WO2009059370A1

WO2009059370A1 - Methods and apparatus for resource management

Info

Publication number: WO2009059370A1
Application number: PCT/AU2008/001650
Authority: WO
Inventors: Rongxin Li; Geoffrey T. Poulton
Original assignee: Commonwealth Scientific And Industrial Research Organisation
Priority date: 2007-11-08
Filing date: 2008-11-06
Publication date: 2009-05-14

Abstract

A method of controlling resource utilisation over a predetermined time interval in a system (100) that includes a plurality of facilities (102). Each facility (102) has a control input which influences a resource utilisation of the facility. For each facility, the method includes first identifying (602) a sequence of values of the control inputs over the time interval that results in a minimal resource utilisation of the facility, by constructing a data structure representing possible sequences of events occurring in the system over the predetermined time interval. Subsequently, the identified sequence of values is modified (604) to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints. Finally, the operation of the facility is controlled (606) over the time interval in accordance with the modified sequence of control values. The invention is particularly well-suited to the management of energy consuming devices or appliances in an energy distribution network. Preferred embodiments of the method utilise a tree structure to encode sequences of facility states, and potential transitions therebetween, enabling the use of efficient algorithms, such as shortest path algorithms, for identifying sequences corresponding with minimum resource utilisation, and for modifying control sequences in order to trade-off resource utilisation against improved satisfaction of other system, facility and user objectives.

Description

METHODS AND APPARATUS FOR RESOURCE MANAGEMENT FIELD OF THE INVENTION

The present invention relates generally to the management and control of resource-consuming facilities. The invention has particular application to improving or optimising energy resource utilisation by multiple energy resource consuming devices, for example during periods of supply restriction. While the invention is described with particular reference to the example of electricity distribution networks (Ze electrical energy consumption), it is applicable to a variety of other types of resources having similar characteristics of utilisation, control and management.

BACKGROUND OF THE INVENTION

Typical electricity networks, like many other systems for the distribution and management of resources, are complex physical entities. Electricity distribution networks consist of generators, transmission networks, transformer stations and distribution networks. Superimposed upon this physical structure is a market structure comprising companies involved in generation, transmission and distribution. In this respect, the modern energy supply industry differs significantly from the more traditional structure.

In the past, the energy supply industry was structured as a regulated monopoly. Under this model, generators were typically "dispatched" (Ze mobilised to deliver power into the grid) by the monopoly supplier in response to the load arising within the monopoly service area. Accordingly, the monopoly operator has complete control over the energy resources within a specified service area. However, in accordance with the restructuring occurring in many modern markets, the dispatch function may be transferred to an independent operator tasked with operating an energy market over the transmission system through which all generators (which may have a number of competitive owners) deliver energy to end consumers.

The independent energy market operator may, for example, operate a wholesale "spot market" in electricity, wherein supply and demand are instantaneously matched in real time through a centrally co-ordinated dispatch process. In accordance with such a scheme, generators offer to supply the market with specific amounts of electricity at particular prices. Such offers may be submitted at regular intervals, for example every five minutes of every day. From all offers submitted, the market operator determines the generators required to produce electricity based on a principle of meeting prevailing demand in the most cost-efficient way. The market operator then dispatches the selected generators into production.

As noted above, in the modern system wholesale electricity prices may be available to industrial users and retail suppliers for each five minute interval. Furthermore, longer-term forecast prices may be made available, for example over the coming 24 hours, and with varying accuracy. This information potentially enables users to regulate their demand in accordance with expected availability, capacity and pricing. It is therefore desirable to implement systems that are able to optimise energy utilisation in order to provide additional improvements in efficiency, reductions in cost, and other benefits potentially available under the modern market structure. One approach is to install a network of autonomous agents on the physical distribution network. The network may include two layers of energy distribution agents, and one layer of consumption manager agents. This provides a tri-level architecture, which may be outlined as follows, in a bottom up fashion. On the consumption level, the energy usage of electrical devices or appliances may be controlled, to a varying degree, by the device manager agents that switch on and off the power supply to the device. A cluster consisting of a plurality of such agents, grouped together according to physical or market driven factors, may be managed by a group optimising agent (GOA) on the lower distribution level. The GOA, in turn, receives a group quota from an upper level distribution agent, which is sometimes referred to as a broker agent.

Accordingly, the roles of a GOA are firstly to optimise energy usage on a per-device basis, and additionally to co-ordinate the consumption between the devices within the group. A method, and system architecture, for providing improved management of distributed energy resources, optimisation of energy consumption, and communications between agents in such a system, is described in International PCT application no. PCT/AU2007/001089, However, there remains an ongoing need for improved methods, systems and apparatus for managing multiple resource-consuming devices subject to constraints, such as system supply caps, while simultaneously seeking to optimise performance with respect to additional local, or system, goals and objectives.

It is accordingly an object of the present invention to address the aforementioned need. SUMMARY OF THE INVENTION

In one aspect, the present invention provides a method of controlling resource utilisation over a predetermined time interval in a system including a plurality of facilities, each of which has a control input which influences a resource utilisation of the facility, the method including the steps of, for each facility: constructing a data structure representing possible sequences of events occurring in the system over the predetermined time interval; identifying within the data structure a sequence of values of the control input over the time interval that results in a minimum resource utilisation of the facility; modifying the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and controlling the operation of the facility over the time interval in accordance with the modified sequence of control values. Advantageously, the method of the present invention may be implemented by a group optimising agent (GOA) of an energy supply network. The GOA may receive resource utilisation constraints, such as a time sequence of supply capacity limits covering a forthcoming predetermined time interval (for example, in five minute intervals) which may be utilised in the method to control resource utilisation of a group of energy consuming facilities over the time interval. While conventional approaches to optimising behaviour based upon such time sequence inputs utilise corresponding time series analysis techniques, methods embodying the present invention advantageously adopt an entirely different approach. In particular, embodiments of the inventive method may utilise suitable data structures, such as a tree structure, to represent possible sequences of events (eg control input values) over the predetermined time interval. Algorithmic approaches for identifying "shortest paths" within such data structures may then be used to identifying sequences of events resulting in minimum resource utilisation. The identified minimum sequence may then be modified to identify alternative sequences of events which advantageously trade-off resource utilisation against improved performance in relation to other specified system and/or facility objectives, while remaining within the specified constraints.

The inventors have found that this unique approach to the control of resource utilisation, at least when advantageously applied to electrical energy consuming facilities, is able to provide excellent performance while requiring reasonable, and achievable, processing resources for implementation. Furthermore, it is reasonably expected that similar principles may be applied to a variety of other types of resources. For example, it is readily foreseeable that embodiments of the invention may be applied to other energy resources, such as gas distribution networks, and also to other utilities, such as water supply. The invention may also have application to areas such as traffic flow management, in which the resources may be identified with, for example, roads, bridges, tunnels and the like, all of which have limited capacity per unit time, and wherein the controllable facilities include traffic lights, toll gates, and so forth. The invention may also have application in communications networks, wherein the available resources include transmission capacity in communications links, as well as storage or buffer capacity in switching nodes.

In another aspect, the invention provides an energy distribution system including a plurality of energy consuming facilities, each of which has a control input which influences a resource utilisation of the facility, and at least one controller operatively associated with a corresponding group of facilities, wherein the controller is adapted to, for each facility within the associated group: identify a sequence of values of the control input over a predetermined time interval that results in a minimum resource utilisation of the facility; modify the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and control the operation of the facility over the time interval in accordance with the modified sequence of control values. In yet another aspect, the invention provides a controller for use in an energy distribution system which includes a plurality of energy consuming facilities, each of which has a control input which influences a resource utilisation of the facility, the controller being operatively associated with a corresponding group of facilities, and being adapted to, for each facility in the group: identify a sequence of values of the control input over a predetermined time interval that results in a minimum resource utilisation of the facility; modify the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and control the operation of the facility over the time interval in accordance with the modified sequence of control values.

Preferably, the controller includes: at least one microprocessor; at least one input/output interface device, such as a network interface, for providing control inputs to the associated group of facilities; and at least one memory/storage device operatively associated with the microprocessor, wherein the memory/storage device includes executable instruction code which, when executed by the microprocessor, causes the controller to implement the steps of, for each facility: identifying a sequence of values of the control input over the time interval that results in a minimum resource utilisation of the facility; modifying the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and controlling the operation of the facility over the time interval in accordance with the modified sequence of control values.

In preferred embodiments, the controller includes and/or implements a GOA.

The controller is preferably further interfaced to a network, via which the controller receives operating information, such as system constraints provided by a broker agent, and via which the controller is also able to transmit information, such as present and forecast status information, which may be utilised by a broker agent, and/or other agents within the system, in their own operations.

One system architecture suitable for providing such communications between agents is described in International PCT application no. PCT/AU2007/001089, however it will be appreciated that other means of communicating information within a system, such as an energy distribution system, may also be utilised.

According to preferred embodiments, the step of modifying the identified sequence of values includes an iterative process of trading off resource utilisation against satisfaction of system and/or facility objectives within the specified resource utilisation constraints. For example, the facilities may have preferred patterns of operation, and the process may include iterating from the identified sequence of values of the control input corresponding with minimum resource utilisation towards said preferred patterns, to the maximum extent allowed by the specified resource utilisation constraints. Alternatively, the process may include iterating from the preferred patterns of operation towards the identified sequence of values of the control input corresponding with minimum resource utilisation to the minimum extent required by the utilisation constraints.

In accordance with preferred embodiments of the invention, each facility has an operating state which is selected from at least two available operating states. The operating state of each facility may switch between the available operating states in accordance with the control input, and at least one internal condition of the facility. The resource utilisation of each facility typically depends at least upon the current operating state of the facility. Preferred methods according to the invention further include providing a model of each facility which enables the resource utilisation of the facility to be estimated over the predetermined time interval as a function at least of an input sequence of control values. Typically, the resource utilisation of the facility over the time interval is also dependent upon the initial state of the facility, and the initial value of any internal conditions. Accordingly, these parameters are also preferably included within the facility models.

By way of example, a refrigeration unit may simply be modelled by an operating state which is either "on" or "off". An internal condition of the refrigeration unit is the interior temperature, which may be allowed to vary between acceptable upper and lower temperature limits. A control input to the refrigerator is its power supply, ie disconnecting the refrigeration unit from the power supply under external control effectively forces the unit into the "off state. A mathematical model of the refrigeration unit, based upon current temperature, operating state, and control input values over a predetermined time interval, enables the energy consumption of the unit, and the corresponding variations in internal temperature, to be modelled over the time interval.

In a preferred embodiment, an energy resource utilisation constraint includes a capacity constraint, ie a restricted supply of available energy over the predetermined time interval. This constraint may vary over the time interval, and for example may consist of a series of energy caps spanning each one of a plurality of sub intervals, for example five minute sub intervals within a 30 minute total interval. Additional constraints may include permissibility constraints. For example, in the case of the refrigeration unit described above, a sequence of control inputs should not be selected which results in the internal condition of the refrigeration unit moving outside the acceptable upper and lower temperature limits. Other examples of permissibility constraints include user constraints, such as cost constraints.

In accordance with preferred embodiments, specified objectives might include operating optimality of facilities. For example, a particular type of facility might have a preferred "free running" operating mode, which would be followed in the absence of external control. On the assumption that such a free running mode is locally optimal, and/or otherwise desirable, the formulation of a corresponding objective may be, for example, to minimise the total restriction of energy supply to the facility, to minimise the frequency of interference with the free running mode of the facility, to control the facility in order to optimise a final internal state at the end of the predetermined time period, or to provide for more desirable future performance of the facility over a plurality of time intervals. Yet other objectives may be defined, such as minimising operating costs.

In a further aspect, the present invention provides a method of controlling resource utilisation over a predetermined time interval in a system including a plurality of facilities, each of which has a control input which influences a resource utilisation of the facility, the method including the steps of, for each facility: constructing a data structure in the form of a tree structure which encodes sequences of facility states in nodes thereof, and transitions between said states in branches thereof, whereby the tree structure represents possible sequences of events occurring in the system over the predetermined time interval, and wherein a cost at each said node, and one for each said transition, are assigned according to a corresponding facility resource utilisation; identifying within the tree structure a sequence of values of the control input over the time interval that results in a minimum resource utilisation of the facility; modifying the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and controlling the operation of the facility over the time interval in accordance with the modified sequence of control values.

In a particularly preferred embodiment, the sequence of values of the control input resulting in minimum resource utilisation, and modified sequences of values, are determined using a shortest path algorithm over the tree structure. Further preferred features and advantages of the present invention will be apparent to those skilled in the art from the following description of preferred embodiments of the invention, which should not be considered to be limiting of the scope of the invention as defined in any of the preceding comments, or in the claims appended hereto. BRIEF DESCRIPTION OF THE DRAWINGS

Preferred embodiments of the invention are described with reference to the accompanying drawings, in which like reference numerals refer to like features, and wherein:

Figure 1 illustrates a hierarchical structure of a network of autonomous agents in accordance with a preferred embodiment of the invention;

Figure 2 is a schematic diagram illustrating an exemplary microprocessor based apparatus for implementing a GOA according to a preferred embodiment of the invention; Figure 3 is a graph illustrating an energy consumption limit for a time interval T in accordance with an embodiment of the invention;

Figure 4 is a state transition diagram of a two state device controlled by an external agent according to an embodiment of the invention; Figure 5 is a graph illustrating a device model for a refrigeration unit according to an embodiment of the invention;

Figure 6 is a flow chart illustrating a method of controlling resource utilisation in accordance with a preferred embodiment of the invention;

Figure 7 is an illustration of a binary tree representing state transitions and switching sequences in accordance with an embodiment of the invention;

Figure 8 is a flow chart illustrating, consumption minimisation according to a preferred embodiment of the invention;

Figures 9A and 9B show a flow chart illustrating optimisation according to a preferred embodiment of the invention; Figure 10 is a graph illustrating a cost-effective trade off between usage and intervention according to an embodiment of the invention;

Figures 11A and 11 B are graphical illustrations of short-term optimisation according to an embodiment of the invention;

Figure 12 is a graphical illustration of optimisation of device control according to an embodiment of the invention; and

Figure 13 is a graphical illustration of an alternative optimisation of device control according to an embodiment of the invention. DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

Figure 1 illustrates a hierarchical structure 100 of a network of autonomous agents 104, 106, 108, according to which preferred embodiments of the invention may be implemented. An exemplary electricity network includes a plurality of electrical devices or appliances 102, each of which may be at least partially controlled by an associated device manager agent 104. A cluster, or group, of such device manager agents 104 are managed by a group optimising agent (GOA) 106.. The GOA 106, in turn, receives a group quota from the upper level distribution agent, or broker agent, 108. As will be appreciated, communications between the various agents may be achieved using various information communication technologies, including, but not limited to, internet or other data network technologies. Similarly, control of the electrical devices 102 by the associated manager agents 104 may be via similar communications technologies, via direct internal or external electrical interfaces, and/or by other means that will be apparent to persons skilled in the art. Figure 2 is a schematic diagram illustrating a number of key components of an exemplary microprocessor based apparatus 200 which may be used to implement a GOA 106. It will be appreciated, however, that Figure 2 does not show all peripherals, interfaces and components of the microprocessor system 200, which are well known in the art but which are not relevant to the present discussion. The apparatus 200 includes a microprocessor 202, interfaced in a conventional manner to a network interface 204. The network interface 204 provides access to a data network, such as a wired (eg Ethernet) or wireless (eg WLAN) network utilising suitable communications protocols, for example internet protocols, for operation. The processor 202 is also interfaced to one or more memory or storage devices 206. The memory or storage device 206 relevantly contains program instructions for execution by the processor 202, for carrying out various operations of the apparatus 200, including those related to the operation to the GOA 104. As will be appreciated, the memory or storage device 206 will also contain program instructions for execution by the processor 202 for performing a variety of other supporting functions, including various operating system functions of the apparatus 200.

While the exemplary apparatus 200 includes only a single communications interface 204, it will be appreciated that multiple interfaces may be provided for communicating with a broker agent 108, and various device manager agents 104. The roles of a GOA 106 are firstly to optimise energy usage on a per- device basis, and further to coordinate the consumption between the devices 102 within the group. As noted above, the GOA 106 receives a group quota from a broker agent 108. According to exemplary embodiments, the group quota is an upper limit of collective energy consumption over a subsequent time period T. The upper limit may be a constant directly defining maximum power consumption for the group (ie an energy cap) for the entire period, or a time-varying but piecewise constant function, or some alternative information that enables such a function to be derived. For the purposes of preferred embodiments described herein, a function L(t) is assumed for the consumption limit having the following form:

where: u(t) = { 1 0 ^X', o °th ^<e *rw ^≤is ^δte

In the foregoing equation, the piecewise limits Ij are fixed constants, and J is a set which fully and uniformly parameterises the continuous interval between time to and time t₀ + T.

Figure 3 is a graph 300 which illustrates an exemplary energy consumption limit in accordance with the above equation. A piecewise constant limiting function L(t) commences at time to 302 and terminates at the end of a period of duration T, 304. The overall period is divided into intervals of duration δ_t, 306, over each of which is defined a constant energy cap 308.

It should be understood that each of the piecewise constant energy cap values I_j may be an estimate subject to subsequent adjustments, ie the function may represent an indicative forecast of energy availability that helps to promote longer-term optimality.

In order to perform its functions, the GOA 106 is required to model the behaviour of each device 102 within its group, in order to predict energy requirements under the control of the device manager agents 104. For this purpose, in accordance with preferred embodiments the GOA 106 implements appropriate mathematical models of the various devices. By way of example, a simple model is described herein which is applicable to a device having two states (ie binary), simply defined as "on" and "off¹. When such a device is in the "on" state, it consumes a fixed quantity of energy, whereas in the "off' state it consumes no energy. The actual state of the device 102 is determined by its own operating rules, as well as external input from a corresponding device manager agent 104. Again, in a simple exemplary case, the external control consists of enabling or disabling a power supply to the device. As will be appreciated, in these circumstances the device manager agent 104 is able effectively to force a corresponding controlled device 102 into the "off' state, regardless of its preferred operating rules. The operation of the device 102 in the absence of external control by the device manager agent 104 is herein termed "free-running operation". Figure 4 is a state transition diagram 400 of such a two state device controlled by an external agent. The transition diagram 400 shows the "off state 402 of the device, and the "on" state 404. An internal state parameter p determines switching between the states 402, 404, which in the absence of external control will occur in accordance with free-running operation of the device. An external control signal c is able to force the device into the "off' state 402. As will be readily appreciated, more complex devices may be modelled using state transition diagrams having a larger number of states, with one or more control inputs having two or more possible values. It should therefore be understood that the simple examples provided herein, of binary devices having binary control inputs, is exemplary only, and not limiting of the overall scope of embodiments of the invention.

For further specificity in the examples which follow, a simple device model enabling state changes to be predicted using a piecewise exponential temporal function pι(t) is assumed, again without limitation to the overall scope of the invention. A graph 500 illustrating the operation of such a model is shown in Figure 5. An example of a device which may exhibit this behaviour is a refrigeration unit. The device has an internal condition which varies with time, and corresponds with the state parameter p of the transition diagram 400. In the case of a refrigeration unit, the internal condition is temperature. Referring again to Figure 5, the temperature (or analogous internal condition) is constrained to lie between a lower limit 502 and an upper limit 504. When the unit is switched off, the temperature rises in accordance with an exponential characteristic. Similarly, when the device is switched on, the temperature decreases in accordance with a different exponential characteristic. The solid line 506 in the graph 500 represents the free-running operation of the refrigeration unit, whereas the dashed line 508 represents an alternative operating characteristic in which an intervention occurs, via the control signal c forcing the unit to switch off at a particular time instant t_c 510. Following this intervention, the illustrated characteristic revers to free-running operation.

The function of the GOA 106 is to determine a "switching plan" for the control inputs of the various devices 102 under its supervision. By modelling the operation of the various devices, the GOA 106 seeks to find a sequence of control inputs for each which will satisfy the energy consumption constraint received from the broker agent 108, while simultaneously seeking to optimise the operation of each individual device 102 with respect to an objective function.

Embodiments of the present invention are particularly directed to the solution of this general problem.

To formalise the problem, in the case of the exemplary binary device (eg refrigeration unit), the primary goal of the GOA 106 is to devise switching plans qι(t), wherein each such plan is a binary function of time, such that:

i_ei Jto+jδt In the above equation, the parameter dj is the required power consumption of a device identified by an index i when it is in the "on" state. It is assumed that each device consumes no energy when the "off' state. Additionally, the switching plans must also be allowable by the hardware specifications of each device 102, and should satisfy further user requirements, and posses, to the greatest extent possible, desired characteristics (such as frequency and/or phase characteristics). The terms "optimisation" and "constrained optimisation" are used herein to describe processes having the goal of meeting such requirements and desiderata.

Embodiments of the invention may provide various additional advantages and benefits, according to implementation details. For example, if there are two (or more) choices for switching plans of one or more devices 102, any or all of which will satisfy the constraints, specifications, user requirements, and so forth, then additional criteria may be utilised to select the "best" plan. For example, it may be desirable to select a plan which leaves the system in a more favourable state for continuing control over future time intervals. The problem of optimising switching plans, in accordance with exemplary embodiments of the invention, is cast initially in terms of time series. In particular, the time interval, already divided into sub-intervals in accordance with the energy consumption constraints received from the broker agent 108, may be sub-divided further into smaller sub-intervals. The "major" intervals are identified, as previously, by an index j, while the sub-intervals within each such interval are identified by an index k. A complete, two dimensional, switching plan for all N devices controlled by the GOA (106), and identified by index i, may be represented as a function Sj(i, k). Every element of this discrete two dimensional function is a binary element, /e has the value 0 or 1.

The capacity constraint is then defined by the following equation:

The internal sum in the above equation, over the discrete two dimensional binary function, may be computed using a well established efficient algorithm for population count. The limit Mj is the cap Ij, scaled in accordance with the time resolution used.

While it is clearly possible to enumerate all possible switching plans over the indices i, j, k, it is apparent that in general only a subset of these will satisfy the capacity constraint. Furthermore, not all of the plans satisfying this constraint will be permissible, in the sense of being compatible with device specifications, user requirements, and so forth. For example, a switching plan that allows one or more refrigeration units to stray outside the specified temperature range, is an impermissible plan. Additional user constraints may exist, such as cost constraints. Formally, if Sιo is the space of all permissible plans for a given device i, then the permissibility constraint may be expressed as follows: so(v) <≡ ^§*o

Having defined a set of permissible plans, which meet all applicable constraints, the task of the GOA 106 is then to identify and implement an optimised plan. It should be understood that there may be no one unique definition of "optimal". For the purposes of the present discussion of exemplary embodiments, it is taken to be plausible to assume that the free-running operation of a device is optimal for the device's operation, and that it is accordingly preferable to operation resulting from intervention by an external control agent. In terms of the graph 500 in Figure 5, the solid operating characteristic 506 is preferable to the dashed operating characteristic 508. Of course, in some circumstances this assumption may be invalid, and some other assumption may be used instead.

Without loss of generality, an optimal plan satisfies the following equation:

S₀ = argmin ^Z ^ωi9 {h{h ')) so{i_>-)ς${o Λ ∑ diso(i,k)<mo _i igi,fceκ

In the foregoing equation, the function g() is an undesirability measure of a proposed plan, and Wj is a preference weighting assigned to the device having index i. In the example as described herein, each device has been assigned a unit weighting, however a preferential system of treatment based upon the criticality of a device's function may be desirable in some situations. The proceeding equation may be understood to express the task of seeking the complete two dimensional binary switching plan S_j(i, k), for which the (weighted) undesirability function is a minimum.

Given the assumption that it is desirable to avoid interfering with the free- running operation of each device, there are at least the following two possibilities (for a binary pattern) for defining the undesirability measure: • minimal total amount of resource restriction, /e minimum total duration for which a device is forced into the "off' state; • minimal frequency of interference.

Taking the second measure (ie minimal frequency of interference) as an example, it may be assumed that the undesirability of a proposed switching pattern is proportional to the number of instances in which a device agent 104 interferes with the running of its associated device 102. The number of interferences is equal to the following cardinality: g (s_o{i, ')) = I {so{i, k- l) = lΛs_o(i, k) = 0 : k G KAp < P_h}

In the foregoing equation, P_h represents the upper limit 504 of Figure 5 at which the free running device will switch "on". It should be understood that the foregoing definition of an undesirability measure is not necessarily unique. For example, another possibility that has been considered is to define undesirability as the Hamming distance between a proposed plan and the optimal switching pattern for each device. This is an appealing approach, which may be extremely effective with appropriate modification, but in an unmodified form has the drawback that a mere phase shift (which may result from a single, short term, interference) may result in a large distance value.

In order to achieve longer term optimality, as noted above, given multiple choices for a plan over the present time interval, it is often more desirable to choose a plan which has a higher probability of providing improved performance over future time intervals. An objective function that reflects this is:

argmm ωiff (so(v)) +

so(v)eSjo Λ J2 diso(i,k)<mo

iei,fceκ jeJSΛj≠o where:

/β(Sy |So) - rrij, /β(Sy |s₀) > rrij

0, otherwise where:

In the foregoing, v(j) is a monotonically decreasing function of j that reflects the increasing uncertainty regarding the capacity constraints in future time intervals, and hence the decreasing importance at the initial time of the corresponding cost terms.

The minimal energy function f() is thresholded at the level of the consumption limit, so that it does not contribute to the cost if it is below the limit.

The appropriate weighting coefficients may be determined as a result of negotiation with the broker agents over the forecast cap values. It should be possible to predict the initial values based upon previous caps.

An optimisation directed to arriving at a preferred final internal state may be defined as follows: SQ = argmin 2__. ω«7 (3o(v)) +

^«0^')/(Sy|(S_fl|5o)) so{i,-)£$io Λ' ∑) diso(i,k)<mo i ie ≠o

The second term in the above equation imposes a preference for a proposed switching plan that maximises the permissible search space for a plan during the subsequent period. Turning now to the solution of the optimisation problem, embodiments of the present invention convert the problem of pattern selection to one of minimal path discovery. The general method applied in preferred embodiments is illustrated by the flow chart 600 of Figure 6. In step 602, there is identified, for each device, a sequence of values of the corresponding control input over the relevant time interval T that results in a minimum resource utilisation of the device (consumption minimisation). In particular, a data structure is constructed, as described in greater detail hereafter, which represents possible sequences of events occurring in the system over the predetermined time interval, and the sequence corresponding with minimum resource utilisation is indetified within the data structure. At step 604, the identified sequence of control input values of each device is modified to improve the satisfaction of specified system and/or device objectives, within specified resource utilisation constraints (constrained optimisation). Finally, at step 606, the device manager agents 104 are instructed by the GOA 106 to control operation of the corresponding devices 102 over the time interval T in accordance with the modified (ie optimised) sequence of control values.

Traditional approaches to this optimisation problem may have employed techniques such as time series analysis. Embodiments of the present invention adopt an entirely different, unique and advantageous approach. In particular, a data structure is constructed that encodes all of the state transitions permitted by the devices' operating constraints, ie all possible switching plans are encoded within a suitable data structure for evaluation. In preferred embodiments, the chosen data structure is a tree structure, and in the examples provided herein this is a binary tree structure (since the control input is binary). Alternative data structures may be applicable, as will be apparent to persons skilled in the art. A minimum energy requirement may be found for a relevant time period by finding a corresponding minimum path on the tree. In this respect, it is assumed that the group's consumption limits, determined by a fellow agent in the distribution network after taking relevant information into account, will at least allow the minimal requirement. Subsequently, any allowance in excess of the minimal requirement is distributed, in the most cost effective way by alternative paths on the tree, to the devices to allow them to run as freely as possible (according to the chosen methods) under the given limit. This may be accomplished by moving iteratively towards the default running patterns to the maximum extent allowed by the limits, starting from the most cost-effective tradeoffs. Alternatively, processing may begin with the free operating patterns, and iteratively moved towards the minimal patterns to the minimum extent required by the group cap.

As will be appreciated by persons skilled in the art, optimal paths may be found on each of the cost trees using the well known Dijkstra minimal path algorithm. The Dijkstra algorithm is a greedy algorithm that computes a minimal path on a graph with non-negativeiy weighted edges from a single source in a single pass.

For the purposes of a general illustration, Figure 7 shows a graph 700 of a pruned binary tree representing a selection of state transitions and switching sequences. The tree is displayed sideways, parallel to the time axis (Ze the horizontal axis), representing all permissible switching state sequences and transitions for the device. Each branching of the tree represents a choice, at the corresponding time instant, between the switching states, given the previous sequence of such choices represented by prior branches. The pruned tree 700 omits branches corresponding with impermissible sequences. As will be appreciated, at least one of the particular series of branchings will correspond with a minimum energy usage. However, there may be numerous other branches of the tree 700 which are permissible, and which represent a lower level of interference with the free running operation of the device.

Figure 8 is a flow chart 800 illustrating further detail of the consumption minimisation step 602. All permissible state transitions are encoded as nodes and edges on a binary tree. For each tree, a minimal (cost) path from root to treetop is determined. The costs along these paths are summed up over all trees {ie for all controlled devices 102). The main product of this computational stage is:

where:

This enables the GOA 106 to supply sufficient information to the broker agent 108 to allow the latter to determine a realistic cap function for the group in question. More particularly, at step 802 the GOA 106 receives current state information from all managed devices 102. This enables the modelling of the energy consumption and internal states of the devices over the planning interval. At step 804, the free running energy requirements are computed over the interval. Additionally, at step 806 the permissibility of different sequences of state transitions is determined, and a tree including all allowable transitions is constructed, and associated primary costs (energy consumption) is determined for each node of the tree. At step 808, the Dijkstra algorithm is used to determine the minimum energy paths. Additionally, corresponding energy requirements are computed, and the secondary costs (Ze the switching costs relating to the undesirability of interference) are computed. Accordingly, at step 810 the GOA 106 is in a position to report on the minimum energy requirements, and free- running energy requirements, of the devices 102 in the group.

The flow charts illustrated in Figures 9A and 9B relate to the subsequent modification step 604. In particular, there are two "blocks" 900, 901 depicted in the flow charts, labelled B2 (900) and B3 (901 ). Block B2 (900) builds up an optimal candidate pool while waiting for a cap function to be provided by the broker agent 108. Here, continuing from the minimal consumption path resulting from the steps 800, a succession of next best paths are determined using the Dijkstra algorithm. Block B3 (901) performs a constrained optimisation on the basis of the candidate pool and the cap function L, received from the broker agents 108. The strategy in B3 is to iteratively exchange energy usage for either less need of interference, or a more desirable exit state, until the group cap is reached.

More particularly, decision step 902 assesses whether a group consumption limit L has been received yet from the broker agent 108. If so, then decision step 904 determines whether the cap is less than a current "worst" candidate. If not, then there may be more candidate paths available that would satisfy the cap, and control returns to step 906. This decision step determines whether all possible candidate paths have been traversed, and if not then at step 908 the next best candidate is identified using Dijkstra algorithm, and a corresponding energy cost SR1 is determined.

Once the cap function L has been received, and all candidate paths under the limit accumulated, control moves to B3 (901 ), in which step 910 is an initialisation step. The constrained optimisation then loops through steps 912, 914 and 916, in which the candidate pool is traversed from the most cost effective trade off to less cost effective trade offs, until the cost limit defined by the cap is reached.

The computation in block B3 (901 ) is illustrated in the graph 1000 of Figure 10. The horizontal axis 1002 is the primary cost (/© energy consumption), while the vertical axis 1004 is the secondary cost (ie switching). Each point, eg 1006, represents the primary and secondary costs corresponding with a particular candidate path. "Lowest energy" points eg 1008, are located on a path 1010 for which the ratio of secondary costs to primary costs has the steepest slope. This line represents the most cost effective trade offs between energy usage and agent-to-device intervention. Accordingly, the optimal plan ultimately chosen is defined by the path in the tree represented by a point on the line 1010 which minimises the secondary costs, while remaining under the capacity limit. For example, in the case represented by the graph 1000, if the cap is set at "10", then the selected path corresponds with the point 1012. The discussion now turns to some exemplary simulation results. In the following examples, a simplified family of device models is employed, having the following parameterised internal state functions: u_\ J -θd(l - e^"^) + P_ft, t moά O < U₀ { s_u[l — e ^υo ) + -π, otherwise ie a repeating sequence of rising/falling exponential functions bounded by Pi and Ph-

Figures 11A and 11 B illustrate a short-term optimisation example. The simulated device is a very large commercial electrical appliance. With reference to Figure 11 A, the graph 1102 shows the internal state of the device in free running mode, while the graph 1104 is the corresponding power status. The device consumes 10 kilowatt hours of energy during a 10 minute period. The graphs 1106 and 1108 represent, respectively, the internal state and power status of the device controlled in accordance with a minimum energy plan, for the same period. The minimum energy requirement is 1 kilowatt hour. By applying a group consumption limit of 2.8 kilowatt hours, the GOA is able to control the device so as to consume 2 kilowatt hours, achieving a 50% reduction^' in agent-to-device intervention, as illustrated in the graphs 1110 and 1112. The corresponding binary tree 1114 is also illustrated in Figure 11 B, and the path 1116 corresponds with the optimised pattern under the consumption cap.

It should be recognised that the maximum energy reduction which can be sustained over a long period through performance adjustment, is determined by the nonlinearity of the internal properties of the devices under control (for example, the internal temperature of a refrigeration unit). While the most aggressive intervention is needed in order to actually achieve the maximum reduction, which is often undesirable as noted previously, it may be important to understand the maximum limit.

Accordingly, simulation experiments have been conducted utilising functions having differing degrees of nonlinearity.

In one set of experiments, a series of 100 trials was performed for a simulated appliance. In each trial, the initial states of the device were randomised and independent. Statistical results were then obtained from the experimental results. The table below represents the statistical results for a near-linear device model, showing a maximum energy reduction of 28.78% achieved compared with free-running.

The following table shows results of similar experiments using a more highly nonlinear device, showing this time a 46.01% saving in energy.

Finally, an even more highly nonlinear device was simulated demonstrating a 69.87% saving in energy.

The experimental results clearly confirm the expectation that greater reductions in energy consumption are possible with more highly nonlinear devices.

Further experiments involving the control of groups of devices having relatively high nonlinearity have demonstrated that the degree of intervention required may be significantly reduced through trade-off of reduced energy consumption under a fixed supply cap. For example, in one experiment an average of 1.51 interventions was needed to achieve maximum energy reduction, with the minimal energy consumption being 1.1 kilowatt hours on average. By setting a cap of 4.06 kilowatt hours, over a 30 minute period, optimised running patterns consumed 1.77 kilowatt hours on average, and brought down the need for agent-to-device interference to a mean of 0.89 times. This represents a 41.06% reduction in interference. However, these experiments included only a single device in the group, and the group agent was therefore only able to utilise 43.60% of the energy allocated to the group.

However, by increasing the number of devices in the group to 10, with an energy cap of 38.16 kilowatt hours, the GOA was able to reduce intervention by 69.78% while utilising 91.43% of the available consumption limit.

Further examples are illustrated in Figures 12 and 13. Figure 12 illustrates optimisation for minimal intervention. In particular, the optimisation effects on one device within a group of four devices are demonstrated. In graph 1202 the internal condition (eg temperature) of the device is illustrated under a minimum energy consumption plan, and the corresponding power status is illustrated in the graph 1204. Given a group consumption limit of 22.8 kilowatt hours, 8 kilowatt hours was allocated to the device, with the resulting internal condition illustrated in the graph 1206, and power status in graph 1208. A 60% reduction in intervention is achieved.

Further experiments of this type have demonstrated that with continuing reduction in the available cap, thereby forcing the device closer to the minimal energy consumption plan, the optimal pattern is "broken up" moving from a lower frequency "free running" pattern to a higher frequency minimal consuming pattern.

Figure 13 illustrates optimisation for best exit temperature on a large commercial refrigerator managed by a GOA. The graphs 1302 and 1304 represent the internal condition (Je temperature) and corresponding power status for minimum consumption. The graphs 1306 and 1308 show corresponding internal condition and power status under a group consumption limit of 14.2 killowatt hours, with 14 kilowatt hours allocated to the device, and applying the additional constraint of achieving an optimal exit state (Ze lowest temperature). It is once again emphasised that the foregoing described embodiments of the invention are intended to be exemplary only, and should not be considered limiting of the scope of the invention, as defined in the following claims.

Claims

CLAIMS:

1. A method of controlling resource utilisation over a predetermined time interval in a system including a plurality of facilities, each of which has a control input which influences a resource utilisation of the facility, the method including the steps of, for each facility: constructing a data structure representing possible sequences of events occurring in the system over the predetermined time interval; identifying within the data structure a sequence of values of the control input over the time interval that results in a minimum resource utilisation of the facility; modifying the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and controlling the operation of the facility over the time interval in accordance with the modified sequence of control values.

2. The method of claim 1 which is implemented in an energy supply network, wherein the facilities are energy consuming devices or appliances, and resource utilisation is an energy resource utilisation.

3. The method of claim 1 or claim 2 wherein the step of modifying includes an iterative process of trading-off resource utilisation against satisfaction of system and/or facility objectives within the specified resource utilisation constraints.

4. The method of claim 3 wherein the facilities have preferred patterns of operation, and the process includes iterating from the identified sequence of values of the control input corresponding with a minimum resource utilisation towards said preferred patterns, to the maximum extent allowed by the utilisation constraints.

5. The method of claim 3 wherein the facilities have preferred patterns of operation, and the process includes iterating from said preferred patterns towards the identified sequence of values of the control input corresponding with minimum resource utilisation, to the minimum extent required by the utilisation constraints.

6. The method of any one of the preceding claims wherein each facility has an operating state which is selected from at least two available operating states, and wherein the operating states of each facility switches between the available operating states in accordance with the control input, and at least one internal condition of the facility.

7. The method of any one of the preceding claims further including the step of providing a model of each facility which enables the resource utilisation of the facility to be estimated over the predetermined time interval as a function at least of an input sequence of control values, and which further includes calculating minimum resource utilisation, and resource utilisation of modified sequences of control input values, utilising said facility models.

8. The method of any one of the preceding claims wherein the constraints include capacity constraints and/or permissibility constraints.

9. The method of any one of the preceding claims wherein the step of identifying a sequence of values of the control input over the time interval that results in a minimum resource utilisation of the facility includes applying a shortest path algorithm to the data structure.

10. The method of any one of the preceding claims wherein said events includes selections of control input values.

11. The method of any one of the preceding claim wherein said events include transitions in facility operating states.

12. The method of any one of the preceding claims wherein the data structure is a tree structure encoding sequences of facility states in nodes thereof, and transitions between said states in branches thereof, and wherein a cost at each said node, and one for each said transition, are assigned according to a corresponding facility resource utilisation.

13. The method of claim 12 wherein the sequence of values of the control input resulting in minimum resource utilisation, and modified sequences of values, are determined using a shortest path algorithm over the tree structure.

14. The method of any one of the preceding claims which is implemented by a group optimising agent (GOA) of an energy supply network.

15. A method of controlling resource utilisation over a predetermined time interval in a system including a plurality of facilities, each of which has a control input which influences a resource utilisation of the facility, the method including the steps of, for each facility: constructing a data structure in the form of a tree structure which encodes sequences of facility states in nodes thereof, and transitions between said states in branches thereof, whereby the tree structure represents possible sequences of events occurring in the system over the predetermined time interval, and wherein a cost at each said node, and one for each said transition, are assigned according to a corresponding facility resource utilisation; identifying within the tree structure a sequence of values of the control input over the time interval that results in a minimum resource utilisation of the facility; modifying the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and controlling the operation of the facility over the time interval in accordance with the modified sequence of control values.

16. The method of claim 15 wherein the sequence of values of the control input resulting in minimum resource utilisation, and modified sequences of values, are determined using a shortest path algorithm over the tree structure.

17. An energy distribution system including a plurality of energy consuming facilities, each of which has a control input which influences a resource utilisation of the facility, and at least one controller operatively associated with a corresponding group of facilities, wherein the controller is adapted to, for each facility within the associated group: identify a sequence of values of the control input over a predetermined time interval that results in a minimum resource utilisation of the facility; modify the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and control the operation of the facility over the time interval in accordance with the modified sequence of control values.

18. A controller for use in an energy distribution system which includes a plurality of energy consuming facilities, each of which has a control input which influences a resource utilisation of the facility, the controller being operatively associated with a corresponding group of facilities, and being adapted to, for each facility in the group: identify a sequence of values of the control input over a predetermined time interval that results in a minimum resource utilisation of the facility; modify the identified sequence of values of the control input to improve satisfaction of specified' system and/or facility objectives within specified resource utilisation constraints; and control the operation of the facility over the time interval in accordance with the modified sequence of control values.

19. A controller for use in an energy distribution system which includes a plurality of energy consuming facilities, each of which has a control input which influences a resource utilisation of the facility, the controller being associated with a corresponding group of facilities and including: at least one microprocessor; at least one input/output interface device for providing control inputs to the associated group of facilities; and at least one memory/storage device operatively associated with the microprocessor, wherein the memory/storage device includes executable instruction code which, when executed by the microprocessor, causes the controller to implement the steps of, for each facility: identifying a sequence of values of the control input over the time interval that results in a minimum resource utilisation of the facility; modifying the identified sequence of values of the control input to improve satisfaction of specified system and/or facility objectives within specified resource utilisation constraints; and controlling the operation of the facility over the time interval in accordance with the modified sequence of control values.

20. The system of claim 17, or the controller of claim 18 or claim 19, wherein the controller includes or implements a group optimising agent (GOA).