Method and system for determining a propagation delay in a fluid transport system

Abstract

A method of determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system, the fluid transport system comprising one or more conduits for transporting the fluid from at least the adjustment location to at least the downstream location, wherein a change of said fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or more conduits, wherein the method comprises: providing a plurality of property propagation models, each property propagation model being configured for modeling propagation of the fluid property between the adjustment location and the downstream location, each property propagation model being based on a respective candidate propagation delay; evaluating a performance of each of the plurality of property propagation models; selecting one or more of the plurality of property propagation models based on the evaluated performance; determining an estimated propagation delay from the one or more candidate propagation delays of the selected one or more property propagation models.

Description

[0001]Method and system for determining a propagation delay in a fluid transport systemTECHNICAL FIELD The present invention relates to the determination of a propagation delay in a fluidtransport system, in particular in a fluid-based energy distribution system, e.g. in adistrict energy system, such as a district heating system or a district cooling system.BACKGROUNDIn a fluid transport system, a fluid is transported along one or more conduits, e.g. fromat least one distribution source via a network of conduits to a plurality of recipients. Forexample, in a fluid-based energy distribution system, water or another fluid is used as amedium to transport thermal energy between a distribution source and a plurality ofrecipients. Examples of fluid-based energy distribution systems include heating systems and cooling systems. In particular, examples of fluid-based energy distribution systemsinclude district energy systems, such as district heating or cooling systems. In a districtheating system, water is heated at a district heating plant and transported to a pluralityof consumers via a network of supply conduits and returned from the consumers to thedistrict heating plant via a network of return conduits. District heating or coolingsystems, which are sometimes also referred to as heat or cooling networks or heatinggrids or cooling systems, may have a variety of different sizes and, depending on theirsize, district heating systems may sometimes also be referred to as “campus heating system,” “Fernwärmesystem”, “Nahwärmesystem”, or the like. For the purpose of the present disclosure, the term district heating system is intended to refer to a heating system configured to distribute heat from one or more distribution sources to a plurality of different buildings via a network of supply and return conduits, wherein the district heating system uses a fluid as a medium for transporting the heat through the network.Similarly, the term district cooling system is intended to refer to a cooling systemconfigured to distribute cooling from one or more distribution sources to a plurality ofdifferent consumers, e.g. to different buildings or other structures, via a network ofsupply and return conduits, wherein the district cooling system uses a fluid as a medium. Here and in the following, district heating and district cooling systems will collectively be referred to as district energy systems. It will be appreciated that some embodiments of a district energy system may be configured to provide heating as well as cooling. Fluid-based energy distribution systems as disclosed herein also find applications as heating and / or cooling systems for individual buildings or other commercial orresidential structures. Accordingly, the term fluid-based energy distribution system asused herein is intended to include district energy systems as well as heating and / or cooling systems of individual buildings or other commercial structures where thermal energy is distributed from one or more distribution sources to a plurality of recipients bytransporting a fluid via a network of supply and return conduits.Other examples of fluid transport systems include waste water systems, where fluid istypically transported from multiple source locations to one or more central waste watertreatment plants. Generally, in a fluid transport system fluid is transported between one or more source locations and one or more destination locations. It will be appreciated that the terms source location and destination location are not intended to imply that the fluid flow is necessarily created at the source location and terminated at thedestination location. For example, a source location may be any node or other locationalong a conduit or within a network of conduits from which fluid flows to one or more downstream destination locations along the conduit or within the network of conduits.For example, a source location may be a mixing node, at which fluid from two or moreincoming flow paths are mixed to form an outgoing combined flow. Similarly, the fluid flow may continue beyond a destination location, e.g. in systems where the fluid is returned from the destination locations to a source location, e.g. along a suitable return flow path. Accordingly, a destination location may be any node or other location along a conduit or within a network of conduits where fluid flow arrives from one or more upstream source locations. In fluid transport systems, the fluid flows along the one or more conduits. Certain properties of the fluid, such as fluid temperature, concentrations of the constituents ofthe fluid, or the like, also propagate along the one or more conduits with the fluid flow,i.e. the propagation of such properties along the one or more conduits is an advective propagation. Accordingly, when a property of the fluid changes at one location along the conduit, the change of the fluid property propagates along the conduit where the propagation speed of the change of the fluid property depends on the flow rate of thefluid. Accordingly, a change of the fluid property occurring at an upstream locationcauses a delayed change of the fluid property at a downstream location. For the purpose of the present description, the upstream location will also be referred to as the adjustment location, and the time delay between a change of the fluid property occurring at an upstream location and a resulting change of the fluid property at a downstream location will be referred to as a propagation delay. In many fluid transport systems, it is desirable to determine the propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. For example, it may be desirable to determine the propagation delay in order to providea more efficient control of the fluid property at the downstream location by means ofaffecting adjustments of the fluid property at the upstream adjustment location.For example, in a fluid-based energy distribution system, it is often desirable to controlthe temperature of the fluid at a downstream location close to the recipients of theenergy. However, the system may only be capable of affecting a change of the fluidtemperature at an adjustment location, which may be upstream from the locationwhere the change of the fluid property is desired. In many fluid transport systems, the distance between the upstream adjustment location and the downstream location maybe considerable. For example, in a district energy system, the adjustment location maybe at or near a district heating or cooling plant while the downstream location may be alocation close to the recipients, possibly several kilometers from the district heating orcooling plant. Accordingly, there may be a considerable delay from when a change of the fluid property is affected at the adjustment location until the change is actuallynoticeable at the downstream location. Moreover, the propagation delay is oftenunknown a priori and / or may vary over time, e.g. depending on the current energyconsumption and, hence, flow rate at the various recipients.It is a problem related to the control of fluid transport systems that the propagationdelay inherent in these systems is flow-dependent and variable, and that the delay isoften unknown a priori. It thus remains desirable to provide a method for determiningthe propagation delay associated with the propagation of a fluid property of a fluid in a fluid transport system, e.g. in order to allow for an efficient control of the fluid property at a downstream location despite a priori unknown and / or varying propagation delaysthat nevertheless influence the propagation dynamics in a significant manner. Anexample of a fluid transport system is a district energy system with long pipes andvarying flow rates in which the fluid temperature of the fluid arriving at the recipients isto be controlled. It will be understood, however, that, for the purpose of the determination of the propagation delay, the fluid property does not necessarily have to be actively adjusted at the adjustment location. In this respect, the term adjustment location is merely intended to refer to a reference location for determining the propagation delay between a change of a fluid property occurring at the adjustment location and a corresponding change of the fluid property occurring at a downstream location. It is further desirable to provide a method for determining a propagation delay that iscost efficient and easy to install. For example, it is desirable to provide a method fordetermining propagation delays of temperature changes that can be implemented in temperature controllers that are easy to commission without detailed a prioriknowledge of the propagation delays of the system.It is further desirable to provide a method for determining propagation delays that requires relatively few sensors for sensing the fluid property in question. It is further desirable to provide a method for determining propagation delays that does not require detailed knowledge of the physical structure of the network of conduits between the adjustment location and the downstream location, such as knowledgeabout pipe dimensions, the number and location of branching points, etc.SUMMARYIn view of the foregoing, it remains desirable to provide a method and system fordetermining a propagation delay of a fluid property of a fluid in a fluid transport systemthat solve one or more of the above problems and / or that have other benefits, or that atleast provide an alternative to existing solutions.According to one aspect, disclosed herein are embodiments of a method of determininga propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system, the fluid transport system comprising one or more conduits for transporting the fluid from at least the adjustment location to at least the downstream location, wherein a change of said fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or more conduits, wherein the method comprises:- providing a plurality of property propagation models, each property propagationmodel being configured for modeling propagation of the fluid property between the adjustment location and the downstream location, each property propagation model being based on a respective candidate propagation delay,- evaluating a performance of each of the plurality of property propagation models,- selecting one or more of the plurality of property propagation models based on theevaluated performance;- determining an estimated propagation delay from the one or more candidatepropagation delays of the selected one or more property propagation models. Accordingly, embodiments of the method disclosed herein use a bank of models to model the propagation of the fluid property in the system, where different models are based on different candidate delays. The process evaluates the performance of the models and determines an estimated propagation delay based on the evaluation. This is a process that can be implemented in a computationally efficient manner, thus making the method applicable for decentral implementation, in an edge device, a pump, or other device with limited computational power. Moreover, embodiments of the method can be implemented without detailed knowledge of the structure of the conduits of the fluid transport system such as pipedimensions, etc. In particular, at least in some embodiments, the method only needs toreceive actual values of the fluid property at the adjustment location and at the downstream location, and the method does not rely on detailed knowledge about otherdetails of the system, e.g. knowledge of the current flow rate at different locations alongthe network of conduits. Accordingly, embodiments of the method can efficiently be commissioned in respect of a fluid transport system. Embodiments of the method disclosed herein are suitable for fluid transport systems where the flow rate varies over time and / or is unknown, as the property propagation models do not receive the flowrate as an input, or include the flow rate as a model parameter, or otherwise requireknowledge of, or make assumptions of, the flow rate. The fluid transport system may bea variable-flow fluid transport system. Accordingly, embodiments of the method disclosed herein are suitable for a wide variety of fluid transport systems and for a wide variety of control strategies for controlling the fluid transport system, including variable- flow strategies. Evaluating the performance of each of the plurality of property propagation models may comprise using said property propagation model to compute one or more estimated values of the fluid property at the downstream location and calculating one or more errors between respective ones of said one or more estimated values and corresponding one or more obtained actual values of the fluid property at the downstream location; and wherein selecting one of the plurality of property propagation models is based on the one or more calculated errors.Obtaining the actual value of the fluid property at the downstream location maycomprise receiving sensor data indicative of the actual value. The process may receivethe sensor data directly from a sensor, or from a SCADA system, or otherwise. Thechoice of sensor depends on the fluid property to be controlled. For example, the fluid property may be a fluid temperature or a concentration of a component of the fluid, or another fluid property that propagates along the one or more conduits with the fluid flow. When the fluid property is a fluid temperature, the sensor comprises atemperature sensor for measuring the fluid temperature at the downstream location.When the fluid property is a concentration of a component of the fluid, the sensorcomprises a sensor for measuring such concentration at the downstream location. In some embodiments, the plurality of property propagation models includes one ormore adaptive models. Accordingly, the method may further comprise recursivelyupdating the one or more adaptive models based on obtained actual values of the fluid property at the adjustment location and / or on obtained actual values of the fluid property at the downstream location. Each adaptive model may comprise one or more adaptable model parameters, and recursively updating may comprise recursively updating the one or more adaptable model parameters of the respective adaptive models. When some or all of the property propagation models are adaptive models, the number of models can be reduced while still allowing for a reliable selection of one ormore candidate delays that best reflect the actual propagation delay of the fluidtransport system. In particular, as the models are recursively adapted based on actual values of the fluid property, the risk is reduced that the model selection is unduly influenced by other model assumptions, such as assumed gain parameters, and / or the like.For the purpose of the present disclosure, recursively updating is intended to refer to acomputation based on an equation that relates a current value of at least one model parameter with at least one earlier value of the model parameter, in particular a value of the model parameter at a preceding discrete time step. In some embodiments, the one or more adaptive models include one or more adaptive filters, in particular linear adaptive filters, thereby providing a method that can be implemented in a particularly computationally efficient manner while providing anaccurate determination of the propagation delay. In particular, embodiments of themethod disclosed herein are suitable for a real-time control of the fluid transport system and can be implemented by a pump controller, an edge device or the like. The deviceimplementing the method does not need to receive information about the network structure of the fluid transport system, the control strategy, the fluid flow, the environmental temperature, etc. Accordingly, the method can be implemented with lowlatency and low communication requirements. The recursive adaptation is also memoryefficient, as the system only needs to maintain a history of observed values of the fluid property corresponding to the longest candidate propagation delay. In some embodiments the adaptive filter is a FIR filter. The candidate propagation delays of the respective property propagation models may be respective model-specific, non-adaptable model attributes of the respective adaptive model, i.e. each model may have a fixed candidate delay associated with it. In particular, at least in some embodiments where the property propagation models are adaptive models, the candidate propagation delay associated with each model is not adapted bythe recursive update of the adaptive model. The candidate delay may be a constantparameter of the model, a part of the model structure, or otherwise. In some embodiments, the method further comprises detecting one or more excitationproperties of an input signal to the respective property propagation models. Generally,the excitation properties may be indicative of the ability of the input signal to excitedifferences between the property propagation models, such that the excited differencescan be captured when observing the outputs of the property propagation models. Inparticular, the input signal may be indicative of the actual value of the fluid property at the adjustment location. The actual value of the fluid property at the adjustment location may be obtained from a suitable sensor or otherwise, e.g. from a control process that controls adjustment of the fluid property at the adjustment location. The one or more excitation properties may be indicative of a variation of the input signal, in particular a variation of the input signal other than a variance attributable to noise, in particular measurement noise. The excitation property may be indicative of a degree of variation of the input signal. The one or more excitation properties may be detected in a number of ways, e.g. by determining a gradient of the input signal, by determining a low-pass-filtered variance of input signal values obtained over a measurement time, and / or otherwise. Recursively updating the one or more adaptive models may comprise selectively updating the one or more adaptive models only when the detected one or more excitation properties fulfil a predetermined excitation condition, in particular only when the input signal includes sufficient variability to allow the model parameters of theadaptive models to be determined, in particular to be uniquely determined. Thecorresponding excitation condition may thus be chosen to be that the input signalpersistently excites the adaptive models. In some embodiments, the excitation conditionmay be chosen to be that the gradient or a low-pass filtered variance of the input signalexceeds a predetermined threshold, or otherwise. Accordingly, an improved modelselection and, hence, a more accurate determination of the propagation delay can be achieved. To this end, in some embodiments, the process may recursively update a measure for detecting persistent excitation. The measure for detecting persistent excitation may bedifferent from the variance of the input signal. If said measure is not above apredetermined threshold, the process may determine that the system corresponds to a steady state. In that case, the process may use a cross-correlation analysis to detect whether the input signal may be considered to represent white noise. If the process determines that the input signal may be considered to represent white noise, theprocess may calculate the noise variance and use the calculated noise variance formodel selection, e.g. by setting a variance in a probability density function for modelselection at a level at or above the calculated noise variance. In alternativeembodiments, the process may calculate the variance of the input signal regardless of whether the signal is persistently exciting or not. The process may then low pass filterthe calculated variance and use the low-pass filtered variance as a persistent excitationmeasure. If the low-pass filtered variance drops to a low value (e.g. lower than apredetermined threshold), the process may use the low-pass filtered variance todetermine the variance of the probability density function for use in the modelselection. While this alternative approach may be less computationally involving thancalculating cross correlation of a signal, the approach based on the cross-correlation may be more robust. In some embodiments, selecting one or more of the plurality of property propagation models comprises selecting the property propagation model that fits the measurementsbest or that is the most likely model. The inventors have realized that the determinationof a propagation delay from one or more candidate property propagation models, which is / are selected based on a best fit or maximum likelihood criterion, provides an efficientyet accurate determination of the propagation delay. In this respect, the best fitcriterion may be chosen such that the selected property propagation model is the model resulting in the smallest calculated errors between the actual values of the fluid property at the downstream location and the corresponding estimated values by the property propagation model, e.g. based on an RMSE or mean-square comparison. The maximum-likelihood criterion may be based on a statistical determination of which property propagation models provide a series of estimated fluid property values thatwith the maximum likelihood corresponds to an expected series of actual values, wherethe expected series of actual values may depend on a predetermined assumption of thenoise properties of the system. The maximum-likelihood criterion may e.g. be based ona Bayesian model selection or other suitable model selection methods.Accordingly, in some embodiments, the method utilizes Bayesian model selection to identify the most probable property propagation model and selects said identified most probable property propagation model rather than selecting the model with the smallest residuals. Bayesian model selection is advantageous as it incorporates prior knowledge about the probability distribution of the errors and therefore is less sensitive to data outliers and noisy measurements. When tracking time-varying propagation delays, which are functions of the flow rate, it can be assumed that these evolve smoothly over time, due to the inertia of the fluid. Hence, the model error distribution for the next timestep can be expected to be in the vicinity of the error distribution calculated at thecurrent time step. Therefore, Bayesian model selection is also very suitable for robustlytracking delays influenced by varying flow rates.In some embodiments, the method further comprises computing an input signal noisevariance of an input signal, the input signal being indicative of the actual value of the fluid property at the adjustment location; wherein evaluating the performance is based at least in part on the computed input signal noise variance. For example, the computed input signal noise variance may be used in the model selection, e.g. the Bayesian model selection, to obtain improved probability estimates of the error for a given delay. The input signal noise variance may for example be calculated using recursive filtering or cross correlation analysis of the input signal history. In some embodiments, the method comprises detecting one or more excitation properties of the input signal, and selectively computing the input signal noise variance only when the detected one or more excitation properties fail to fulfil a predetermined excitation condition, in particular only when the input signal is not persistently excitedor when the signal variability can predominantly be attributed to measurement noise.Accordingly, an accurate estimate of the noise-induced signal variance is provided, which, in turn, may improve the model selection. The predetermined excitation condition may be the same condition used for the purpose of deciding whether or not to recursively update the adaptive models, or it may be a different condition indicative of the variability of the input signal. Accordingly, in some embodiments, the method comprises:- recording the input signal over a period of time to obtain an input signal history,- selecting a part of the obtained input signal history for which the recorded input signalvalues represent random variables having a distribution that fulfils one or more predetermined criteria,- selectively calculating the input signal noise variance for only the selected part of theobtained input signal history. The predetermined criteria may e.g. include that the recorded input signal values represent independent and identically distributed random variables, optionally with theexception of a predetermined maximum amount of outliers. This may be detected byperforming an auto-correlation analysis of the signal. e.g. by verifying that the signal isonly correlated with itself above lag 1 to an extend that it can be considered randomlycoincident. Accordingly, in some embodiments, the input signal noise variance iscalculated only when the recorded input signal values represent measurement noise ofan otherwise steady input signal. Here, the condition that the recorded input signalvalues represent independent and identically distributed random variables correspondto a situation where the noise at time k does not depend on noise at any other time inthe selected part of the input signal history and that the distribution of noise at time k isthe same distribution as at all other times within the selected portion of the input signalhistory. It will be appreciated that, in practice, this condition may be fulfilled in anapproximate manner only, e.g. within predetermined acceptance margins.In some embodiments, selecting one or more of the plurality of property propagation models comprises selecting a single model of the one or more property propagation models based on the evaluated performance; and determining the estimated propagation delay comprises selecting the candidate propagation delay of the selected property propagation model as the estimated propagation delay, thereby providing asimple selection process. Alternatively, the method may select more than a single modeland determine the estimated propagation delay from the candidate propagation delaysof the selected models, e.g. as an average of the candidate propagation delays of theselected models, or otherwise. This embodiment may provide a more accurate estimateof the propagation delay, e.g. in situations where two property propagation modelshave similar likelihoods or otherwise perform similar to each other.In some embodiments, the method further comprises resetting the one or more of the adaptive models responsive to a predetermined reset criterion. Resetting may comprise setting one or more model parameters of the one or more adaptive models to respective initial values, e.g. predetermined initial values. The reset criterion may include lapse of a predetermined time period, e.g. such that one or more of the adaptivemodels are periodically reset. Other examples of reset criteria include a criterion basedon computed likelihoods of the respective adaptive models as computed during the model selection. For example, one or more of the adaptive models may be reset responsive to the distribution of models fulfilling one or more predetermined resetcriteria, e.g. responsive to one of the adaptive models having a likelihood much largerthan the sum of likelihoods of the remaining adaptive models. In some embodiments resetting one or more of the adaptive models may comprise resetting all models, e.g. such that the distribution of models fulfills a predetermined criterion, e.g. a uniform distribution, or a distribution where the difference in likelihoods between the mostlikely model and the remaining models is more pronounced.Generally, the fluid may be water or another liquid. Examples of the fluid property include a fluid temperature, a concentration of a component of the fluid being transported, etc. or another fluid property that propagates at least predominantly in an advective manner. Examples of a fluid transport system include a fluid-based energy- distribution system where a fluid is used as a medium for transporting energy, e.g. a district heating and / or cooling system. Other examples include a waste water system, a fluid distribution system, etc. The present disclosure relates to different aspects including the method described above and in the following, corresponding apparatus, systems, methods, and / or products, each yielding one or more of the benefits and advantages described in connection with one or more of the other aspects, and each having one or more embodiments corresponding to the embodiments described in connection with one or more of the other aspects and / or disclosed in the appended claims. In particular, according to another aspect, disclosed herein are embodiments of a method of controlling a fluid property at a downstream location of a fluid transport system by adjusting the fluid property at an adjustment location, the downstream location being downstream from the adjustment location, the fluid transport system comprising one or more conduits for transporting the fluid from at least the adjustment location to at least the downstream location, wherein a change of said fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or more conduits, wherein the method comprises:- performing the acts of the method for determining a propagation delay of an advectivepropagation of a fluid property between an adjustment location and a downstream location of a fluid transport system as disclosed above or in the following so as to determine an estimated propagation delay of the fluid property between the adjustment location and the downstream location, -controlling adjustment of the fluid property at the adjustment location based at least in part on the estimated propagation delay. Controlling adjustment of the fluid property may be based on a suitable control strategy known as such in the art, e.g. as described in Bin Zhou et al.: “Truncated predictor feedback for linear systems with long time-varying input delays”, Automatica, Volume 48, Issue 10, October 2012, Pages 2387-2399, or otherwise. Controlling adjustment of the fluid property may comprise causing a change of the fluid property at the adjustment location. To this end, the method may directly control a mechanism for affecting a change of the fluid property at the adjustment location, e.g. by controlling a valve, a pump, a dosing pump, a heater, a chiller, a boiler and / or thelike. Alternatively or additionally, the method may compute an adjustment of the fluidproperty to be affected at the adjustment location, and communicate the computed adjustment of the fluid property to a control system for controlling such mechanism, thereby causing the control system to affect the change. Generally, the computedadjustment may be indicative of a desired value or set point of the fluid property at theadjustment location, i.e. the computed adjustment may be indicative of a result of the adjustment to be performed. In the following, the desired value of the fluid property at the adjustment location will also be referred to as a target value of said fluid property atthe adjustment location. Alternatively or additionally, the computed adjustment may beindicative of a required change or increment of the fluid property at the adjustment location, i.e. the computed adjustment may be indicative of a magnitude of the change to be affected. As used herein, the term increment is intended to encompass both positive and negative increments. Yet alternatively or additionally, the computed adjustment may be represented in another suitable form.Embodiments of the methods disclosed herein may be computer-implemented.Accordingly, further disclosed herein are embodiments of a data processing systemconfigured to perform the steps of the method described herein. In particular, the dataprocessing system may have stored thereon program code adapted to cause, whenexecuted by the data processing system, the data processing system to perform thesteps of the method described herein. According to another aspect, disclosed herein are embodiments of a control system for controlling a fluid property of a fluid in a fluid transport system, the control systembeing configured to perform the steps of the method as disclosed herein. In particular,the control system may include a data processing system as disclosed herein. The control system may be a PLC-based system, a computer-implemented system, a SCADA system, or the like.Embodiments of the control system and / or of the data processing system may beembodied as a single computer or other data processing device, or as a distributed system including multiple computers and / or other data processing devices, e.g. a client-server system, a cloud-based system, etc. The control system and / or the data processingsystem may include a data storage device for storing the computer program and,optionally, sensor data and / or other data indicative of obtained actual values of the fluidproperty. The control system and / or the data processing system may include acommunications interface for receiving data indicative of actual values of the fluidproperty, such as sensor data, e.g. from one or more sensors, in particular from a sensorconfigured to sense the fluid property at the downstream location and to provide sensordata indicative of the actual value of the fluid property at the downstream location. Thecontrol system and / or the data processing system may receive the data indicative ofactual values of the fluid property directly or indirectly from the one or more sensors orother data source via a suitable wired or wireless communicative connection, e.g. via a suitable communications network, or otherwise.The control system and / or the data processing system may comprise or becommunicatively coupled with one or more actuators or other control mechanism for affecting a change of the fluid property at the adjustment location. The actuator orcontrol mechanism may e.g. include one or more valves, pumps or other devices forselectively mixing fluid having different respective values of the fluid property, or otherwise. When the fluid property is the temperature of the fluid, the actuator or control mechanism may include a boiler, heat exchanger, chiller or other device for changing the temperature of the fluid at the adjustment location.The control system and / or the data processing system may provide a user-interface forallowing a user to monitor operation of the fluid transport system. The data processingsystem may also issue warnings or alerts or other notifications, e.g. responsive todetected communication failures, e.g. audible or visual alerts, alerts communicated via e-mail, SMS, or other forms of notifications, and / or the like. The control system and / or the data processing system may comprise or be communicatively coupled to a first sensor for acquiring the actual value of said fluidproperty at the downstream location. Optionally, the control system and / or the dataprocessing system may comprise or be communicatively coupled to a second sensor for acquiring an actual value of the fluid property at the adjustment location.Another aspect disclosed herein relates to a fluid transport system comprising a controlsystem and / or a data processing system as disclosed above or in the following.The fluid transport system may further comprise one or more conduits for transportingthe fluid from at least the adjustment location to at least the downstream location,wherein a change of the fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or moreconduits. The fluid transport system may further comprise a first sensor for acquiringthe actual value of said fluid property at the downstream location. For the purpose of the present disclosure the one or more conduits will also be referred to as a network of conduits. Yet another aspect disclosed herein relates to embodiments of a computer programconfigured to cause a data processing system or a control system to perform the acts ofthe method described above and in the following. A computer program may compriseprogram code means adapted to cause a data processing system or a control system to perform the acts of the method disclosed above and in the following when the programcode means are executed on the data processing system or the control system. Thecomputer program may be stored on a computer-readable storage medium, inparticular a non-volatile storage medium, or embodied as a data signal. The storagemedium may comprise any suitable circuitry or device for storing data, such as a RAM, a ROM, an EPROM, EEPROM, flash memory, magnetic or optical storage device, such as a CD ROM, a DVD, a hard disk, and / or the like. BRIEF DESCRIPTION OF THE DRAWINGS Preferred embodiments will be described in more detail in connection with the appended drawings, where:FIGs.1A-B schematically show an example of a fluid transport system,FIG.2 schematically illustrates an example of a process for determining a propagationdelay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system,FIG.3 schematically illustrates a more detailed view of an example of a process fordetermining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. FIG.4 schematically illustrates a flow diagram of an example of a process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system, FIG.5 schematically illustrates a more detailed flow diagram of an example of a process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system, FIG.6 schematically illustrates a flow diagram of yet another example of process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. DETAILED DECRIPTION In the following, embodiments of various aspects disclosed herein will be described in the context of temperature control in a district heating system, also referred to as adistrict heating grid. It will be appreciated however that the various embodimentsdisclosed herein may also be applied to other types of systems, e.g. other types of fluid- based energy distribution systems or other types of fluid transport systems where the determination of the propagation delay of temperature changes is desirable, or even to fluid transport systems where the determination of the propagation delay of another fluid property is desired, e.g. a waste water system where the determination of the propagation delay of changes of a concentration of a component of the fluid is desired.FIGs.1A-B schematically show an embodiment of a fluid transport system, generallydesignated by reference numeral 1. While FIG.1A shows a schematic view of the overall fluid transport system, FIG.1B shows a more detailed view of a part of the fluidtransport system. The fluid transport system 1 comprises one or more conduits fortransporting a fluid. In the present embodiment, the fluid transport system transports the fluid between a distribution source 3 and a number of recipients 4. In the presentexample, the fluid transport system 1 is a district heating system, the fluid is water, andthe distribution source 3 is a district heating plant, which provides heated water.However, it will be appreciated that various embodiments of the methods, apparatus and system disclosed herein may also be applied to other types of fluid transportsystems, in particular to other types of fluid-based energy distribution systems. It will beappreciated that, in other embodiments, the fluid transport system may transport the fluid from another type of source location to another type of destination location.The fluid transport system 1 comprises a supply network 2 of one or more supplyconduits 5 for transporting the fluid from the distribution source 3 to the plurality ofrecipients 4. In the example of a district heating system, the recipients may be domestic and / or commercial buildings, such as family homes, apartment buildings, office buildings or other commercial buildings. It will be appreciated, however, that otherembodiments may include other types of recipients. Moreover, different embodimentsmay have different numbers of recipients. The supply network of supply conduits 5 mayinclude a main supply line and a plurality of branch lines, branching off from the mainsupply line to respective recipients 4. The supply network may comprise multiple zones2a-2e, each comprising one or more recipients 4. The zones may be delimited byrespective zone boundaries 9 along the supply conduits 5, e.g. by respective nodes ofthe network of supply lines where the fluid flows into a zone, or otherwise. However,other embodiments may have a different network structure. The fluid transport system1 further includes a return network of return conduits 6 (see FIG.1B, not explicitlyshown in FIG.1A) for returning fluid from the respective recipients 4 to the distributionsource 3. The supply conduits and / or return conduits may be pipes or other suitabletypes of conduits. In the schematic illustration of FIG.1B, for the sake of simplicity ofillustration, only a single zone 2a is illustrated, and the network of supply conduits 5 andthe network of return conduits 6 are each illustrated as an individual pipe. It will beappreciated, however, that many embodiments may have a more complex networkstructure, where the network of supply conduits comprises more than one conduit (e.g.as illustrated in FIG.1A) and / or where the network of return conduits comprises morethan one conduit. Similarly, for the sake of simplicity of illustration, only a singledistribution source 3 is illustrated in FIGs.1A-B. It will be appreciated, however, that some embodiments may have more than one distribution source, or no distribution source at all.The fluid transport system 1 may further comprise one or more pumps (not explicitlyshown in FIGs.1A-B) for pumping fluid between the distribution source 3 and therecipients 4. The distribution source 3, in this example the district heating plant, comprises amechanism for heating (or otherwise adjusting the temperature of) the fluid to betransported from the distribution source 3 to the recipients 4, i.e. the distribution source3 is capable of controlling the temperature of the fluid leaving the distribution source 3. Accordingly, as best illustrated in FIG.1B, the system is capable of controlling thetemperature ^^of the fluid at an adjustment location 7 along the network of supplyconduits, where the adjustment location is located at, or close to, the distributionsource 3. The distribution source 3 receives return fluid from the recipients 4 via thenetwork of return conduits 6. It will be appreciated that the adjustment location doesnot need to be at or proximal to the distribution source 3. For example, in some embodiments, the fluid transport system may include a mixing node where fluid flows of respective temperatures can be mixed so as to control the temperature of the resulting combined flow. Such a mixing node may thus serve as another example of an adjustment location. Generally, temperature control is an important task of district energy grid operators. Typically, grid operators are bound by a minimum supply temperature requirement andseek to provide a specific temperature at some distal downstream location 8 in thenetwork, downstream from the adjustment location 7. In the example of FIG.1B, thedownstream location 8 is a location along the supply conduit 5 at or close to a zoneboundary 9. However, it will be appreciated that the downstream location may bepositioned at any location along the supply conduit, downstream from the adjustmentlocation, where temperature control is desired and / or where knowledge of the effect oftemperature changes caused at the adjustment location is desired. Typically, thedownstream location may be selected to be at some point of the network of conduitsthat is relevant for the performance of the fluid transport system, e.g. for the heating performance at the recipients 4 of the district heating grid. Often, the downstreamlocation 8 is several kilometers away from the district heating plant 3, where thetemperature of the water fed into the network can be adjusted. Generally, in some embodiments, the distance (measured as the length of the shortest path along the oneor more conduits) between the adjustment location and the downstream location maybe rather large, e.g. at least 1 km or several kilometers. In some embodiments, a fluidtransport system may include more than one downstream location and / or more thanone adjustment location. Generally, due to the dynamics of fluid flow, particularly time-varying transport delays,temperature control in a fluid transport system is a nontrivial control task. In order tofacilitate control the temperature at, or near, the downstream location, it is desirable to have knowledge of the actual propagation delay between the time at which a change ofthe temperature is caused at the adjustment location and the time at which such achange causes a change of the temperature at the downstream location.To facilitate determination of the propagation delay, the fluid transport systemcomprises a data processing system 10 that computes an estimated propagation delaybetween an adjustment of the temperature ^^ at the adjustment location 7 and aresulting change of the temperature ^^ at the downstream location 8. The dataprocessing system 10 may be a computer-implemented control system, a PLC-basedsystem, a suitably programmed computer, and / or the like. The data processing system10 may implement a SCADA system or another suitable type of control and / ormonitoring function. In the example of FIG.1B, the data processing system 10 isillustrated as a block separate from the distribution source 3. However, it will be appreciated that the data processing system 10 may be implemented as an integral part of the distribution source 3. In particular, the data processing system 10 may beintegrated into, or communicatively coupled to, the temperature control system of adistrict heating plant to cause the temperature control system to affect the requiredtemperature change at the adjustment location 7. The data processing system 10 maye.g. be communicatively coupled to a control system of the distribution source 3 by awired or wireless connection or otherwise. In some embodiments, the data processingsystem may be embodied as a pump controller of a pump of the fluid transport system, or otherwise integrated into another device of the fluid transport system that has computational capabilities.The data processing system 10 may be programmed to perform an embodiment of amethod for determining the propagation delay associated with the fluid transport system, e.g. one of the embodiments described below, or otherwise. Additionally, thedata processing system 10 may implement an embodiment of the control methoddisclosed herein for controlling the temperature at the adjustment location 7.Generally, various embodiments of the method disclosed herein can determine propagation delay robustly, even when the propagation delay varies over time. This is animportant improvement, as it allows proper tuning of the temperature controller withrespect to the estimated delay. Embodiments of the method disclosed herein avoidproblems of tracking references, if delays are shorter than a predetermined nominaldelay, or of becoming unstable, if delays are longer than a predetermined nominaldelay.Various embodiments of the process disclosed herein only require knowledge about anactual temperature ^^ at the downstream location 8 and about an actual temperature^^at the adjustment location 7, while they do not require information about flow rates, pipe geometry, propagation delays, temperature distributions along the conduits, etc. To this end, the fluid transport system 1 may include a temperature sensor 80configured to measure the temperature ^^ of the fluid in the supply conduit 5 at thedownstream location 8, e.g. at a location where it is desirable to control thetemperature of the fluid. Optionally, the fluid transport system 1 may further include aninlet temperature sensor 70 configured to measure the temperature ^^ of the fluid inthe supply conduit 5 at the adjustment location 7, i.e. at a location where a change of the temperature can be affected. However, the temperature adjustment affected by thedistribution source 3 at the adjustment location 7 typically occurs without significantdelay and is sufficiently accurate so that an additional inlet temperature sensor for measuring the actual temperature ^^at the adjustment location may not be required. Instead, the data processing system 10 may receive information about the actual temperature at the adjustment location from the distribution source, e.g. from a temperature controller of the distribution source.The data processing system 10 may be communicatively coupled to the temperaturesensor 80 and, optionally, to the inlet temperature sensor 70, e.g. directly or indirectly,for example via a SCADA or control system of the distribution source 3. Thecommunication between the temperature sensor(s) and the data processing system 10 may be wired or wireless. FIG.2 schematically illustrates an example of a process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. The process is implemented by a data processing system 10, for example the dataprocessing system of the fluid transport system of FIGs.1A-B, or otherwise. The processdetermines the propagation delay of a property of a fluid that is transported through a network of one or more conduits 5 of a fluid transport system.A known problem in fluid transport is the presence of propagation delays, which dependon the details of the network of conduits 5 through which the fluid is being transported between the adjustment location and the downstream location. For example, if it is assumed that the network of conduits 5 consists of a set of cylindrical pipes, the propagation delay will be a function of both the flow in the pipes, the cross-sectional area, and length of the pipes. The network of conduits 5 may be rather complex, as the geometry of the pipes may change along the network, and the fluid flow may branch offin a tree structure. Therefore, based on the pipe geometries and physical structure ofthe network of conduits, it is often difficult if not impossible to calculate or make flow models based on the pipe geometry that accurately represent the fluid flow such thatthe transport delays in these systems can be computed from flow models.The present disclosure provides a method that estimates the propagation delayassociated with the propagation of a fluid property using only information of the actual,e.g. measured, fluid property at two locations, namely the adjustment location and the downstream location, which is fluidly connected to the adjustment location by the network of conduits 5. Accordingly, the adjustment location may be considered the inlet to the network of conduits 5, and the downstream location may be considered the outlet of the network of conduits 5. In FIG.2, the fluid property at the adjustmentlocation 7 will also be referred to as input ^ to the network of one or more conduits 5,while the fluid property at the downstream location will also be referred to as output ^of the network of one or more conduits 5. The network of one or more conduits 5 maythus simply be considered as an unknown system which imparts and a priori unknownpropagation delay. The network of one or more conduits 5 may also be considered as imparting a gain to the fluid property, and various embodiments of the method disclosed herein further estimate a corresponding gain factor of the network of one or more conduits 5. Various embodiments of the process include two main stages: (1) A modelling stage 110 comprising one or more property propagation modelsconfigured to model the behaviour of the system, i.e. of the one or moreconduits 5, based on input-output data.(2) A model selection stage 120, comprising selection of the model that best, e.g.with maximum likelihood, represents the system behaviour.The property propagation models serve as a bank of candidate models from which theprocess selects one or more models that best (e.g. with maximum likelihood) representthe system behaviour. Each of the property propagation models are associated with a respective propagation delay and the process determines an estimate of the propagation delay of the system from the selected one or more models, selected from the bank of candidate models. In particular, the model selection stage may select one of the candidate models and output the propagation delay associated with the selected model as an estimate of the propagation delay of the system. The property propagation models may be adaptive. In particular, they may be (orinclude) adaptive filters used to fit the behaviour of the system based on the input-output data. Instead of adaptive models, the process may utilize non-adaptive models.However, embodiments based on non-adaptive models may require a larger set of candidate models, e.g. so as to reflect different choices of model parameters, e.g.different gain parameters, which would otherwise be adaptively adjusted in an adaptivemodel.The input and output data may be represented as a time series of input values ^(^) andoutput values ^(^), respectively, where k represents discrete time. For example, whenthe fluid property is temperature, the input and output values may be temperature values. In particular, in the transport system of FIGs.1A-B, the input may be the temperature at the adjustment location 7 and the output may be the temperature atthe downstream location 8, i.e. ^ = ^^ and ^ = ^^.In the following, examples of the modelling stage 110 and of the model selection stage 120 will be described in more detail. FIG.3 schematically illustrates a more detailed view of an example of a process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. The process is implemented by a data processing system 10 and determines the propagation delay of a property of a fluid that is transported through a network of one or more conduits 5 of a fluid transport system, all as described in connection with FIG.2. To this end, the process includes two main stages, namely a modelling stage 110 and a model selection stage 120, also as described above in connection with FIG.2. In the embodiment of FIG.3, the property propagation models are a bank of adaptivefilters. The modelling stage 110 comprises a filter calculation block 111, an errorcalculation block 112 and a filter update block 113. The model selection stage 120 comprises a model selection block 121 and a delay computation block 122. At the modelling stage 110, the process calculates and updates multiple filters, i.e. afilter bank. Each filter represents the input / output behaviour between the input ^ andthe output ^ for a given propagation delay. For example, with ^ being the current timestep, the filter ^^(^) = ^(^ − ^)^^ , selects the input collected ^ samples ago and scalesit by a weight ^^ to model the current output ^(^). To do this, the process may storethe input history of the scalar input ^(^) up to the length of the filter bank. At the errorcalculation block 112, the process calculates the errors between the current measuredoutput and the outputs of the respective filters. Based on these errors, at the filterupdated block 113, the process updates the weights ^^of the filters such that the filtersfit the output y(k) as well as possible. A preferred way of doing this is to select a function^(⋅) so the weights are updated by minimizing the error in a least square sense, forexample ^^^^(^)^ = ^^(^) ⋅ ^^^^(^^(^)). Here, ^(^) denotes the stored input history.At the model selection block 121, the process calculates probabilities of the errors byBayesian model selection. The probability of an error at a given delay, i.e. ^(^^|^^), maybe calculated under the assumption that the errors are independent and identicallydistributed random variables with a suitable assumed probability density function, e.g.Gaussian noise with a given noise variance. At the delay computation block 122, theprocess selects an updated propagation delay associated with the next timestep,denoted as ^^(^ + 1), as the delay associated with the maximum probability.In the present and other embodiments, the filter bank that is calculated and updatedduring the modelling stage 110 comprises a set of adaptive FIR filters. The adaptive FIR filter vector structure of the present embodiment corresponds to a set of parallel coupled pipe models with differing delays. However, in other embodiments other filter structures can be used. The filter structure of the present example may be related to the network of conduits 5and, in particular, to the system of FIGs.1A-B as follows: The system outputtemperature ^^ in the fluid transport system of FIGs.1A-B may, in the case of a districtheating system, be considered as being bounded from below by the ambienttemperature ^^, which is the temperature surrounding the pipes in the transport network, and by the input temperature ^^from above: ^(^) is the gain factor of the pipe network and ^(^) is the delay through the pipenetwork, both at time ^. The model corresponds to the following FIR filter: The above relations may e.g. be obtained by observing that the spatiotemporaltemperature profile of primarily advective fluid flow in a cylindrical pipe has the analytical solution: Where ^^is the ambient (ground for buried pipes) temperature, ^^^is the temperatureat the pipe inlet, ^^ is a loss parameter, and ^(^) is the time-varying pipe transportdelay given by the pipe length, circumference, and time-varying flow rate. Assuming ^^nonnegative it can be shown that this function is bounded from below by ^^(^) andbounded from above by ^^^(^). Analogous considerations apply to a cooling system. Forexample, it will be appreciated that the loss parameter ^^will be negative in a cooling system.Assuming that the loss parameter and delay are both unknown, the gain factor ^(^) andthe exponent ^(^) are both unknown. A set of linear adaptive filters may neverthelessbe constructed by first defining a vector of n possible coefficients (gain factors): ^(^) = [^^(^) ^^(^) … ^^(^)]^(3) and then defining the input vector: ^(^) = [^^(^) ^^(^ − 1) … ^^(^ − ^ + 1)]^(4)By elementwise scalar multiplication of ^ and ^ a filter output vector may be obtained:^(^) = [^^(^)^^(^) ^^(^)^^(^ − 1) … ^^(^)^^(^ − ^ + 1)]^(5) This corresponds to a vector of potential system models with each successive element inthe vector corresponding to a successively larger delay and individual gain factors. Thegain factors may be updated using least mean squares updating as ^^(^ + 1) = ^^(^) + ^(^^) (6)with ^(^) being a suitable update function.At the model selection stage 120, the delay estimate can be obtained by updating the delay probability function with a Bayesian update law, e.g. as described in the following, or otherwise:For the purpose of model selection, it may be assumed that the measurements from thesystem follow a suitable measurement model, e.g. a model of the type: ^(^) = ^(^) + ^(^) (7)Where ^(^) is the measured output, ^(^) is the true output, and ^(^)~^(0, ^^) isi.i.d Gaussian noise. The model error for each individual model in equation 5 is then: ^^(^) = ^(^) − ^^(^)^^(^ − ^ + 1) (8)= ^(^) + ^(^) − ^^(^)^^(^ − ^ + 1),which implies that, if the delay and gain estimates ^^(^), ^^(^) are equal to their truevalues, one obtains ^(^) = ^(^), which in turn implies that: Using the assumption that residuals are mutually independent, a most likely candidate model may then be selected by using standard Bayesian model selection, i.e. the most likely sequence of model errors can be computed conditional on the error distribution in equation 8, using: Where ^th^is the icandidate delay and ^^(^) is the error sequence of the ithmodel up totime ^. It may be worthwhile noting that, due to its multiplicative nature, the probabilityestimates may, in some embodiments, be periodically reset to avoid that the individualmodel probabilities dissipate.The delay estimate at time ^ may e.g. be determined as: where ^∗ is the maximizer of equation (10) and Δ^ is the sampling interval.The above describes one preferred implementation of the delay estimator. However, itwill be appreciated that various modifications may be made. In particular, variousadditions may be made to improve the robustness of the estimation process. Forexample, the assumed probability density function for the error may be selected to account for the adaptation process, for example centrally dense but non-normal. Also, excitation of the input signal may be detected such that the filter is only updated if theinput signal has sufficient variation (e.g. only when the input signal is persistentlyexciting or fulfils another suitable excitation criterion). On the other hand, in steadyperiods, where the input signal is not sufficiently rich to update the filter, the noisevariance of the input signal may advantageously be estimated. The result being that the noise variance used in the Bayesian model selection (in particular in the probability density function of the nominator of equation (10)) gives better probability estimates of the error for a given delay. The input signal noise variance may for example be calculated using recursive filtering or cross correlation analysis of the input signal history ^(^), that is already stored in the above embodiment. This input history is also well- suited for calculating the excitation of the input signal, for example as suggested by: R. Bitmead, "Persistence of excitation conditions and the convergence of adaptiveschemes," in IEEE Transactions on Information Theory, vol.30, no.2, pp.183-191, March1984, doi: 10.1109 / TIT.1984.1056898Some embodiments of the process, including some or all of the above modifications, willnow be described with reference to FIGs.4-6. In particular, FIG.4 schematically illustrates a flow diagram of an example of a process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. In initial step S1, the process is initialized. This step may include the setting of one or more constants and / or initial values of model parameters. At step S2, the process receives the measured input values, e.g. the temperature at the adjustment location, and determines one or more excitation properties of the input signal. At step S3, the process determines whether the calculated one or more excitation properties fulfil a predetermined excitation criterion, e.g. whether the input signal is persistently exciting. If the excitation criterion is fulfilled, the process proceeds at steps S4 and S5 which implement the modelling stage and the model selection stage, respectively, e.g. asdescribed in connection with FIGs.2 and / or 3 above, or otherwise. In step S4, theadaptive filters are updated and in step S5, a model is selected. If the excitation criterion is not fulfilled, the process proceeds to step S6 and determines whether the process has a previously stored updated estimate of the noise variance available, e.g. by determining whether the time since the previous computation of an estimated noise variance exceeds a predetermined threshold, or otherwise. If the stored estimate of the noise variance is sufficiently up-to-date, the process returns to step S3; otherwise the process proceeds at step S7 to compute an estimate of the noise variance and then returns to step S3. It will be appreciated that the test of step S6 may beomitted, and an updated noise variance may be computed whenever the input signalreflects a steady situation. However, the test of S6 may be useful if estimation of thenoise variance is done using numerically expensive computations such as cross correlation calculation. The determination of the excitation properties may be performed such that the processdetects whether the input signal is stationary / steady or varies over time, other thanstatistical variance due to noise. For example, the process may calculate the gradient of the stored input signal history to distinguish a steady period in the input signal (e.g. corresponding to a sufficiently small gradient) from a period with sufficient signal variability for performing the model update (e.g. corresponding to a sufficiently large gradient). When the process detects a steadyperiod in the input signal, the process may cut out a steady period of the input signalhistory, e.g. a period between consecutive gradient changes above a predefinedthreshold, e.g. the longest period in the input history where gradient changes are belowthe predefined threshold. The process may then subtract the mean of the steady periodfrom the steady period so as to obtain a zero-mean steady period. The process may then calculate the autocorrelation of the steady period, i.e. the cross correlation of the signalwith itself. The process may then calculate an actual confidence interval for apreselected confidence level, i.e. the interval is a function of the number of samples inthe steady period. The process may further check that the number of outliers outsidethe confidence interval is less than the confidence level. If so, the process may update the estimate of the variance. It will be appreciated that the computation of the signal variance may be performed in adifferent manner. For example, in an alternative embodiment, the process may computethe variance of the input signal history ^(^) and low-pass filter the computed variance,e.g. as V(k) = ^*Var(U(k)) + (1-^)V(k-1), where ^ is a suitable constant defining theresponse of the low-pass filter and where Var() is a function is a function that computesthe variance of the input signals included in the history U(k) at time k. Hence, V(k) is thelow-pass filtered variance of U(k). FIG.5 schematically illustrates a more detailed flow diagram of an example of a process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. The example of FIG.5 is similar to the example of FIG.4, except that embodiments of the modelling stage S4 and the model selection stage S5 are illustrated in more detail. In particular, the modelling stage S4 of this example includes a filter calculation step S41, an error calculation step S42 and a filter update step S43, e.g. as described in connection with the corresponding blocks 111-113, respectively, of FIG.3. The model selection stage S5 comprises a model selection step S51 and a delay computation step S52, e.g. as described in connection with the corresponding blocks 121 and 122, respectively, of FIG. 3. FIG.6 schematically illustrates a flow diagram of yet another example of process for determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system. The example of FIG.6 is similar to the example of FIG.5, except that the modelling stage S5 of the process of FIG.6 includes an additional conditional reset of the filter bank. In particular, at additional step S44, before proceeding to the model selection stage S5, the process determines whether the filters of the filter bank have degenerated. If so, the process proceeds at step S45 to reset the filter bank before returning to step S3. If the filter bank is not degenerated, the process proceeds at the model selection stage S5 as described above. A degeneration of the filter bank may e.g. occur due to the multiplication in the Bayesian model selection. The detection of a degeneration may e.g. be performed bycalculating the sum of probabilities not associated to the most probable tap in the filterand resetting the filter if this sum is sufficiently small. That is, if ^ is the most probabletap in the filter of length ^^^^, the filter is reset if: In fluid transport systems, the flow and therefore the transport delay is not expected toexperience large jumps. Therefore, the filter reset is preferable done so the most probable filter tap is maintained. This can for example be done by distributing the tap probabilities according to a Cauchy Distribution, a Normal Distribution, or any other centrally dense distribution and maintaining the most probable tap after the reset. It is worthwhile to note that the embodiments disclosed above also provide estimates of the gain parameter, e.g. as the gain parameter of the selected model. Accordingly, the above embodiments provide a process for the co-estimation of loss and time delays in fluid transport systems dominated by advective flow. Various embodiments of the process disclosed herein only require sensors at two points (or otherwise knowledge of the actual fluid property at these points) and require no information about pipe geometry or flow rate. Some embodiments utilize an adaptive filter in vector form and aFIR filter representation of the pipe that is derivable from an analytical solution to thetransport equation. Generally, various embodiments of the method disclosed herein provide an estimate of the propagation delay in a fluid transport system based on two measurements, wherein the process comprises: ^Calculation of a model that represents a fluid property or other fluid streamvariable that travels by advection. ^Recursively updating the model.^ Calculating the errors between the output measurement and the estimates frommultiple models. ^Selecting the model that fits the measurements best or is the most likely model.^ Selecting the delay associated with the best or most likely model.^ Optionally, detecting excitation properties of the input signal describingwhether the input signal is suitable for model estimation (e.g. being persistently exciting). Optionally, the model calculation is done only if the input signal ispersistently exciting. ^Optionally, using a calculated variance of the input signal for selection of thebest or the most probable model and / or deeming the input signal persistently exciting or not. Optionally, the variance of the input signal is calculated only inperiods where the model is not updated. Optionally, the variance is onlycalculated for a part of the stored input history, that under a given outlier percentage can be considered independent and identically distributed random variables. It will be appreciated that a number of modifications may be made to the process and system described herein. While the various aspects disclosed herein have mainly been described in the context of a district heating system, it will be appreciated that they may also be applied to othertypes of fluid transport systems. In particular, the fluid transport system may not necessarily transport fluid between a single source location and multiple recipients or other destination locations. Instead, the fluid transport system may include multiple source locations and / or only a single destination location. Similarly, the adjustment location does not need to be proximal to a distribution source, but may be located at any node or other location of a network of conduits or at a location along a single conduit. For example, the adjustment location may be located at a mixing node where two flow paths can be joined, e.g. so as to mix fluid flows of respective temperature soas to adjust the outgoing flow from the mixing node. Similarly, the downstream locationdoes not need to be proximal to a recipient or destination location, as long as thedownstream location is downstream from the adjustment location. Similarly, while thevarious aspects disclosed herein have mainly been described in the context of temperature control, it will be appreciated that other embodiments may be used to determine propagation delays associated with another fluid property that propagates along the conduit with the fluid flow. Some embodiments of the various aspects disclosed herein may be summarized as follows: Embodiment 1: A method of determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system, the fluid transport system comprising one or more conduits for transporting the fluid from at least the adjustment location to at least the downstream location, wherein a change of said fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or more conduits, wherein the method comprises:- providing a plurality of property propagation models, each property propagationmodel being configured for modeling propagation of the fluid property between the adjustment location and the downstream location, each property propagation model being based on a respective candidate propagation delay,- evaluating a performance of each of the plurality of property propagation models,- selecting one or more of the plurality of property propagation models based on theevaluated performance;- determining an estimated propagation delay from the one or more candidatepropagation delays of the selected one or more property propagation models. Embodiment 2: The method according to embodiment 1, wherein evaluating the performance of each of the plurality of property propagation models comprises using said property propagation model to compute one or more estimated values of the fluid property at the downstream location and calculating one or more errors between respective ones of said one or more estimated values and corresponding one or more obtained actual values of the fluid property at the downstream location; and wherein selecting one of the plurality of property propagation models is based on the one or more calculated errors. Embodiment 3: The method according to any one of the preceding embodiments, wherein the plurality of property propagation models includes one or more adaptive models and wherein the method further comprises recursively updating the one or more adaptive models based on obtained actual values of the fluid property at the adjustment location and / or on obtained actual values of the fluid property at the downstream location. Embodiment 4: The method according to embodiment 3, wherein the one or more adaptive models include one or more adaptive filters, in particular linear adaptive filters. Embodiment 5: The method according to any one of embodiments 3 through 4, wherein each adaptive model comprises one or more adaptable model parameters, and wherein recursively updating comprises recursively updating the one or more adaptable model parameters of the respective adaptive models. Embodiment 6: The method according to any one of embodiments 3 through 5, wherein the candidate propagation delays of the respective property propagation models are respective model-specific, non-adaptable model attributes of the respective adaptive models. Embodiment 7: The method according to any one of embodiments 3 through 6, further comprising detecting one or more excitation properties of an input signal, the input signal being indicative of the actual value of the fluid property at the adjustment location; and wherein recursively updating the one or more adaptive models comprises selectively updating the one or more adaptive models only when the detected one or more excitation properties fulfil a predetermined excitation condition, in particular only when the input signal is persistently excited. Embodiment 8: The method according to any one of embodiments 3 through 8, further comprising resetting the one or more of the adaptive models responsive to a predetermined reset criterion. Embodiment 9: The method according to any one of the preceding embodiments, wherein selecting one or more of the plurality of property propagation models comprises selecting the property propagation model that fits the measurements best or that is the most likely model. Embodiment 10: The method according to any one of the preceding embodiments, further comprising computing an input signal noise variance of an input signal, the input signal being indicative of the actual value of the fluid property at the adjustment location; wherein evaluating the performance is based at least in part on the computed input signal noise variance. Embodiment 11: The method according to embodiment 10, comprising detecting one or more excitation properties of the input signal, and selectively computing the input signal noise variance only when the detected one or more excitation properties fail to fulfil a predetermined excitation condition, in particular only when the input signal is not persistently excited. Embodiment 12: The method according to embodiment 10 or11, comprising:- recording the input signal over a period of time to obtain an input signal history,- selecting a part of the obtained input signal history for which the recorded input signalvalues represent random variables having a distribution that fulfils one or more predetermined criteria,- selectively calculating the input signal noise variance for only the selected part of theobtained input signal history. Embodiment 13: The method according to embodiment 12, wherein selecting comprises selecting a part of the obtained input signal history for which the recorded input signal values represent independent and identically distributed random variables, optionally with the exception of a predetermined maximum amount of outliers. Embodiment 14: The method according to any one of the preceding embodiments, wherein the fluid property is a fluid temperature. Embodiment 15: The method according to embodiment 14, wherein the fluid transport system is a district energy system, in particular a district heating and / or district cooling system. Embodiment 16: The method according to any one of the preceding embodiments, wherein selecting one or more of the plurality of property propagation models comprises selecting a single model of the one or more property propagation models based on the evaluated performance; and wherein determining the estimated propagation delay comprises selecting the candidate propagation delay of the selected property propagation model as the estimated propagation delay.Embodiment 17: A method of controlling a fluid property at a downstream location of afluid transport system by adjusting the fluid property at an adjustment location, the downstream location being downstream from the adjustment location, the fluid transport system comprising one or more conduits for transporting the fluid from at least the adjustment location to at least the downstream location, wherein a change of said fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or more conduits, wherein the method comprises:- performing the acts of the method according to any one of the precedingembodiments to determine an estimated propagation delay of the fluid property between the adjustment location and the downstream location, -controlling adjustment of the fluid property at the adjustment location based at least in part on the estimated propagation delay.Embodiment 18: A computer program configured to cause a data processing system,when executed by said data processing system, to perform the steps of the method according to any one of the preceding embodiments.Embodiment 19: A data processing system configured to perform the steps of themethod according to any one of embodiments 1 through 17.Embodiment 20: A control system for controlling operation of a fluid transport system,the control system comprising a data processing system according to embodiment 19. Embodiment 21: The control system according to embodiment 20, comprising a first sensor for acquiring actual values of said fluid property at the downstream location, and, optionally, a second sensor for acquiring actual values of the fluid property at the adjustment.Embodiment 22: A fluid transport system comprising a data processing system asdefined in embodiment 19 and / or a control system as defined in any one of embodiments 20 through 21. Various embodiments of the method described herein may be computer-implemented. In particular, embodiments of the method may be implemented by means of hardware comprising several distinct elements, and / or at least in part by means of a suitably programmed data processing system. In the apparatus claims enumerating several means, several of these means can be embodied by one and the same element, component or item of hardware. The mere fact that certain measures are recited in mutually different dependent claims or described in different embodiments does not indicate that a combination of these measures cannot be used to advantage. It should be emphasized that the term "comprises / comprising" when used in this specification is taken to specify the presence of stated features, elements, steps or components but does not preclude the presence or addition of one or more other features, elements, steps, components or groups thereof.

Claims

CLAIMS 1. A method of determining a propagation delay of an advective propagation of a fluid property between an adjustment location and a downstream location of a fluid transport system, the fluid transport system comprising one or more conduits for transporting the fluid from at least the adjustment location to at least the downstream location, wherein a change of said fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or more conduits, wherein the method comprises:- providing a plurality of property propagation models, each property propagationmodel being configured for modeling propagation of the fluid property between theadjustment location and the downstream location, each property propagation modelbeing based on a respective candidate propagation delay,- evaluating a performance of each of the plurality of property propagation models,- selecting one or more of the plurality of property propagation models based on theevaluated performance;- determining an estimated propagation delay from the one or more candidatepropagation delays of the selected one or more property propagation models.

2. The method according to claim 1, wherein evaluating the performance of each of theplurality of property propagation models comprises using said property propagationmodel to compute one or more estimated values of the fluid property at thedownstream location and calculating one or more errors between respective ones ofsaid one or more estimated values and corresponding one or more obtained actual values of the fluid property at the downstream location; and wherein selecting one of the plurality of property propagation models is based on the one or more calculated errors.

3. The method according to any one of the preceding claims, wherein the plurality of property propagation models includes one or more adaptive models and wherein the method further comprises recursively updating the one or more adaptive models based on obtained actual values of the fluid property at the adjustment location and / or on obtained actual values of the fluid property at the downstream location.

4. The method according to claim 3, wherein the one or more adaptive models include one or more adaptive filters, in particular linear adaptive filters.

5. The method according to claim 4, wherein the one or more adaptive models include one or more adaptive FIR filters.

6. The method according to claim 4 or 5, wherein the plurality of property propagationmodels includes a plurality of adaptive filters, each relating observed values of the fluid property at the adjustment location at an earlier time to corresponding values of the fluid property at the downstream location at a current time, wherein the earlier time and the current time differ by a respective candidate propagation delay.

7. The method according to any one of claims 3 through 6, wherein each adaptive modelcomprises one or more adaptable model parameters, and wherein recursively updating comprises recursively updating the one or more adaptable model parameters of the respective adaptive models.

8. The method according to any one of claims 3 through 7, wherein the candidate propagation delays of the respective property propagation models are respective model-specific, non-adaptable model attributes of the respective adaptive models.

9. The method according to any one of claims 3 through 8, further comprising detecting one or more excitation properties of an input signal, the input signal being indicative of the actual value of the fluid property at the adjustment location; and wherein recursively updating the one or more adaptive models comprises selectively updating the one or more adaptive models only when the detected one or more excitation properties fulfil a predetermined excitation condition, in particular only when the input signal is persistently excited.

10. The method according to any one of claims 3 through 9, further comprising resetting the one or more of the adaptive models responsive to a predetermined reset criterion.

11. The method according to any one of the preceding claims, wherein the plurality ofproperty propagation models only receive information of the actual fluid property at the adjustment location and at the downstream location and require no information about pipe geometry or flow rate.

12. The method according to any one of the preceding claims, wherein selecting one ormore of the plurality of property propagation models comprises selecting the property propagation model that fits the measurements best or that is the most likely model.

13. The method according to claim 12, wherein selecting one or more of the plurality ofproperty propagation models is based on a statistical determination of which property propagation model provides a series of estimated fluid property values that with a maximum likelihood corresponds to an expected series of actual values, where the expected series of actual values depends on a predetermined assumption of the noise properties of the system.

14. The method according to any one of the preceding claims, wherein transporting the fluid from at least the adjustment location to at least the downstream location occurs at a variable and / or unknown flow rate.

15. The method according to any one of the preceding claims, further comprisingcomputing an input signal noise variance of an input signal, the input signal beingindicative of the actual value of the fluid property at the adjustment location; wherein evaluating the performance is based at least in part on the computed input signal noise variance.

16. The method according to claim 15, comprising detecting one or more excitation properties of the input signal, and selectively computing the input signal noise variance only when the detected one or more excitation properties fail to fulfil a predetermined excitation condition, in particular only when the input signal is not persistently excited.

17. The method according to claim 15 or 16, comprising:- recording the input signal over a period of time to obtain an input signal history,- selecting a part of the obtained input signal history for which the recorded input signalvalues represent random variables having a distribution that fulfils one or more predetermined criteria,- selectively calculating the input signal noise variance for only the selected part of theobtained input signal history.

18. The method according to claim 17, wherein selecting comprises selecting a part of the obtained input signal history for which the recorded input signal values represent independent and identically distributed random variables, optionally with the exception of a predetermined maximum amount of outliers.

19. The method according to any one of the preceding claims, wherein the fluid property is a fluid temperature.

20. The method according to claim 19, wherein the fluid transport system is a district energy system, in particular a district heating and / or district cooling system.

21. The method according to any one of the preceding claims, wherein selecting one or more of the plurality of property propagation models comprises selecting a single model of the one or more property propagation models based on the evaluated performance; and wherein determining the estimated propagation delay comprises selecting the candidate propagation delay of the selected property propagation model as the estimated propagation delay.

22. A method of controlling a fluid property at a downstream location of a fluid transport system by adjusting the fluid property at an adjustment location, the downstream location being downstream from the adjustment location, the fluid transport system comprising one or more conduits for transporting the fluid from at least the adjustment location to at least the downstream location, wherein a change of said fluid property occurring at the adjustment location propagates along the one or more conduits in dependence of the fluid flow along the one or more conduits, wherein the method comprises:- performing the acts of the method according to any one of the preceding claims todetermine an estimated propagation delay of the fluid property between the adjustment location and the downstream location, -controlling adjustment of the fluid property at the adjustment location based at least in part on the estimated propagation delay.

23. A computer program configured to cause a data processing system, when executedby said data processing system, to perform the steps of the method according to any one of the preceding claims.

24. A data processing system configured to perform the steps of the method accordingto any one of claims 1 through 22.

25. A control system for controlling operation of a fluid transport system, the controlsystem comprising a data processing system according to claim 24.

26. The control system according to claim 25, comprising a first sensor for acquiring actual values of said fluid property at the downstream location, and, optionally, a second sensor for acquiring actual values of the fluid property at the adjustment.

27. A fluid transport system comprising a data processing system as defined in claim 19and / or a control system as defined in any one of claims 25 through 26.

Images