving main power grid information and micro-grid resource information;

predicting, for a predetermined period of time:

future micro-grid power consumption,

micro-grid power resource availability, and

a possibility of the main power grid becoming unavailable, the predictions being based, at least in part, on information received in the receiving step;

determining an efficient operation plan of using a micro-grid resource and main grid power for the predetermined period of time, the efficient operation plan being based, at least in part, on the predictions of the prediction unit; and

outputting the efficient operation plan.

8. A non-transitory computer readable storage medium containing instructions for causing a computer to perform the computer-implemented power management method of Claim 7.