CN116557992A

CN116557992A - Method for predicting cold energy demand in cold supply pipe network

Info

Publication number: CN116557992A
Application number: CN202210105548.7A
Authority: CN
Inventors: 郭静波; 弗朗索斯·考尔托特; 孙晔琦
Original assignee: Electricite de France SA
Current assignee: Electricite de France SA
Priority date: 2022-01-28
Filing date: 2022-01-28
Publication date: 2023-08-08

Abstract

The invention relates to a method for predicting the cooling capacity demand in a cooling pipe network, comprising: -establishing an initial library comprising numerical models of thermal properties of the building, each model corresponding to a respective building in the cooling network and comprising a limited number of parameters and being adapted to calculate the cooling demand of the respective building; -calibrating the parameters using the historically measured cold demand and weather data as boundary conditions, thereby obtaining a parameterized numerical model; -simulating the parameterized numerical model during a training period to obtain a simulated historical cold demand time series; -obtaining a set of fixed weights, the weighted sum of the time series of historical cold demands matching the cold demands measured during the training period, wherein the set of fixed weights is obtained by a statistical method; -simulating the parameterized model during a prediction period, with the weather forecast data as boundary conditions, obtaining a predicted cold demand time sequence; and-applying a fixed weight to the predicted cold demand time series, obtaining a cold demand prediction by a weighted sum of the predicted cold demand time series. The invention improves the prediction precision and efficiency of the cold energy demand.

Description

Method for predicting cold energy demand in cold supply pipe network

Technical Field

The invention relates to the field of regional cooling station optimal control, in particular to prediction of regional cooling water flow and return water temperature, which is used for optimizing production, transportation and distribution of cooling water in the operation of a cooling system.

Background

During operation of the district cooling project, the amount of chilled water produced is based on the customer's cooling capacity demand. Depending on the cooling demand, the district cooling system may establish a control strategy and thereby set relevant plant operating parameters.

In the district cooling project, the core equipment is refrigeration equipment, typically chillers. By consuming electricity or heat (steam, hot water or flue gas), the refrigeration device may produce chilled water through a refrigeration cycle. Large refrigeration systems, including district refrigeration systems, are often equipped with energy storage devices, sometimes also utilizing the thermal inertia of the pipe network as an energy storage means. The performance and operating costs of a refrigeration device depend on the load factor of the device, external conditions and energy prices. Knowing the customer's cooling needs in advance allows the plant operator to optimally arrange the operation of each refrigeration and energy storage device to minimize operating costs.

Thus, accurate refrigeration demand predictions are critical to the operation of regional refrigeration stations. At the same time, chilled water is transported through the cooling network, so that the cooling network always reacts later than the adjustment of the operating parameters. In summary, timely and accurate cold demand prediction is a precondition for regional refrigeration station operation optimization.

In the prior art, the cooling capacity demand in the cooling pipe network is mainly predicted by the following method:

1. experience/manual method

Mainly used by operators, which are based on empirical curves (relationship between outdoor temperature and chilled water temperature), this method relies on the experience of the operators, often with insufficient accuracy.

Another solution is to directly repeat the demand profile of the previous day (or to correct it according to weather conditions), which is also often not accurate enough, which reduces the refrigeration production efficiency and affects the operating costs.

2. Statistical/machine learning method

This is currently the vast majority of existing commercial solutions. These methods can give accurate results based on operational data under similar conditions in the past, and typically a good machine learning requires at least one complete cold season of operational data, which is typically not available for newly built regional cold projects. Even for regional cooling stations that have been operated for more than 3 years, the cooling demand is changing due to the continuous connection of new customers, and what the machine has learned is not applicable to new conditions.

3. Physical model method

This method is rarely used today and relies on building thermal performance numerical models to simulate the cold demand. The main reason that this approach is not popular is that it requires detailed information about the building envelope, the cold distribution equipment and the usage situation, in particular information about the use of refrigeration and the behaviour of the user, which has a great influence on the cold demand.

Thus, prior art methods and tools for cold demand prediction do not provide timely and accurate results. This results in the area cooling system not being able to operate in an optimal condition without accurate prediction.

Disclosure of Invention

It is noted that the present invention aims to overcome the drawbacks and problems of the prior art.

To this end, according to one aspect of the invention, the invention proposes a method for predicting the cooling demand in a cooling pipe network, comprising:

-establishing an initial library comprising a set of building thermal performance numerical models comprising a plurality of models, each model corresponding to a respective building in the cooling network, and each numerical model comprising a limited number of parameters and being adapted to calculate the cooling demand of the respective building;

-calibrating said parameters in each of said set of building thermal performance numerical models using historical cold demand measurements and historical weather data of the respective building as boundary conditions, thereby obtaining a set of parameterized numerical models;

-simulating the set of parameterized numerical models during a preset training period, wherein the training period is terminated at a current time to obtain a set of simulated historical cold demand time sequences corresponding to each numerical model of the set of parameterized numerical models, respectively;

-obtaining a set of fixed weights respectively assigned to each of said simulated historical cooling demand time series, such that a weighted sum of said set of simulated historical cooling demand time series matches the cooling demand measured during said training period, wherein the set of fixed weights is obtained by a statistical method;

-simulating the set of parameterized models during a preset prediction period, wherein the prediction period starts at the current time, and weather forecast data during the prediction period is used as a boundary condition, so as to obtain a set of predicted cold demand time sequences respectively corresponding to each numerical model in the set of parameterized models; and

-applying the obtained fixed weights to the set of predicted cold demand time sequences in order to obtain a prediction of cold demand during the prediction period by a weighted sum of the set of predicted cold demand time sequences.

The method according to the invention combines a physical model and a statistical method. With the building's physical and thermal performance model, the cooling demand can be inferred under unobserved weather conditions. Furthermore, due to the use of statistical methods, the impact of user behavior can also be captured. Therefore, the invention greatly reduces more than 90% of training data required by the statistical method. By using the new method according to the invention, a good prediction of the cold demand can be achieved with only a short history of operating data, for example one week.

Optionally, the refrigeration requirement includes a total flow and/or a return water temperature of the cooling pipe network, and other physical quantities indicative of the refrigeration requirement.

Optionally, each numerical model includes a linear or nonlinear differential equation describing the heat transfer with the limited number of parameters.

Optionally, the limited number of parameters is a physical parameter of the building and/or a building usage scenario. Further preferably, the physical parameters of the building include floor area, wall area, thermal inertia, glass area, heat transfer coefficients of air handling devices, thermal resistance of walls, windows and floors, solar transmittance, and the building usage scenarios include occupancy, indoor temperature settings, ventilation times, lighting, internal heat gain of equipment and personnel, and heat gain from solar radiation.

Optionally, after calibrating the parameters of the set of building thermal performance numerical models, expanding the set of parameterized numerical models by modifying at least part of the physical parameters and the building usage scenarios to obtain a set of expanded parameterized numerical models, wherein the physical parameters are modified using random selection within a defined distribution centered on the calibrated parameters. Therefore, a model library with more models is obtained so as to cover the cold load characteristics of more buildings in the cold supply pipe network, and the prediction accuracy is improved.

Alternatively, the set of parametric numerical models can also be expanded into a set of expanded parametric numerical models based on measurements taken on buildings connected to the cooling network. Preferably, the measurements taken on the buildings connected to the cooling network are obtained by a set of sensors installed in the buildings, within a defined time, to calibrate the parameters. Advantageously, at least one of the set of sensors measures the flow rate taken by the building from the cooling network, the water supply temperature of the cooling network, the return water temperature of the cooling network, the indoor temperature at a selected location within the building, the indoor air humidity at a selected location within the building, and the occupancy rate.

Alternatively, the statistical method is linear regression.

Optionally, the training period is 3 to 10 days before the current time.

Optionally, the prediction period is 24 to 48 hours after the current time.

According to another aspect of the invention, the invention also proposes an apparatus comprising a processor configured to execute instructions stored on a computer readable medium to perform the above method.

Other aspects, features and advantages of the present invention will be discussed in the following detailed description, and the contents of the present invention, as well as the technical effects obtained, will be apparent to those skilled in the art based on the following examples.

Drawings

It is to be understood that all of the features, variations and/or specific embodiments of the invention may be combined in various combinations, except where clearly contradicted or incompatible.

Other features and advantages of the invention will be apparent from reading the following detailed description, given as non-limiting illustrations, in conjunction with the accompanying drawings, in which:

fig. 1 is a schematic diagram of one embodiment of a method for predicting the cooling demand in a cooling pipe network according to the present invention.

Detailed Description

The following are exemplary embodiments according to the present invention. The following relevant definitions are used to describe exemplary embodiments and are not intended to limit the scope of the invention. As the embodiments described herein are exemplary, they can also be extended to modifications relating to the function, purpose and/or structure of the present invention.

FIG. 1 illustrates one embodiment of a method for predicting cooling demand in a cooling pipe network. The exemplary method may include the steps of:

step 1 (in the figure (1))

An initial library is created containing a set of building thermal performance numerical models M1 based on, for example, a system of linear and nonlinear differential equations describing heat (and humidity) transfer. The set of models comprises at least one building thermal performance numerical model, and each model corresponds to a (or a class of) typical building (such as a hotel, an office building and the like) connected with a cooling pipe network. The models can timely calculate the cold energy demands of the corresponding building, such as the chilled water flow and the backwater temperature, and can also calculate the indoor temperatures of different positions of the building.

These models have a limited number of parameters, including physical parameters related to building geometry, building envelope, and cooling capacity distribution equipment, and a limited number of building usage scenarios including and typical days. Physical parameters include, for example, floor area, wall area, thermal inertia, glass area, heat transfer coefficient of the air handling unit, thermal resistance of walls, windows, and floors, solar energy transmittance. Building usage scenarios include occupancy, indoor temperature settings, ventilation times, lighting, internal heating of equipment and personnel, and heat obtained directly or indirectly from solar radiation.

More specifically, the model includes a system of equations with the above parameters, and a solver that solves the system of equations given inputs (boundary conditions): outdoor temperature, outdoor humidity, solar position and solar radiant quantity, and internal heat gain. The output of the equation set is flow and return water temperature.

Further, the equations of the model describe heat and mass (humidity) transfer in the building, including heat transfer from the building envelope, heat gain or loss from ventilation, heat gain from the interior of lighting, equipment and people, latent heat released during condensation of water in the air, and heat gain from solar radiation, including direct light entering from the glass and heat transfer from the building envelope. Different models may have different sets of equations. The following is an exemplary set of equations for one model:

Q _buildings ＝Q _envelop +Q _air +Q _{internal heat} +Q _latent +Q _solar

wherein:

Q _buildings for the purpose ofThe model corresponds to the cooling capacity requirement of a (typical) building

Q _envelope For heat transfer through building envelope

Q _air Heat exchange load for ventilation of indoor and outdoor air

Q _{internal heat} For obtaining heat from the interior, including personnel, lighting and equipment in the building

Q _latent Is the vaporization latent heat of water vapor in the air

Q _solar Is the heat of solar radiation

Meanwhile, according to a typical relation curve of the flow or backwater temperature of the chilled water and the cold quantity, the flow and backwater temperature of the chilled water are calculated according to the cold quantity demand. These typical relationships depend on the actual operating strategy of the zone cooling project.

Step 2 (2 in the figure)

Parameters in the model are calibrated to accommodate past historical measurement data for their corresponding individual buildings (there is no time period limitation in the present invention, but preferably there is at least one season (3 months of data)). The method uses corresponding historical weather data (the same period as measured data, the data come from a weather station closest to regional cold supply projects) as boundary conditions of a model, and an optimizer based on successive approximation simulation is used for searching for a proper combination of physical parameters and a use scene so that the chilled water flow and the backwater temperature calculated by the model are matched with actual measured values. Ideally, the model and calibration parameters used are close to, but not necessarily from, the model and calibration data of the project to be predicted. In other words, the model for a typical building may be applied to other similar buildings, such as hotels and offices of similar scale. The result of this step is a set of calibrated parameterized models M2. The number of models in this set of calibrated parameterized models M2 is typically 5 to 20, one for each (typical building).

Further, by way of example, the above-described history data is measured by a sensor (as shown in fig. 1). The sensor can obtain:

-historical total chilled water flow and water supply and return temperatures;

measurement of parts of the buildings which may represent typical ones (but not necessarily actual ones) connected to the cooling network, in order to obtain the same

Necessary data: chilled water flow, water supply and return water temperatures

Optional data: indoor temperature, indoor humidity

The following additional necessary input data need to be obtained:

-historical weather data including periods and locations of cooling networks and buildings under test, including:

necessary data: outdoor dry bulb temperature.

Optional data: cloud or solar radiation, outdoor air humidity, wind speed and direction.

Optional data: access to a prediction of the usage of a building (e.g., hotel reservations).

-predicting weather forecast during the period.

Step 3 (optional, in the figure (3))

Alternatively, for cooling networks with a large number of buildings (regional cooling), the set of "parameterized models" can also be expanded by modifying some physical parameters using random choices within a defined distribution range centered on the parameters of the "parameterized models", and modifying the choice of the building usage scheme (random or systematically).

The purpose of this step is to expand the number of models in the parameterized model to cover more possibilities for building thermal performance in the regional cooling program. For each "calibration parametric model", the parameter values may be centered on a normal distribution (e.g., using a normal distribution formula) A new parameter value is randomly selected in accordance with a preset standard deviation for each parameter. In this way, more parameter numerically models can be created with new physical parameters. Meanwhile, the use scene can be modified, can be randomly modified like physical parameters, and can also be according to a building operation strategy systemAnd modified.

The result of this step is an extended set of parametric numerical models M3, which also includes the set of calibrated parametric numerical models M2 obtained in step 2. Typically, the number of parameterized models in this large database is about 10 times the typical building selected in step 1. All of these parameterized models may be named for their features and stored in a database, and subsequently new models may be added as needed. All these existing parameterized models can then be selected from the database in a subsequent step.

Of course, it should be pointed out that this step is not necessary, especially in the case of (typical) buildings in the cooling network, the calibration parametric model already covers the existing building well without the need to scale up the model library.

Step 4 (4) in the figure

The simulation is performed on a set of parameterized numerical models M2 or a set of extended parameterized numerical models M3 for a preset historical period of time ending at the current time, called a "training period", depending on the condition of the project, for example, 5 days to 10 days. All physical parameters and usage scenarios of the "parameterized model" in step 3 are input into the physical model of step 1. And simulating all parameterized models in a training period by taking historical weather data of the position of the cold supply pipe network and the historical chilled water supply temperature as boundary conditions. The result of this step is a time series set of historical cold demands (flow and return water temperature) simulated correspondingly by each parameterized model during the training period.

Step 5 (5 in the figure)

Each time series (e.g., traffic time series) is given a fixed weight using statistical methods so that the weighted sum of all time series matches the measured cold (e.g., chilled water flow) during the training period (the weight may be 0). Likewise, other time series (e.g., each return water temperature) may be given a fixed weight. The result of this step is, for example, a set of weights for the flow time series, and/or a set of weights for the return water temperature time series.

For example, the above weights may be obtained by the following exemplary model and procedure using a regression algorithm.

|Y _regression -Y _measure |→0

Y _regression For the flow of the cold supply pipe network calculated by regression algorithm

p is the number of buildings

Beta is the fixed weight of each building calculated by regression algorithm

Y _i For the flow of chilled water in the ith building

C is the constant intercept

Y _measure For measuring the flow rate of the obtained cold supply pipe network

In statistics, linear regression is a linear method that models the relationship between a dependent variable and one or more independent variables (also referred to as output and input, response, and characteristics). In view of the historical dataset of responses and features, the linear regression model assumes that the relationship between responses and features is linear. Therefore, the equation of the model is:

Y＝Xβ+ε

wherein the method comprises the steps of

In this embodiment, Y represents a flow or return water temperature time sequence measured in the "training period", X represents a flow or return water temperature time sequence obtained by simulation for P building in the "training period", β represents a weight given to a simulation value of P building, and ε represents a difference between a weighted sum of all flow or return water temperature simulations and an actually measured flow or return water temperature. Beta is an estimate of beta by minimizing Y/u during the "training period _Prediction Is the difference between the predicted flow or return water temperature time sequence and the Y measured flow or return water temperature time sequenceThe sum of squares results, i.e., denoted as the sum of squares of the Residuals (RSS). The minimization process is called the general least squares method (OLS), and is represented by the following formula:

this is a linear least squares method used to estimate the unknown parameters in a linear regression model. Epsilon measurement of Y/u _Prediction And Y.

Of course, other regression models can be used in the present invention to obtain the respective weights.

Step 6 (6 in the figure)

The "parameterized model" is simulated within a preset "prediction period" that starts from the current time. The simulation is bordered by forecasted weather data for the location of the cooling network, where the weather forecast is from the nearest weather station. The result of this step is a set of time series of future cooling demands corresponding to the "parameterized model" set simulated by the "forecast period", i.e. a time series of predicted cooling demand corresponding to each of the parameterized models, respectively.

Statistically, at β and X/u _Prediction On the basis of (1), the simulation data of the P building in the prediction period is obtained by the following formula, wherein Y/u _Prediction The method is used for predicting the refrigerating capacity demand in a period, namely predicting the refrigerating water flow and the backwater temperature.

Over time, the historical data may change accordingly. Thus, if the above process can be repeated on a frequency (e.g., every hour), the prediction results can be updated in time (e.g., every hour).

Further, to evaluate the effectiveness of the model, mean Square Error (MSE) and Mean Absolute Error (MAE) may be utilized, as follows. The values of both indicators are used to evaluate the difference between the measured and predicted time series of cold demand (flow or return water temperature) during the training period, with the error always being positive and decreasing as the error approaches zero.

Wherein n is the number of the time series of the measured flow or the backwater temperature in the training period, y _i Andthe i-th measured and predicted flow or return water temperature, respectively.

Step 7 (7 in the figure)

Applying the weight set obtained in step 5 to the time series set obtained in step 6, calculating the total cold energy demand in the prediction period by weighted summation, such as the predicted flow rate and the predicted backwater temperature, and obtaining by the following formula:

wherein Y/u _Prediction The method is used for predicting the cooling capacity requirement of the cooling pipe network.

Through the above exemplary steps, the method according to the present invention can accurately and timely predict the cooling capacity demand in the cooling pipe network within a certain time in the future.

Furthermore, in the present embodiment, steps 4 to 7 are repeated for each new prediction. While steps 1 to 3 may be performed only once, or repeated less frequently, the set of "parameterized models" is completed with new models, depending on whether there is a substantial change in the building in the corresponding cooling network and its cooling conditions, for example increasing or decreasing the building accessed or increasing or decreasing the cooling (different time check-in rates of hotels).

Preferably, unnecessary models can also be deleted in the "parameterized model" set, for example:

-statistically analysing the weight history attributed to the different models, deleting the underutilized models; and/or

By analysing the history of the time series, the redundant model is statistically deleted, e.g. the result is too close to the other models.

As shown in fig. 1, the method according to the present invention can be applied to an entire regional cooling system, which uses sensors and a control platform of a regional cooling station for data monitoring. The predictions of the cooling demand obtained by the method according to the invention as described above are sent to the control and monitoring platform of the regional cooling plant, these predictions will be converted into control signals and transmitted to the refrigeration equipment, and the workload of the refrigeration equipment will then be adjusted to meet the future cooling demand in the cooling pipe network. At the same time, due to the monitoring sensors installed in the cold supply network, real-time operating data of the cold supply network is sent back to the calculation model according to the invention, which can remain trained to optimize the working conditions of the cold supply station.

A more specific example will be given below to illustrate the method according to the present invention for predicting the demand for cold in a cold supply network. This embodiment relates to a regional cooling station whose primary users are office buildings and parts of commercial buildings. The regional cooling station produces chilled water and supplies it to the user through a cooling pipe network.

First, a physical model is built according to the type of building. As described in step 1 above, these models include a set of linear and nonlinear equations describing heat and mass (humidity) transfer. For example, the heat transfer of the exterior wall is based on the following heat transfer equation:

Q＝K*A*ΔT

q is heat (W) transferred through the outer wall

K is the heat transfer coefficient (W/(m) ² *K))

Delta T is the indoor and outdoor temperature difference (DEG C)

Wherein,,

K＝1/(R _i +R _wall +R _o )

r isThermal resistance (inner and outer surface heat transfer resistance, and outer wall heat transfer resistance) (m ² *K/W)

Thus, R is _wall Is one of the physical parameters mentioned in step 1, which will be calibrated in step 2 above.

Meanwhile, the cooling capacity requirement is calculated according to a typical relation curve between cooling power and flow or backwater temperature. These typical relationships depend on the actual operating strategy of the district cooling station.

Next, 10 typical buildings with measurement data were selected in this example. Based on these measurement data, the optimizer solves for the appropriate combination of physical parameters and usage scenarios in step 2, creating a database with "calibration parameterized models". Each "calibration parametric model" has a specific physical parameter and value for the usage scenario. For example, for a selected typical building, the thermal resistance is 4m ² * K/W, considering the common value of the building in the area where the refrigeration project is located, the thermal resistance range of the outer wall can be defined to be 2-10 m ² * Within K/W, the optimizer will then try different combinations of all parameters and schemes to arrive at the most appropriate flow or return water temperature profile to accommodate the measured data. Finally, for one of the "calibrated parametric models", solving yields 5.16m ² * K/W.

The database is then optionally extended to cover the possibility of more building thermal performance. For example, 5.16m obtained in the previous step ² * The thermal resistance of K/W is used as a central parameter value, and a normal distribution is established on the basis of the central parameter value. Then some new parameter values were randomly selected at a standard deviation of + -30%. Finally, a total of 104 parameterized models were obtained.

All these "parameterized models" are then entered into the physical model mentioned in step 1, simulating the chilled water flow and return water temperature during the training period based on historical climate data obtained from the nearest climate station. The training period may be set to 7 days, for example.

Next, in combination with the simulation prediction results, each "parameterized model" gives a proportion (weight) by regression calculation, and calculates the cooling capacity demand (chilled water flow and return water temperature) within 24 to 48 hours in the future, and an example of 24-hour prediction is given in this embodiment. From this, a weighted sum of "parameterized models" yields a prediction of total cooling demand for 104 buildings in the future 24 hours. In this process, an estimate of β, is obtained by the following exemplary equations and statistical methods:

beta can be obtained by minimizing the difference between the measured and predicted values (Y and X) using the common least squares (OLS) method. Applying the formulaWherein in the present embodiment X\u _Prediction 24-hour weather forecast from "parameterized model" to obtain Y/u _Prediction 。

The model according to the present invention results in less index values (mean square error and average absolute error) than the conventional method of simply taking 24 hours data of the previous day as predictions for the next 24 hours. In this example, the average deviation of the method according to the present invention was about 10.4%, whereas the average deviation of the conventional method of repeating the previous day demand curve was 12.2%.

Finally, training and prediction (steps 4 to 7) may be repeated every 1 hour, so that the cooling demand for the next 24 hours may be predicted without interruption.

Compared with the traditional cold demand prediction method, the method can improve the prediction accuracy and efficiency under the condition of using a small amount of training data.

Those skilled in the art will recognize various embodiments and variations and modifications. In particular, it is to be understood that the features, variations and/or embodiments of the invention described herein may be combined with each other, unless clearly contradicted or incompatible. All such embodiments and variations and modifications are intended to be within the scope of the present invention.

Claims

1. A method of predicting cooling demand in a cooling network, comprising:

-establishing an initial library comprising a set of building thermal performance numerical models (M1) comprising a plurality of models, each model corresponding to a respective building in the cooling network, and each model comprising a limited number of parameters and being adapted to calculate the cooling demand of the respective building;

-calibrating said parameters in each of said set of building thermal performance numerical models using the historical measured cold demand and the historical weather data of the respective building as boundary conditions, thereby obtaining a set of parameterized numerical models (M2);

-simulating the set of parameterized numerical models (M2) during a preset training period, wherein the training period ends at a current time to obtain a set of simulated historical cold demand time sequences corresponding to each model of the set of parameterized numerical models (M2), respectively;

-simulating said set of parameterized numerical models (M2) during a preset prediction period, wherein said prediction period starts at said current time, with weather forecast data during said prediction period as boundary conditions, thereby obtaining a set of predicted cold demand time sequences corresponding to each model of said set of parameterized numerical models (M2), respectively; and

-applying the obtained fixed weights to the set of predicted cold demand time sequences so as to obtain a prediction of cold demand during the prediction period by a weighted sum of the set of predicted cold demand time sequences.

2. The method of predicting refrigeration demand in a refrigeration line of claim 1, wherein the refrigeration demand comprises total flow and/or return water temperature of the refrigeration line.

3. The method of predicting cooling demand in a cooling grid according to claim 1, wherein each model comprises linear and/or nonlinear differential equations describing heat transfer with the limited number of parameters.

4. The method of predicting cooling demand in a cooling network of claim 1, wherein the limited number of parameters are physical parameters of a building and/or a building usage scenario.

5. The method of predicting cooling demand in a cooling grid according to claim 4, wherein the physical parameters of the building include floor area, wall area, thermal inertia, glass area, heat transfer coefficients of air handling units, thermal resistance of walls, windows and floors, solar transmittance, and the building usage scenarios include occupancy, indoor temperature settings, ventilation times, lighting, internal heat gain of equipment and personnel, and heat gain from solar radiation.

6. Method for predicting cooling demand in a cooling network according to claim 4, characterized in that the set of parameterized numerical models (M2) is expanded into a set of expanded parameterized numerical models (M3) by modifying at least part of the physical parameters and modifying the building usage scenarios, wherein the physical parameters are modified by using random selection within a defined distribution centered on the parameters of the parameterized numerical models.

7. The method of predicting refrigeration demand in a refrigeration line network of claim 1, wherein the set of parameterized numerical models is expanded into a set of expanded parameterized numerical models based on measurements taken on buildings connected to the refrigeration line network.

8. The method of predicting refrigeration demand in a refrigeration line network of claim 7 wherein the measurements taken of the building connected to the refrigeration line network are taken by a set of sensors installed in the building over a prescribed period of time to calibrate the parameters.

9. The method of predicting refrigeration demand in a refrigeration line of claim 8, wherein the set of sensors includes measuring a flow rate taken by the building from the refrigeration line, a water supply temperature of the refrigeration line, a water return temperature of the refrigeration line, an indoor temperature at a selected location within the building, an indoor air humidity at a selected location within the building, and an occupancy rate.

10. The method of predicting refrigeration demand in a refrigeration line of claim 1 wherein said statistical method is linear regression.

11. The method of predicting refrigeration demand in a refrigeration line of claim 1, wherein the training period is 3 to 10 days prior to the current time.

12. The method of predicting refrigeration demand in a refrigeration vessel network of claim 1, wherein the prediction period is 24 to 48 hours after the current time.

13. An apparatus comprising a processor configured to execute instructions stored on a computer readable medium to perform the method of predicting refrigeration demand in a refrigeration utility network of claim 1.