Control method of an electricity distribution network and control system for implementing said method

Abstract

Control method of an electricity distribution network, as defined above, which method further provides steps of generating synthetic data of the time trend of the energy withdrawn and / or the energy generated or fed into the network for each or at least part of the withdrawal units and / or the generation units that are not provided with the function of detecting and storing the time trend of the energy withdrawn and / or the energy generated or fed into the network, and which steps of generating said synthetic data comprise the generation of said synthetic data by means of an algorithm for estimating said synthetic data based on the data relating to the time trend of the energy withdrawn and / or that produced or fed into the network which are stored in each or a part of a plurality of different energy withdrawal units and / or different energy generation units, and which units are provided in a pre-established boundary with respect to the individual withdrawal units and / or the respective generation units which are without said detection and storage functions. The invention concerns a control system of a distribution network which uses said method, and a method for predicting the time trend of the consumption of a withdrawal unit which is without storage functionality, using the data relating to withdrawal units that do have said storage functionality and are within a boundary of said unit without said storage functionality.

Description

[0001] “CONTROL METHOD OF AN ELECTRICITY DISTRIBUTION NETWORK AND CONTROL SYSTEM FOR IMPLEMENTING SAID METHOD”

[0002] FIELD OF THE INVENTION The present invention concerns a control method of an electricity distribution network and a control system for implementing such method, which network comprises one or more electrical energy generation units that feed the network and a plurality of separate and independent withdrawal points of the electrical energy, in particular distributed over a pre-established geographical area, and wherein at least a part of the withdrawal points is provided with a storage unit for the electricity withdrawal curve measured as a function of time from the corresponding withdrawal unit and with a port for reading the withdrawal curve, and wherein the withdrawal curves stored in at least a part of the withdrawal units of the plurality of withdrawal units are supplied to a central control unit, and wherein at least a further part of the withdrawal units is not provided with a storage unit for the electricity withdrawal curve measured as a function of time from the corresponding withdrawal unit, which control unit is further connected to measurement units for the electricity generation curve of each of the electrical energy generation units and for the operating conditions of the electrical energy units, a control program being stored and executable in the central control unit for controlling the electricity distribution network as a function of the withdrawal curves of the individual withdrawal units and of the electricity generation curves of the individual generation units.

[0003] BACKGROUND OF THE INVENTION

[0004] The progressive abandonment of fossil energy sources and the transition toward energy sources with less environmental impact and toward user devices with higher energy efficiency have led to substantial structural changes in electricity distribution networks.

[0005] Sources of electrical energy generation alternative to fossil and / or hydroelectric sources, such as solar and / or wind and / or other types for example, have made it possible to create small electrical energy generation units, configured to meet at least part of the energy demands of each individual user or small groups of users. Unlike traditional sources of electrical energy, that is, based on the combustion of fossil fuels and / or on the transformation of water energy and / or on the exploitation of nuclear energy, these alternative energy sources have the characteristic of atomizing the energy sources and distributing said sources throughout the territory. In addition, the currently most widespread alternative energy sources, that is, those sources operating on the basis of photovoltaic processes and / or those driven by wind energy, suffer from the fact that they are extremely sensitive to climatic conditions in terms of the amount of energy produced. For photovoltaic sources in particular, there are also problems related to the fact that they do not guarantee an energy flywheel effect that allows to deal with maximum and minimum power peaks, so that in the presence of strong solar radiation, a large amount of energy is generated that is fed into the networks and that can lead to high power peaks, both localized in sub-areas of the distribution network and also affecting larger zones of the distribution network. On the contrary, in the case of low solar radiation, the energy produced by photovoltaic systems drops drastically, while often in parallel the energy demand is greater than in conditions where solar radiation is greater. A possible energy flywheel to control positive and / or negative energy peaks that occur in the form of fast transients is the use of accumulators, such as traditional accumulators and / or even super capacitors. However, these devices are relatively expensive, have relatively short average lives, and require constant functional monitoring.

[0006] As far as wind generators are concerned, these require a certain force of the winds to be efficient, but they have the advantage that they can act as power flywheels that allow to control and compensate for power peaks, within certain limits. It is evident that in a control method and system of an electricity distribution network, or at least a part of such network, it is essential to constantly monitor the operating conditions of the network and intervene in a very short time to implement measures that allow to deal with relatively rapid changes in operating conditions, both from the point of view of the energy required and also from the point of view of the energy generated. In relation to this requirement, it should also be considered that current climatic changes mean that, in many cases, climatic effects are extreme, in relation to their stability over time, that is, climatic changes are more extreme and temporally limited, and this entails a further rapid variability of the power generated and / or of that withdrawn or demanded from a distribution network. In the context of controlling, above all, the energy withdrawals by the different withdrawal units, it is also important to be able to detect anomalies regarding the peaks of energy withdrawn in order to establish whether there are conditions of dispersion of electrical energy caused by damage to the network infrastructure and / or abusive withdrawals of electrical energy.

[0007] In current electricity distribution networks, not all withdrawal units are of the so-called smart-meter type, that is, provided with a hardware / software configuration that allows them to record the energy withdrawn from the network through them in the form of time series, that is, the time trend of the energy withdrawn from the electricity distribution network through the withdrawal unit, and to store the data relating to the curves that represent this time trend in a pre-established time window that is updated with a pre- established updating frequency. A relatively large number of withdrawal units, therefore, is still unable to supply data relating to the time trend of the energy withdrawn, with a consequent lack of the corresponding data for the execution of the prediction algorithm of future consumption in one or more pre-established future time instants and / or in a future time window defined for the prediction.

[0008] This condition not only introduces possible inaccuracies in the prediction at a general level, but does not allow to have reliable predictive data for the withdrawal units that do not allow the detection and storage of the time trend of the energy withdrawn by them.

[0009] The number of withdrawal units is already very high even at a national level, therefore the process of replacing withdrawal units of the type that does not allow detection and storage of the trend of the energy withdrawn, with withdrawal units according to the most advanced configuration, and in particular a configuration that allows said detection and said storage, is not complete, therefore the number of withdrawal units that are without the detection and storage functions is still relatively high and the influence on the prediction processes is still not negligible.

[0010] For a simpler disclosure, hereafter the withdrawal units that are not configured to perform the functions of detecting and storing the time trend of the energy withdrawn by them will be referred to as “registry withdrawal units”, because the available data that characterize these units is essentially related to geographic location, the type of supply in relation to the electrical power made available and / or other physical parameters of the electrical power signal withdrawable, such as the number of phases available, transient power peaks allowed, voltage, frequency, etc., and other indications such as, for example, climate zone and / or also type of user to whom a withdrawal unit is assigned, and optionally also further data which is substantially stable over time. On the other hand, the withdrawal units that have the functionality for detecting and storing the time trend of the energy withdrawn are referred to, for brevity, as “withdrawal units with consumption”. A similar line of thinking can also concern electrical energy generation units that feed at least part of the energy generated into the network, even though, generally, the installation of these units also requires an upgrade of the withdrawal units belonging to the same user and / or private system, not only for technical purposes, but also for energy reporting purposes for calculating economic offsets based on the energy fed into the network, therefore for electrical energy generation units, the reasoning outlined above for the withdrawal units is, although plausible, only technically hypothetical, the impact on the overall reliability of the prediction remaining, however, substantially negligible.

[0011] In the state of the art, methods are known that allow to estimate energy consumption, that is, the time trend of the energy withdrawn from a registry withdrawal unit using the data provided by withdrawal units with consumption present in the same network, and therefore generate an estimate of the time trend of the energy withdrawn from the network through each or at least a part of the registry withdrawal units present in the network in a pre-established time window, and that make the registry withdrawal units virtually identical to those with consumption, thus allowing to treat the registry withdrawal units in an identical way to those with consumption with regard to the predictive process and the functions of the network, or the interventions on the network that are activated and controlled as a function of the outcome of the prediction.

[0012] Considering that the number of withdrawal units is, as already indicated, very large and with a wide distribution across the territory, the reliability of the prediction, especially in relation to identifying anomalies of energy withdrawals resulting, for example, from the comparison of the outcome of the prediction on the energy withdrawal for each withdrawal unit and that actually measured for the withdrawal units with consumption, or that estimated for the registry withdrawal units, is an important factor, since the indication of a possible anomaly has the consequence that at least verification activities are prepared which, in many cases, make it necessary to carry out inspections by technical personnel at the withdrawal units, with the corresponding expenses from the point of view of intervention times and the costs of the interventions themselves.

[0013] In light of the above, the accuracy and reliability of the estimation of the time trends of electrical energy withdrawal for the registry withdrawal units on the basis of the time trends of electrical energy withdrawal stored in a plurality of withdrawal units with consumption is essential to obtain a reliable prediction that does not cause so-called “false positives” or “false negatives”, and the consequences produced thereby. In addition, it should also be considered that, from a technical point of view, not only can estimation methods produce results that differ in terms of reliability, but many of the known methods require the use of complex prediction or classification algorithms which require training steps, needing many training records, an assessment of the withdrawal units characterization variables and / or the labels, that is, the outcomes in the form of reliability parameters or “false / non-false” indication of the estimates, and which, also from a computational point of view, require long times and high computing powers.

[0014] On the other hand, the operations of predicting, and therefore the operations of estimating, the energy withdrawals over time for at least a part, preferably for all the registry withdrawal units have to be repeatable with high frequency over time, preferably continuously or at pre-established time frequencies characterized by relatively short time intervals, preferably of the order of magnitude of single hours or minutes, therefore the techniques currently known, and in particular the techniques that use supervised machine learning algorithms to obtain the estimates of the energy withdrawal trends for the registry withdrawal units, do not meet the criteria of simplicity, speed as well as low computational and cost burden that are required for their effective practical implementation, which therefore goes beyond theoretical effectiveness.

[0015] The document Al-Wakeel Ali et al: “k-means based load estimation of domestic smart meter measurements”, Applied Energy., vol. 194,1 May 2017 (2017-05-01), pages 333- 342, DOI: 10.1016 / j.apenergy.2016.06.046. The purpose of this document is to address technical and operational challenges of smart meter measuring systems. The technical challenges include intermittent communication networks (both mobile and radio frequency); lack of sufficient signal strength; lack of tools to detect mobile network failures; and intemal / extemal positioning of meters. Examples of operational challenges include planned or unplanned system maintenance, software and hardware failures or smart meter malfunctions, and customers unwilling to communicate their energy consumption data. These challenges make smart meter measurements susceptible to time delays or even temporary losses when required by energy suppliers or network operators. This document therefore aims to develop a load estimation algorithm to replace missing measurements and estimate future smart meter measurements. In this sense, load estimation analyzes past measurements and extracts practical information, such as typical customer load profiles, to estimate missing measurements. Home smart meter measurement load estimation is based on k-means cluster analysis. The algorithm applies cluster centers, of previously grouped load profiles, and distance functions to estimate any missing and future measurements.

[0016] SUMMARY OF THE INVENTION

[0017] The present invention aims to provide a control method and a system for implementing such control method that, thanks to relatively simple measures, allows to deal with the greater dynamism of energy demand and generation in an electricity distribution network, or in a part thereof, and that allows to keep the network under control, down to the detail of the individual user devices, both in terms of energy consumption and also in terms of energy production, as well as allowing to set up in a timely fashion controls directed to the network infrastructures and / or structural changes to the network that can prevent transient overloads of the network, whether positive or negative, and / or detect failures of the network infrastructures that cause dispersions, as well as any abusive energy withdrawals, which method is able to reliably consider also energy withdrawal units and / or energy generation units of the registry type, that is, which are not provided with functionalities for detecting and storing the time trends of the energy withdrawn and / or that generated, and to also extend the predictions to these units, or at least to part of them.

[0018] The present invention solves the problem outlined with a control method of an electricity distribution network as defined above, which method further provides steps of generating synthetic data of the time trend of the energy withdrawn and / or the energy generated or fed into the network for each or at least part of the withdrawal units and / or the generation units that are not provided with the function of detecting and storing the time trend of the energy withdrawn and / or the energy generated or fed into the network and which steps of generating the synthetic data comprise the generation of the synthetic data by means of an algorithm for estimating the synthetic data as a function of the data relating to the time trend of the energy withdrawn and / or that produced or fed into the network that are stored in each or in a part of a plurality of different energy withdrawal units and / or different energy generation units, and which units are provided in a pre- established boundary with respect to the individual withdrawal units and / or the respective generation units which are without the detection and storage functions, the steps of generating the synthetic data being carried out before the calculation of the prediction for each or a part of the energy withdrawal units and / or of the energy generation or feed into the network units; and the synthetic data generated being associated, respectively, with each or with a pre- established part of corresponding energy withdrawal units and / or energy generation or feed into the network units, and the synthetic data being used for calculating the prediction of the trend of the energy withdrawn from the network relating to the withdrawal units and / or the generation or feed into the network units which are without the functionalities for detecting and storing the time trend of the withdrawal of energy and / or the generation or feed of energy into the network, respectively.

[0019] Advantageously, in order to calculate the estimate of the data relating to the trend of the energy withdrawn and / or the energy generated or fed into the network for the withdrawal units and / or the generation units of the so-called registry type, there is provided a preventive step of selecting a subgroup of withdrawal units and / or generation units comprising units that have the functionalities for detecting and storing the time trends of the energy withdrawn and / or that generated or fed into the network, there being established criteria defining the limits of the boundary within which to select the withdrawal units and / or the generation units of the type defined as units with consumption, these steps comprising:

[0020] - the definition of one or more characterization variables of the aforementioned withdrawal and / or generation units relating to corresponding description parameters of the characterization variables;

[0021] - the definition of metrics that define a distance value as a function of the characterization variables between each registry-type withdrawal and / or generation unit and each or a part of the withdrawal and / or generation units provided in the network, and the definition of threshold values of said distance values, which threshold values define the boundaries of the boundary and the selection of the withdrawal units and / or of the generation units which fall within the boundary as units whose time trends stored and related to the withdrawal of electrical energy and / or the generation and / or feed of electrical energy into the network are used to estimate the corresponding time trends for a pre-established registry-type unit;

[0022] - the application to the data of the trend of energy withdrawal and / or generation or feed of energy into the electricity network of an algorithm for estimating the synthetic data for the corresponding registry-type electrical energy withdrawal unit and / or electrical energy generation or feed into the network unit, and the unique association of the synthetic data to the corresponding registry unit for the generation of a corresponding virtual unit that simulates a unit of the type with consumption. Several different algorithms, and in particular machine learning algorithms, can be used to estimate the virtual data, that is, calculate the time trend of energy withdrawn and / or energy generated or fed into the network relating to registry-type units based on the corresponding data of a plurality of selected units of the type with consumption. While theoretically all machine learning algorithms can help determine these synthetic values, for the registry-type units, not all can be compatible with the need for a rapid and dynamic adjustment to changes in the positive or negative load conditions of the network. Current timing for implementing monitoring and / or control operations, such as for example those provided in this method and relating to the prediction of energy withdrawal and / or energy feed from and into the network through the individual units, is of the order of one repetition of the method with time intervals between these repetitions of the order of ten or a few tens of minutes, for example, about fifteen minutes. In addition, the number of withdrawal and / or generation units under control and for which the prediction is performed is very high, therefore calculation speed is essential, as is the reliability of the calculation. According to one embodiment, the generation of the synthetic data for one or more of the withdrawal and / or generation units of the type defined as registry units according to the present description, provides that the synthetic data be produced by estimating the time trends of energy withdrawal and / or energy generation or feed into the network for the registry-type withdrawal and / or generation units using a clustering algorithm. One example embodiment provides to use, as clustering algorithm, the algorithm called k-means.

[0023] The k-means algorithm is a known unsupervised clustering algorithm, consisting of a method for dividing a set of objects into a number of K groups called clusters in such a way that within each cluster the objects grouped therein have the highest degree of similarity, while the objects belonging to different clusters have the maximum degree of dissimilarity between them. In the present case, the objects consist of the individual energy withdrawal and / or generation units provided in the network, or in a sub-network thereof, each unit being characterized using a combination of variables whose values define the essential features of these units for the purpose of generating the clusters.

[0024] The clustering process using the k-means algorithm is iterative and, starting with a K number of randomly chosen centroids (representative points), it assigns each withdrawal and / or generation unit to the cluster whose centroid is closest with reference to a metric that takes into account the difference in the values of the variables for each withdrawal and / or generation unit. Once all the units have been assigned to a cluster, the centroids are updated as the average of the points assigned to the cluster. This process is repeated until the centroids converge, that is, until they no longer change significantly, that is, until for each iteration step there is a minimum total variation within the cluster set. The computation of the synthetic data, that is, of the fictitious values of the time trends of the withdrawal and / or generation or feed of electrical energy according to the k-means clustering algorithm, requires preparatory actions that are related to determining the optimal number of clusters and defining the variables that characterize the individual units, in order to divide them into the different clusters. The quality of the clustering process of basically all clustering algorithms, and thus also of the preferred k-means algorithm, depends on the choice of an optimal number of clusters. The number of clusters has an influence on the total variation within the clusters, therefore according to one embodiment the method according to the present invention provides a step of determining the optimal number of clusters based on the analysis of the minimum total variation of the clusters represented by the trend of a cost function as a function of the number of clusters, at least two different methods being used in parallel to determine the optimal number of clusters as a function of the analysis of the trend of a cost function, in particular the so-called Elbow Method and the so-called Silhouette Method. The method is characterized by a specific cost function and a by specific characterization parameter extrapolated from the trend of the cost function and which coincides with the optimal number of clusters.

[0025] The k-means clustering algorithm and the algorithms for determining the number of optimal clusters, in particular with reference to the Elbow Method and the Silhouette Method, are described in greater detail in several publications, including, for example:

[0026] “Introduction to unsupervised clustering with K-Means” published online at https: / / www.diariodiunanalista.it / posts / introduzione-al-clustering-non-supervisionato- con-k-means / Or

[0027] “K Means” Written by Chris Piech. Based on a handout by Andrew Ng. published online at https: / / stanford.edu / ~cpiech / cs221 / handouts / kmeans.html.

[0028] As far as the methods for determining the optimal number of clusters are concerned, a summary description of the methods is contained in the document “Review on Determining of Cluster in K-means Clustering” Trupti m Kodinariya, Prashant Makwana published at the web address https: / / www.researchgate.net / publication / 313554124_Rev iew_on_Determining_of_Cluster_in_K-means_Clustering.

[0029] After determining the optimal number of clusters, the further step of generating a training database for the algorithm and training the k-means algorithm itself is provided.

[0030] According to one embodiment, this step provides to define one or more identifying registry features of the individual energy withdrawal and / or generation units which characterize both the so-called registry units and also the so-called units with consumption, associate each or at least part of the features with corresponding variables and define metrics for calculating the distance in the multidimensional space defined by the variables representing, respectively, each or at least part of the features between the individual withdrawal and / or generation units of the registry type and of the type with consumption, the variables that describe the features as well as the distance values defined by the metrics constitute the records of the training database on the basis of which the k- means algorithm is trained.

[0031] According to a further step, provided in combination with and in succession to the previous steps, the method according to the present invention provides to determine the effectiveness in relation to the discrimination of the individual withdrawal and / or generation units from each other, comprising both the units with consumption and also the registry units which are assimilated to units with consumption only of at least some, or of all the different registry features, defined in the previous step, this identification being performed on the basis of an analysis technique called PCA (Principal Component Analysis) Wold, Svante, Kim Esbensen, and Paul Geladi. “Principal component analysis.” Chemometrics and intelligent laboratory systems 2.1-3 (1987): 37-52. In one embodiment, the geographic component is used as the main feature by grouping the geographical components across eight clusters and assigning membership of the individual withdrawal and / or generation units to one of the eight clusters on the basis of the feature that defines the positioning zone of those units. In one embodiment of the invention, one or more, preferably all the features represented by the following data, are added to the categories of registry features, defined as described above:

[0032] - composition of the household associated with a withdrawal unit, or a meter; - square meters of the home that is served by a withdrawal unit, or by a meter;

[0033] - type of dwelling, such as a detached house or condominium;

[0034] - floor at which the dwelling served by a withdrawal unit, or by a meter, is provided;

[0035] - altitude / meters above sea level at which the dwelling served by the withdrawal unit, or by a meter, is located. According to another characteristic, the result of the clustering process operated on the registry data of the individual withdrawal and / or generation units generates a distribution of the withdrawal and / or generation units across the optimal number of clusters, and the determination of the synthetic data of the time trend of the energy withdrawal and / or the generation or feed of energy into the network for a registry-type withdrawal and / or generation unit is performed using the real data of the time trends of the withdrawal of energy and / or the generation of energy or the feed of energy into the network of a pre- established number of energy withdrawal and / or generation units of the type with consumption which are present in the same cluster comprising the registry-type withdrawal and / or generation unit and which present the smallest distance from the registry-type unit with reference to the variables representing one or more features with discriminatory effectiveness of the energy withdrawal and / or generation units and with reference to the specific embodiment with reference to the geographical position.

[0036] According to one embodiment, the estimated trend for a registry-type withdrawal unit and / or generation unit is calculated by determining the average of the time trends of the energy withdrawal and / or the generation or feed of energy into the network stored in the corresponding units with consumption selected as closest and within a pre-established time window, thus generating a fictitious time trend of the energy withdrawn and / or the energy generated and / or fed into the network relating to the registry-type withdrawal unit and / or generation unit. According to one embodiment, the units with consumption that are closest to a registry unit present in the same cluster are determined by applying an algorithm called KNN. This algorithm is known in the state of the art and is described for example in the document published on the web at https: / / it.wikipedia.org / wiki / K-nearest_neighbors or Peterson, Leif E. “K-nearest neighbor.” Scholarpedia 4.2 (2009): 1883.

[0037] According to an improvement of the step of determining the units of the type with consumption that within a same cluster are closer to a registry-type unit, it is provided to use the KNN algorithm in combination with a maximum threshold of difference between a possible historical consumption value recorded for the registry unit referring to a pre- established past time instant and the consumption measured by the withdrawal units of the type with consumption at the same time instant, the determination of the proximity being measured based on whether or not the consumption difference is greater or less than the maximum threshold, and only those units for which this difference is less than the maximum threshold being considered as closest units, the above obviously when a datum on energy consumption at a certain pre-established previous time instant is available for a registry unit.

[0038] This improvement can provide that the step of comparing the energy consumption described above is performed at the same time as the execution of the KNN algorithm, there being provided, for example, a variable that defines the condition of being lower and higher than the maximum threshold for the difference, and that is part of the variables that characterize the withdrawal units and as a function of which the distances between the units with consumption and a registry-type unit are measured.

[0039] Alternatively, the step of selecting on the basis of the value of the difference with respect to the maximum threshold provided therefor can be a preventive step, or a step that follows the execution of the KNN algorithm.

[0040] According to an improvement, the step of calculating the average provides in combination a step of filtering so-called outliers before the step of calculating the average.

[0041] For the predictive calculation of the consumption in one or more future time instants or in one or more future time periods it is possible to take different methods operating on the basis of different predictive algorithms, in particular machine learning algorithms that generate a prediction of the trend of the energy withdrawn and / or of the energy generated and fed into the network at the one or more future instants and / or in the one or more future time periods on the basis of the real time trends of the energy withdrawn and / or of the energy generated or fed into the network and at the fictitious time trends in the time window for the units with consumption and for the registry units, respectively.

[0042] Also in this case, the selection of the predictive process, that is, of the prediction algorithm, is not obvious, since the same needs of reliability of the prediction result, speed of calculation of the prediction for time instants or time periods that are relatively close and / or short apply, therefore in general machine learning algorithms such as neural networks or suchlike are not suitable because they are relatively slow.

[0043] According to one embodiment, the method of the present invention provides a program for predicting the time trend of the energy withdrawn through at least part, preferably each of the withdrawal units and / or the trend of the generation or feed of energy into the network by at least a part or by each of the generation units in one or more future time instants and / or in one or more future time periods, which prediction program comprises the instructions for the processor of the central control unit, which are related to the steps of executing a regression and statistical classification algorithm of the supervised type, in particular of the type called gradient boosting, such program being retrievable and executable by the central control unit.

[0044] This algorithm is known and is described in greater detail in several publications including, for example, the document published on the internet page https: / / it.wikipedia.org / wiki / Gradient_boosting or Freund and Schapire, 1997; Friedman et al., 2000; Friedman, 2001.

[0045] This algorithm is particularly suitable for the execution of predictive operations, in particular relating to the trend of energy consumption, or the loads to which a pre- established electricity distribution network is subject in a pre-established period of time. The program executed by the central control unit for determining the prediction is stored in a memory of the central control unit and is executable by the central control unit, and comprises the instructions for training the algorithm, as well as the instructions for enabling the central control unit to execute the computation steps provided by the algorithm in its trained condition. According to a further characteristic, the method provides, on the basis of the predictions, steps of executing and / or planning activities for compensating for transient peaks of electrical energy withdrawal and / or generation, which activities are synchronized with the instant in which the occurrence of these peaks has been expected and / or the advance planning, for the moment in which the occurrence of such transient peaks is expected, of structural measures and / or configuration of the connection / s between the withdrawal units and / or the generation units of the network and / or planning of field checks of the condition of the network infrastructures.

[0046] Some examples of such compensation activities can comprise activating links with further distribution networks and / or further zones of the distribution network, thanks to connections that can be activated and deactivated by servo-controlled change-over switches, for example, by the central control unit or by a command unit in turn controlled by the central control unit, and as a function of the prediction relating to the time trend of the withdrawals and / or the generation of electrical energy.

[0047] In this case, for example, an electricity distribution network comprising a plurality of electrical energy withdrawal units and / or a plurality of electrical energy generation units can be divided into sub-networks, each of which is characterized by a different geographical location area of the corresponding withdrawal and / or generation units, and also by the number of such units, as well as the type of withdrawal units with reference to the energy withdrawal curves of the individual withdrawal units, as well as by the type of electrical energy generation units, while between one or more of the sub-networks there are provided electrical connections able to be switched between a closed and an open condition, the switching state of the one or more connections between the sub-networks being automatically commanded into a closed or open condition as a function of the results of the prediction relating to the withdrawal of electrical energy and / or the generation of electrical energy for the future time period, and the closing and / or opening condition of the connections between the sub-networks being varied in the future time period as a function of the variation in the time trend of the withdrawal and / or the time trend of the generation of electrical energy for the various sub-networks.

[0048] Thanks to the aforementioned steps, it is possible to keep the network’s operating conditions stable, preventing transient peaks of excess electricity production or excess demand for electrical energy, by promptly connecting, on the basis of the trends resulting from the aforementioned prediction, parts of an electricity distribution network which, on the basis of the aforementioned predictions, compensate each other in relation to the energy produced and the energy withdrawn. This stabilization measure occurs without delay with respect to the onset of instabilities, and prevents excessive stress on the operating conditions of the network or the need to intervene more drastically in order to stop a positive or negative drift of the energy present in the network. The above mentioned measures can also comprise, alternatively or even in combination with other types of interventions, such as for example the timely creation, before the occurrence of transient events of excessive withdrawal or excessive generation of energy, of compensation stations, such as electrical energy storage stations functioning as an electrical energy flywheel to level out negative and / or positive energy peaks in the network or in a specific sub-network.

[0049] The results of the prediction allow to provide a sizing of such storage stations and possibly also a modular arrangement in order to expand the capacity of these stations.

[0050] In one embodiment, the period to which the prediction of the time trend of the energy withdrawn and / or generated refers is of the order of at least a few days.

[0051] According to another characteristic of the aforementioned method, according to one or more of the embodiments and / or variants described above, for the synthetic data relating to the time trend of the energy withdrawn and / or of the energy generated or fed into the network referring to a registry-type withdrawal unit or generation unit, it is possible to provide a step of validating the reliability and accuracy of the synthetic data before using the synthetic data obtained according to one or more of the embodiments described above for the registry units, which provides a comparison between the synthetic data provided for one or more future instants and the real data measured at the same time as such instants, there being established metrics for measuring such differences and threshold values to discriminate between conditions of reliability and conditions of unreliability.

[0052] According to one embodiment, a first comparison metric provides to calculate the difference between the estimated values for the estimated energy time trend (estimated consumption) and the real values weighted with respect to the available power, averaging the data between the day’s quarters of an hour, according to the following formula:

[0053] A second comparison metric is the object of an alternative embodiment that provides to calculate the difference between the estimated values defined as estimated consumption in the function and the weighted real values (defined as weighted real consumption) with respect to the available power, by performing, for each withdrawal unit, the sum of the values between the day’s quarters of an hour and then averaging the withdrawal units according to the following formula:

[0054] In which UP is a numbering index of the individual withdrawal units provided in the network.

[0055] A third alternative comparison metric provides to calculate the difference between the estimated values (defined as estimated consumption) and the real values measured squared obtained by averaging the values relating to the day’s quarters of an hour according to the following formula:

[0056] When the variations between the predicted values and the values actually measured according to the various metrics exceed a pre-established limit value, it is possible to provide to repeat one or more of the steps described above regarding the definition of the optimal number of clusters, the definition of the registry features to be used to characterize the individual energy withdrawal units and / or the individual energy generation or feed into the network units, the identification among the aforementioned registry features of those with the greatest discrimination effectiveness, or of distinguishing the withdrawal units and / or the generation units, of dividing the units into individual clusters, of executing the clustering algorithm, or the algorithm called k-means, of determining the synthetic data relating to the trends of the energy withdrawn and / or energy generated or fed into the network, and of associating the synthetic data relating to the time trend of the energy withdrawn and / or that generated or fed into the network with a corresponding registry-type unit, thus generating from the registry-type units corresponding virtual units of the type with consumption to be used in combination with the real units of the type with consumption for the prediction of the future trends of the energy withdrawn by the withdrawal units and / or of the energy generated by the units for generating and feeding energy into the network.

[0057] According to one embodiment of the invention, the algorithm trained and coded into a software can constitute a product which can be supplied to a client, such as an electricity distribution operator, in the form to be installed in that operator’s computer systems or in the form of a service accessible for a fee, but not residing in that operator’s systems.

[0058] This is based on the fact that, in order to train the algorithm for estimating or predicting the virtual or synthetic data relating to the consumption of a withdrawal unit of the type defined as registry, and which therefore does not have functions for measuring and storing the curves relating to the time trend of the energy withdrawn, in such a way that the performance of the algorithm and therefore the reliability of the result are relatively high, it is necessary to have a large collection of proprietary consumption data available, as in the case of the Applicant, which large data collection allows to associate with a pod that does not have consumption storage functions the consumption recorded thanks to the pods that instead do indeed have this functionality, such as the smart-meters installed. The invention therefore also concerns a method for the generation of virtual energy withdrawal and / or generation or feed units that simulate corresponding units of the type with consumption, associating with each of the registry-type energy withdrawal and / or generation units synthetic data consisting of the estimate of the time trend in a pre- established time window of the energy withdrawn and / or the energy generated or fed into the network for the corresponding registry-type units as a function of the real data of the time trend in a time window which are detected and stored in a plurality of energy withdrawal units and / or energy generation or feed into the network units of the type with consumption, the set, comprising a pre-established number of withdrawal units and / or generation units of the type with consumption and a pre-established number of registry -type withdrawal units and / or generation units, being subjected to a clustering process in which a certain number of the units of the type with consumption and a certain number of the registry-type units present in the set is attributed to one of a predefined number of clusters, and the synthetic data for a registry withdrawal unit and / or energy generation or feed into the network unit provided in one of the clusters, calculated as a function of the real data of the time trend in a time window, being detected and stored in a certain number of withdrawal units and / or energy generation or feed into the network units of the type with consumption, which units with consumption have, within the same cluster, smaller distances from the corresponding registry unit with reference to a mono or multidimensional space of representation of the variables relating to one or more representative features of the units, in particular one or more registry features.

[0059] According to one embodiment, the method for generating virtual energy withdrawal and / or energy generation or feed units provides to use a supervised algorithm of the type called k-means to perform the clustering process. According to a further characteristic that can be provided in combination with any of the aforementioned previous embodiments and / or characteristics, the method for generating virtual energy withdrawal and / or energy generation or feed units provides to preemptively perform a step of determining the optimal number of clusters for the execution of the clustering algorithm.

[0060] According to one embodiment, the aforementioned step of determining the optimal number of clusters provides to determine this optimal number by means of two different methodologies for determining the optimal number of clusters for the execution of a clustering algorithm, the number being considered optimal when the optimal number determined with the two different methods is to be considered identical within the scope of pre-established tolerances.

[0061] According to another characteristic of the method for generating virtual energy withdrawal and / or energy generation or feed units, which can be provided in combination with one or more of the previous characteristics of the method, there are provided steps of defining and / or selecting a plurality of registry features that are common to the registry units and to those with consumption, and a step of identifying and selecting, from the registry features, those registry features that most determine a differentiation between the different units provided in the network, both of the type with consumption and also of the registry type, the distribution of such units, both those with consumption and also those of the registry type, on the different clusters being determined on the basis of the, or of one or more of the, registry features that most determine a differentiation between the units.

[0062] According to one characteristic, the distribution of the units, both those with consumption and also those of the registry type, on the different clusters constitutes the initial starting distribution of the iterative clustering process of the clustering algorithm, that is, the k-means algorithm.

[0063] According to another characteristic, in combination with one or more of the characteristics and / or embodiments of the method for generating virtual energy withdrawal and / or energy generation or feed units according to one or more of the embodiments or characteristics described above, there is provided a step of validating the synthetic data relating to the time trend of the withdrawal and / or generation or feed of energy for a registry-type unit, which step provides the comparison of the synthetic data determined for pre-established time instants with the real measured data and the quantification of a distance metric of the synthetic data from the real data, on the basis of which to determine the accuracy of the synthetic data in order to integrate, for each registry unit or for a part thereof, the registry data of the registry-type unit with the synthetic data on the time trend of the energy withdrawn and / or that generated or fed into the network, simulating that the registry unit is a unit of the type with consumption.

[0064] This unit for simulating a unit with consumption can then be used for the prediction of the time trend of the energy withdrawn and / or the energy generated or fed into the network by the corresponding registry unit, according to one or more of the steps of the method described above.

[0065] As regards the metric for evaluating the aforementioned comparison, it is possible to use different metrics, both alternatively and also in combination with each other, such as for example the metrics described above with reference to the steps of the method for predicting the trend of the energy withdrawal and / or energy generation or feed into the network in future time instants, on the basis of the time trend of the energy withdrawal and / or energy generation or feed into the network measured in a pre-established time window and stored in the units with consumption.

[0066] According to one embodiment, the synthetic data relating to the estimated trend for a registry-type withdrawal unit and / or generation unit is calculated by determining the average of the time trends of the energy withdrawal and / or the generation or feed of energy into the network stored in the corresponding units with consumption selected as closest and within a pre-established time window, thus generating a fictitious time trend of the energy withdrawn and / or the energy generated and / or fed into the network relating to the registry-type withdrawal unit and / or generation unit. According to an improvement, the step of calculating the average provides, in combination, a step of filtering so-called outliers before the step of calculating the average.

[0067] According to another characteristic, the step of determining the units with consumption closest to a registry unit present in the same cluster is performed using an algorithm called KNN (K-nearest neighbors). This algorithm is known and is described for example in the document published on the web at https: / / it.wikipedia.org / wiki / K-nearest_neighbors.

[0068] The invention also concerns a system for implementing the aforementioned method according to one or more of the embodiments and variants of the method described above, which system comprises: an electricity distribution network, which network in turn comprises one or more electrical energy generation units that feed the network and a plurality of separate and independent withdrawal points of the electrical energy, in particular distributed over a pre-established geographical area, and wherein at least a part of the withdrawal points are provided with a storage unit for the electrical energy withdrawal curve measured as a function of time from the corresponding withdrawal unit, that is, the time trend of the electrical energy withdrawn from the network, and a port for reading the withdrawal curve, and wherein the withdrawal curves are stored in at least a part of the withdrawal units of the plurality of withdrawal units and are supplied to, or can be retrieved from, a central control unit, at least a part of the withdrawal units being instead without the functionalities for detecting and storing the time trend of the electrical energy withdrawal for a pre- established time window, which central control unit is further connected to measurement units of the curve as a function of the electrical energy generation time of at least part or of each of the electrical energy generation units and the operating conditions of the electrical energy units, that is, the time trend of the electrical energy generated, in the central control unit a program being stored and executable comprising the instructions that make the control unit capable of carrying out the processing steps of an algorithm for generating synthetic data of the time trend of the energy withdrawn and / or the energy generated or fed into the network for each or at least part of the withdrawal units and / or of the generation units that are not provided with the function for detecting and storing the time trend of the energy withdrawn and / or the energy generated or fed into the network, and which steps of generating the synthetic data comprise the generation of the synthetic data by means of an algorithm for estimating the synthetic data as a function of the data relating to the time trend of the energy withdrawn and / or that produced or fed into the network that are stored in each or in a part of a plurality of different withdrawal units and / or different energy generation units and which units are provided in a pre- established boundary with respect to the individual withdrawal units and / or the respective generation units which are without the detection and storage functions, the steps of generating the synthetic data being carried out before the calculation of the prediction for each or a part of the energy withdrawal units and / or of the energy generation or feed into the network units, and the synthetic data generated being associated, respectively, with each or with a pre- established part of corresponding energy withdrawal units and / or energy generation or feed into the network units, and the synthetic data being used for calculating the prediction of the trend of the energy withdrawn from the network relating to the withdrawal units and / or the generation or feed into the network units which are without the functionalities for detecting and storing the time trend of the withdrawal of energy and / or the generation or feed of energy into the network, respectively.

[0069] According to a further characteristic, a program is further stored in a memory of the central control unit and is executable for predicting the time trend of the energy consumption of at least part, preferably of each, of the withdrawal units within a pre- established future time period and / or for predicting the time trend of the energy generated by at least part, or by each, of the electrical energy generation units connected to the network in the pre-established future time period, which prediction program comprises the instructions for the central control unit’s processor which are related to the execution steps of an algorithm predictive of the time trend of the consumption relating to at least part, or to each, of the withdrawal units and / or predictive of the time trend of the quantity of energy generated by at least part, or by each, of the generation units in the pre- established future time period as a function of the curves as a function of the time of withdrawal and / or generation, respectively, of the electrical energy by at least said part, or by each, of the withdrawal units and / or the generation units stored therein and relating to a time period prior to the future time period, these predictions being carried out for each, or at least part, of the units that are provided with the functionalities for detecting and storing the time trend of the energy withdrawn and / or that generated or fed into the network, and also for the units that are not provided with the functionalities for detecting and storing the time trend of the energy withdrawn and / or that generated or fed into the network, on the basis of these predictions there being activated units for compensating for the transient peaks of electrical energy withdrawal and / or electrical energy generation synchronized with the time instant in which these transient peaks are expected.

[0070] The compensation units can consist of a plurality of electrical energy storage / transfer flywheels, such as for example one or more storage / transfer stations, comprising one or more electric accumulators or similar devices and / or one or more units of electrical energy generators comprising electromechanical transducers.

[0071] Alternatively or in combination, the distribution network can be provided in combination with a structural configuration that provides the division of the distribution network comprising a plurality of electrical energy withdrawal units and / or a plurality of electrical energy generation units into two or more sub-networks, each of which subnetworks is characterized by a different geographical location area of the corresponding withdrawal and / or generation units, and also by the number of such units, as well as the type of withdrawal units with reference to the energy withdrawal curves of the individual withdrawal units, as well as by the type of electrical energy generation units, while between one or more of the sub-networks there are provided electrical connections able to be switched between a closed and an open condition, the switching state of the one or more connections between the sub-networks being automatically commanded into a closed or open condition as a function of the results of the prediction relating to the withdrawal of electrical energy and / or the generation of electrical energy for the future time period, and the closing and / or opening condition of the connections between the subnetworks being varied in the future time period as a function of the prediction of the time trend of the withdrawal and / or the time trend of the generation of electrical energy for the various sub-networks.

[0072] The invention therefore also concerns a software stored on a stable memory medium and / or on a portable memory medium or on a memory medium existing in the cloud, which software comprises the instructions for a processor or a combination of processors to perform the steps of the method according to one or more of the embodiments and / or characteristics described above, or combinations thereof.

[0073] BRIEF DESCRIPTION OF THE DRAWINGS These and other characteristics and advantages of the present invention will become apparent from the following description of a non-restrictive example embodiment, shown the attached drawings wherein:

[0074] Figure 1 shows a high-level block diagram of an electricity distribution network provided with a plurality of electrical energy withdrawal units and a plurality of electrical energy generation units and, in combination, with a central control unit according to an embodiment of the present invention.

[0075] Figure 2 shows a high-level block diagram of an example embodiment of a unit for withdrawing energy from an electricity distribution network, which unit is of the type defined as “with consumption”.

[0076] Figure 3 shows a high-level block diagram of an example embodiment of a unit for generating energy from an electricity distribution network, which unit is of the type defined as “with consumption”. Figure 4 shows a high-level block diagram of an example embodiment of a central control unit according to the present invention.

[0077] Figure 5 shows an example of curves representing the daily time trend of the energy withdrawn by a withdrawal unit on different dates.

[0078] Figure 6 shows a block diagram of the steps according to an embodiment of the prediction method performed by the system of the present invention.

[0079] Figure 7 shows a flowchart of the method according to the present invention for generating virtual data relating to the estimated trend of the energy withdrawal and / or energy generated or energy fed into the network, based on the time trends stored in the corresponding units of the type with consumption, wherein the process is of the sequential type.

[0080] Figure 8 shows a variant of the method according to figure 7, in which the steps of determining the synthetic data for the registry-type units of the individual clusters are performed in parallel for each cluster.

[0081] Figures 9 and 10 graphically show two examples of determination of the optimal number of clusters based on the method called “Elbow” and on the method called “Silhouette”.

[0082] Figure 11 schematically shows a possible combination of registry features common to the registry units and those with consumption that can be selected to characterize the units for the purposes of the clustering. Figure 12 shows a diagram showing the relevance of the different registry features for the purposes of differentiating the individual registry units and the units with consumption from each other.

[0083] Figure 13 schematically shows the result of the clustering and an embodiment of the process for selecting the units with consumption used for the calculation of the synthetic data relating to a registry unit present in a same cluster.

[0084] Figure 14 shows an example of integration of the registry features of a registry unit with the synthetic data relating to the time trend of the withdrawals and / or of the energy generated or fed into the network for the transformation of the registry unit into a unit with consumption of the synthetic or virtual type.

[0085] DESCRIPTION OF SOME EMBODIMENTS

[0086] In some embodiments, a control method of an electricity distribution network provides to estimate the future consumption of an electricity user device (POD) by means of a specific algorithm, e.g. of the gradient boosting type, which receives as input (i) historical consumption data and (ii) registry data (position, type, etc.) of the user device. In the event that historical data on the user device’s consumption is not available, then the method described here provides to use historical data of a virtual user device. The virtual user device’s historical data is calculated as follows:

[0087] - first, the user devices served are divided into clusters based on the registry features of the user devices, e.g. ZIP code and type of meter (3kWh, 6kWh);

[0088] - in the cluster of the user device of interest, N user devices are selected that are closer, that is, they are less than a threshold distance away from it. The distance being defined in the space of the registry features used to define the clusters;

[0089] - the historical data of the N user devices selected are taken;

[0090] - the historical data of the virtual user device are calculated starting from (e.g. averaging) those of the N user devices selected.

[0091] In some embodiments, the clustering depends on the user devices’ registry features.

[0092] In some embodiments, the method described here makes estimates of future consumption using an algorithm (e.g. a gradient boosting algorithm), and builds the data of a virtual user device to use it as historical consumption data of the user device on which the prediction is made.

[0093] With reference to figure 1, this schematically shows an electricity distribution network that is divided into a plurality of sub-networks SUBnet 1 , SUBnet 2 to SUBnet N, wherein N is any natural number whatsoever.

[0094] Each sub-network comprises a plurality of withdrawal units indicated as POD 1 , POD2, POD3, POD4, PODn, wherein n is a natural number greater than 4.

[0095] Similarly, each sub-network can comprise one or a plurality of electrical energy generation units indicated as Genl, Gen2, Gen3, Gen4, Genn, wherein n is any natural number whatsoever greater than 4.

[0096] The above notation is intended to indicate that each sub-network can have a pre- established number greater than 1 of electrical energy withdrawal units, generically called PODs, and any number whatsoever of electrical energy generation units. The electrical energy generation units are shown for completeness in the embodiment of figure 1 , but may also not be provided in one or more or all of the sub-networks.

[0097] Typically, the withdrawal units consist of the so-called “PODs” which are associated with an electrical energy supply contract, the conditions of the supply and the subjects to whom the supply is intended, as well as the payment tariffs for the energy supplied and the payment terms, and a geographical location.

[0098] With regard to the generation units, these are increasingly widespread and serve the same user devices served by a withdrawal unit. There is no limitation regarding the type of electrical energy generators which can be, for example, photovoltaic panels and / or wind generators and / or even hydrodynamic generators, or other types of so-called energy harvesting systems.

[0099] As can be seen from the block diagram, some of the withdrawal units and some of the energy generation units, and in particular those indicated with POD1, POD3, Genl and Gen3 in the three sub-networks, do not have connections with the network. This represents the fact that there is no possibility for these units to communicate with the central control unit 110. However, the connection to the sub-network, with regard to the lines for supplying the electrical energy through the POD unit and the lines for feeding the electrical energy generated by the Genl and Gen3 units remains present, although not explicitly shown. Furthermore, for simplicity, the same units have been highlighted in the various subnetworks as being without the functionality for communicating with the central unit, this however is only for ease of description and has no relation to the network’s technical or configuration characteristics, the units without the communication functionality being any units whatsoever and also provided in any number whatsoever in each sub-network. The units without communication functionality indicate the so-called registry-type units which, unlike those of the type with consumption, do not comprise members that allow to detect the time trend of the energy withdrawn and / or that generated and fed into the network, and which would therefore be excluded from the possibility of calculating a prediction, for the same units, of the future trend of the energy withdrawn and / or that generated or fed into the network.

[0100] Therefore, communication with the central unit 110 in order to access the curves of the time trend of the energy withdrawn and / or the energy generated or fed into the network that are stored in the corresponding unit is not possible, and this is evidenced by the graphic notation of the drawing.

[0101] According to a non-limiting embodiment, the example embodiment shown further presents a plurality of change-over switches, indicated overall by the block 100, which change-over switches allow two or more sub-networks to be connected to each other, or to be connected to or isolated from each other. The change-over switches are an example of implementation of members that compensate network instability caused, for example, by transient peaks in the withdrawal of electrical energy or the feed of electrical energy generated by the one or more generation units into the network.

[0102] An alternative provided in the present example configuration can provide one or more energy accumulators 101, 201, M which are stably connected to one or more of the subnetworks, or which can be connected, each one or in any combination of them whatsoever, to one or more of the sub-networks by means of associated connection / isolation switches 100.

[0103] By the term “energy accumulators” we mean energy storage devices in a broad sense, which are not limited to static accumulators of the electrochemical type but also comprise inertial-type accumulators, such as electromechanical accumulators or suchlike, which in some cases can also consist of the alternators of wind generators connected to the network.

[0104] The individual withdrawal units POD2, POD4, PODn and / or one or more generators Gen2, Gen4, Genn, on the other hand, have communication interfaces for communicating with a central control unit 110, these units being of the type defined as units with consumption, that is, provided with devices for detecting and storing the time trend of the energy withdrawn from the network and / or that generated or fed into the network, respectively.

[0105] With reference to the withdrawal units of the type with consumption, these are provided with a management processor that runs a firmware comprising the instructions to make the management processor capable of performing the functionalities provided for the withdrawal units.

[0106] According to an essential characteristic of the present invention, the withdrawal units with consumption comprise a meter of the time trend of the withdrawal of electrical energy or, alternatively, they are configured by means of the firmware to execute that function.

[0107] The measured values are stored as a function of time, in the form of energy withdrawal time curves in a memory of the corresponding withdrawal unit, which memory is accessible to the central control unit 110 by means of the communication interface indicated above.

[0108] The central control unit 110 is further connected with maintenance activity planning units 120 and / or an upgrade planning unit 130 and / or a unit for verifying the accuracy of the consumption detected for each withdrawal and / or generation unit 140.

[0109] Similarly, it is also possible to provide that the energy generation units of the type with consumption are provided with local management units that are provided with sensors to measure the time trend of the energy generated and fed into the respective sub-network. Also in this case, the sensors can be of a hardware type controlled by the generation unit’s management unit, or they can consist of a firmware that configures the management unit in such a way as to make it capable of performing the measurement functions.

[0110] The time trend of the energy generated by a generation unit is stored in a memory of the same generation unit, which memory is accessible to the central control unit by means of a communication interface for communicating with each or at least part of the electrical energy generation units provided in the sub-networks.

[0111] With reference to figure 2, this shows a non-limiting embodiment example of a possible configuration of a withdrawal unit of the type with consumption according to the present invention.

[0112] The withdrawal unit comprises a CPU 200 which runs a firmware for managing various peripherals provided in the withdrawal unit. The firmware is stored in a dedicated memory or in a dedicated memory area of a common mass memory of the withdrawal unit, indicated with 210. It is possible to communicate with the CPU 200 and with the various peripherals by means of a human-machine interface indicated overall with 220. This can comprise various devices, such as display devices, selection devices, data string input devices and / or command devices. These devices can consist of one or more humanmachine interface devices provided in current computers or computer systems.

[0113] The CPU 200 controls a unit 230 for measuring the energy delivered by the withdrawal unit for a pre-established time interval. In addition, the CPU 200 controls a communication interface, preferably for bidirectional communication with the central control unit 110, which communication interface is indicated with 240.

[0114] In dedicated memories 250 and 260, or in dedicated memory areas of a common mass memory, there are respectively stored the data relating to the curves of the time trend of the energy that has been withdrawn by means of the withdrawal unit, or delivered thereby, and optionally diagnostic data that indicate the functional status of the withdrawal unit itself and / or of one or more of the peripherals that compose it.

[0115] This hardware configuration can also be provided for the energy generation units of the type with consumption, being suitably configured to operate in combination with electrical energy generation devices. These devices can be of different types and operate according to different physical and / or physical / chemical principles, and are indicated overall by block 370. Even in the case of the electrical energy generation units, these comprise a management processor 300 that controls the functionalities of the various peripherals provided in such units. In particular, according to the more generic embodiment shown in figure 3, the functional control of the peripherals is implemented by means of a firmware that is stored in a dedicated memory or in a dedicated memory area represented by block 310.

[0116] The CPU 300 controls a unit 330 for measuring the energy produced by the generator 370 and fed into the network during a pre-established time interval. In addition, the CPU 300 controls a communication interface, preferably for bidirectional communication with the central control unit 110, which communication interface is indicated with 340.

[0117] It is possible to communicate with the CPU 300 and with the various peripherals by means of a human-machine interface indicated overall with 320. This can comprise various devices, such as display devices, selection devices, data string input devices and / or command devices. These devices can consist of one or more human-machine interface devices provided in current computers or computer systems.

[0118] Dedicated memories 350 and 360, or dedicated memory areas of a common mass memory, are provided in which there are respectively stored the data relating to the curves of the time trend of the energy that has been generated and fed into the network by means of the generation unit, or by the generator 370, and optionally diagnostic data that indicate the functional status of the generation unit itself and / or of one or more of the peripherals that compose it.

[0119] On the other hand, the corresponding withdrawal units and the corresponding generation units of the registry type differ structurally from what is shown in figures 2 and 3, respectively, at least in the fact that they do not provide a unit 230 for measuring the energy delivered for a pre-established time interval and a memory 250 for the curves of the time trend of the energy withdrawn, and at least in the fact that they do not have a memory 350 for the curves of the time trend of the energy generated or fed into the network.

[0120] Obviously, it is possible for the registry units to have configurations in which further units, shown in the configurations of figures 2 and 3 for the corresponding units of the type with consumption, are missing. Figure 4 shows a high-level block diagram of an embodiment of the central control unit. Also in this case, the central control unit 110 has at least one CPU 400 that manages several peripherals. In different areas of a common mass memory, or in dedicated memories indicated with 410, 450, 460, 480, there are stored, respectively, a firmware for managing the peripherals that contain the instructions for making the peripherals capable of performing the functions provided, for training the predictive algorithm, a software comprising the instructions for making the CPU capable of performing the steps provided by the prediction algorithm, and optionally diagnostic data relating to the central control unit’s peripherals or other units associated with or controlled by the central control unit.

[0121] It is possible to communicate with the CPU 400 and with the various peripherals by means of a human-machine interface indicated overall with 420. This can comprise various devices, such as display devices, selection devices, data string input devices and / or command devices. These devices can consist of one or more human-machine interface devices provided in current computers or computer systems.

[0122] The CPU 400 controls one or more drivers 430 for commanding one or more members for automatic intervention on the network for the execution of operations aimed, for example, at compensating for conditions of network instability and / or transient peaks of network load and / or feed of energy into the network itself, such as for example those described with reference to figure 2, and which are in the form of servo-controlled change- over switches and / or stations for storing / releasing electrical energy. In addition, the CPU 400 controls a communication interface, preferably for bidirectional communication with the central control unit 110, which communication interface is indicated with 440 and through which it can access the memories of the withdrawal units and / or of the generation units for the collection of the data relating to the time trend curves of the energy withdrawn by the withdrawal units and that generated and fed into the network by the generation units.

[0123] According to yet a further characteristic, the central control unit can provide a memory, or a memory area, indicated with 490 and in which there is stored a software for planning network interventions and / or inspections for verifying the actual conditions of the network and / or of the withdrawal units and / or of the generation units.

[0124] With reference to the data stored in the memories of the withdrawal units relating to the curves that represent the time trend of the energy withdrawn through the withdrawal units, figure 5 shows these curves that represent the electrical energy withdrawn from the network through a withdrawal unit and over the course of a time interval of essentially one day and on different dates.

[0125] Again, according to a further characteristic, the central control unit comprises a comparator 491 which can be either of the hardware or software type and which is configured to perform a comparison between the time trends of the energy withdrawn and / or that generated of a pre-established withdrawal unit and / or of a pre-established generation unit, as defined by the predictive algorithm, and the data actually measured in coincidence with the time instants which these values refer to.

[0126] In figure 4, the central control unit 110 further provides a dedicated memory or a dedicated memory area indicated with 492, in which a program is stored that encodes the instructions for the CPU 400, the execution of which makes the processor and the corresponding peripheral units capable of performing the steps of a method for transforming the registry units into virtual units of the type with consumption, using the time trend of the energy withdrawn and / or of the energy generated or fed into the network in a pre-established time window stored in a pre-established number of units of the type with consumption to estimate synthetic data of the time trend of the energy withdrawn and / or that generated or fed into the network for the registry-type units.

[0127] Figure 6 shows a flow chart illustrating the steps of the method for predicting the energy withdrawn by each or at least part of the withdrawal units and / or the energy generated or fed into the network by each or at least part of the generation units that is the object of the present invention and is implemented by means of the system according to the present invention. The prediction method is applied to all the units present in the network, both those of the registry type and also those of the type with consumption, integrating the registry features that define the individual units of the type with consumption and of the registry type with the synthetic data of the trend of the energy withdrawn and / or of the energy generated or fed into the network calculated for the individual registry-type units, on the basis of the aforementioned trends stored for the same time window in the units of the type with consumption and selected according to pre-established criteria among the units with consumption present in the network. The process of selecting the units with consumption and the method for determining the data relating to these trends stored in the units with consumption are also the object of the present invention and are described in more detail in the examples of figures 7 to 14, while in the prediction process according to the example of figure 6, the prediction is carried out for all units, both those of the type with consumption and also those of the registry type, the latter being transformed into synthetic or virtual units of the type with consumption, by associating the registry-type characterization variables with the synthetic data relating to the time trend of the energy withdrawn and / or that generated or fed into the network calculated as generically indicated above and according to the method of the present invention, of which example embodiments are shown in figures 7 to 14.

[0128] As already described above, the method according to the embodiment of figure 6 provides to perform a prediction of the trend of the withdrawal of electrical energy and / or of the feed of electrical energy in one or more future time instants or periods, which prediction is based on historical data of the these trends of each or all the withdrawal and / or generation units, both of the registry type and transformed into virtual units of the type with consumption, and also of the type with consumption, that is, of real stored data or estimated synthetic data relating to the time trend of the withdrawal of electrical energy and / or the feed of electrical energy by each of at least a part of the electrical energy withdrawal units and / or generation units, respectively of the type with consumption and registry, provided in an electric distribution network and / or in a sub-network thereof, in a time window prior to the instant such data is collected.

[0129] In order to perform the aforementioned prediction, a preferred embodiment of the invention provides to use a machine learning algorithm suitably trained on the basis of this data.

[0130] Among the various possible machine learning algorithms, it has proven advantageous to use an algorithm called “gradient boosting”, which determines the predictive values by means of a regression based on known data. The algorithm is known, and more detailed information can be found from various publications, including the one referenced above.

[0131] Since the acquisition of the data relating to the time trend of the energy withdrawn by the individual withdrawal units and / or fed into the network by the individual generation units continues according to a pre-established measurement sequence, the data on the electrical energy withdrawn and / or on that generated which are calculated by means of the predictive algorithm can in turn be compared with the data actually measured at the future time instants or in the future time periods. This allows to verify the accuracy between these values, from which not only to test the quality of the prediction, but also to extract indications on possible anomalies that may be related to failures of the units that make up the network infrastructure and / or even related to malicious energy withdrawal actions that exclude the withdrawal devices.

[0132] In addition, the predictions also allow to deal in a timely and near real-time manner with the occurrence of conditions of instability or excessive load on the network that may compromise its operation or cause damage or other malfunctions. There are multiple possible variants. Figure 6 shows the block diagram of an example embodiment and, in relation to what has been illustrated, we will indicate the possible optional steps of the process which in the drawing are instead explicitly shown for the purpose of completeness of the embodiment with reference to the possible variants.

[0133] The method shown is a dynamic method that provides to measure the time trend of the electrical energy withdrawn by each of the withdrawal units that are provided in a network or in a sub-network thereof, or by a part of these units in a time window of pre-established length, that is, which, with respect to a pre-established instant, extends backward in time for a pre-established time distance.

[0134] The time window typically has a substantially predetermined length that can be modified but is kept fixed during the execution of the method and moves forward in time continuously or at pre-established time intervals, so that the stored data are constantly updated by adding new data measured in the time interval in which the time window advances, and by deleting the oldest data.

[0135] Therefore, according to this embodiment, the process begins as shown in step 600, with the definition of the time sequence of the movement in time of the time window for the storage of the data deriving from the measurement of the energy withdrawn from the network and / or the energy fed into the network for each or for part of the withdrawal units and / or the generation units.

[0136] In step 601, a time sequence of instances in which the stored data are red is defined. In each instant of the reading sequence, the data stored by the individual withdrawal and / or generation units present in the network or in a sub-network are retrieved.

[0137] Steps 602 through to 604 are optional steps which, however, in the most refined and preferred embodiment, integrate the information relating to the real stored data for the units of the type with consumption and to the synthetic data estimated for the registrytype units relating to the energy withdrawn and the energy generated and fed into the network, with information useful for generating more reliable predictions.

[0138] In particular, in step 602, so-called registry data are added to the data obtained from the curves of the time trend measured for the energy withdrawn and / or that fed into the network. These data are fixed and include the indications relating to the settings of the withdrawal units in relation to the maximum values of energy that can be withdrawn by them, and to other settings of the withdrawal and / or generation units.

[0139] In step 603, information regarding the geographical positioning of the various withdrawal and / or generation units is added while in step 604, time data are added, such as month, week, time.

[0140] These data are supplied to the prediction algorithm for its training in the execution of the predictive calculation, and in particular for the calculation of the prediction values by means of a regression process performed in particular and preferably by means of a socalled gradient boosting algorithm, as indicated in steps 605 and 606.

[0141] In step 607, a time period is defined for which to calculate, by means of the predictive algorithm, the values of the time trend of the energy withdrawn and / or that generated by a certain withdrawal unit and / or by a certain generation unit. In step 608, the algorithm is executed ,and the prediction curves are generated for the time trend of the energy withdrawn and / or that fed into the network in the prediction time period defined in step

[0142] 607.

[0143] Subsequently, the data subject to the prediction undergo post-processing in order to assess the need to perform planning for, or changes to, actions of maintenance and / or upgrade and / or stabilization of the network’s operating conditions, as indicated in step 609.

[0144] The process of measuring the data relating to the time trends of the energy withdrawn and of that fed for each withdrawal unit and for each generation unit of the type with consumption, and the process of calculating the synthetic data relating to the time trends of the energy withdrawn and of that fed for each withdrawal unit and for each generation unit of the registry type, is continued according to a time sequence of forward movement of the time window, defined in step 600. The data measured and stored in the memories of the withdrawal units and / or of the generation units, as indicated in step 610, and also the synthetic data estimated for the registry type units, are then updated, while for the data measured at the same future time instants or in the future time periods for which the prediction values have been calculated, in step 611, it is provided to carry out a comparison between the data actually measured at these instants and the corresponding data supplied by the prediction, the results of this comparison can be used either to verify the performances of the prediction algorithm, and possibly make corrections, as indicated in step 612, or they can be used to perform a step 613 of verifying the accuracy of the energy consumption of the individual withdrawal units and / or of the energy fed by the individual generation units with predictive data, in order to establish whether the network’s infrastructure is malfunctioning and / or damaged, or whether the energy withdrawal or the energy fed are subject to abusive withdrawal activities that exclude the passage through the electrical energy withdrawal units or generation units, for which a step of implementing or planning control inspections is performed, as indicated in step 614.

[0145] Furthermore, the results of the comparison relating to the registry-type units can also be used for a verification of the accuracy of the synthetic data generated by means of the estimation process described in detail below.

[0146] It is possible to set a threshold value for the result of the accuracy verification that can be, for example, in the form of a comparison between values obtained with the prediction process and values actually measured. When this threshold value is exceeded, then it is possible to provide a step of planning an inspection at the withdrawal unit and / or the generation unit for which this comparison has been calculated. This step is indicated with reference number 614.

[0147] One embodiment provides that the measurement of the energy withdrawn by a withdrawal unit and / or that fed into the network by a generation unit is carried out at intervals comprised between 1 and 15 minutes, while the time window within which to store the measured data is of about two months from the time of the last measurement.

[0148] The time frequency for updating the synthetic data relating to the time trend of the energy withdrawn and / or that generated or fed into the network by the registry-type units can also be different from that provided above for the update, that is, the forward shift of the time window for detecting and / or storing the time trend of the energy withdrawn and / or that generated or fed into the network by the units with consumption.

[0149] Also in one embodiment, the time period for which to determine the prediction of the time trend of the energy withdrawn and / or that fed can vary from about one day up to a few days, for example three or more days.

[0150] With reference to figures 7 to 14, these refer to some embodiments of the estimation method, that is, the method for generating synthetic data relating to the time trend of the energy withdrawn and / or that generated or fed into the network in the pre-established past time window for the registry-type units, in order to transform them into synthetic or virtual units of the type with consumption usable in the prediction process, such as for example that of the embodiment described with reference to figure 6, in a similar way and together with the real units of the type with consumption.

[0151] The method generally provides steps of subjecting the units of the type with consumption and the registry-type units provided in the network to a clustering process on the basis of a pre-established number and a pre-established type of registry features that define mono or multidimensional spaces in which to place the individual units and in which to define metrics for the distance between the units of the registry type and those of the type with consumption, following the clustering process, defining within the individual clusters a pre- established number of units with consumption selected from those that, in the mono or multidimensional space of the features, are closest to a target registry-type unit, repeating this step for each registry-type unit in each of the clusters generated in the step of executing the clustering process, using the real data relating to the time trends of the energy withdrawn and / or that generated or fed into the network detected and stored in the selected units of the type with consumption, for each of the registry-type units to calculate synthetic data relating to the time trend of the energy withdrawn and / or that generated or fed into the network for the corresponding registry-type unit, and associating these synthetic data with the characterization space of the registry unit, transforming it into a synthetic or virtual unit of the type with consumption to be used in a similar way to the units of the type with consumption for the prediction step, as described above according to an embodiment of the prediction step.

[0152] A non-limiting example embodiments of the steps of the process for determining the synthetic data relating to the time trend of the energy withdrawn and / or that generated or fed into the network for each or for at least part of the registry-type units present in the network is shown in figure 7 and refers to the use of a clustering algorithm applied to the units of the type with consumption and to those of the registry type, using a representation of such units on the basis of variables relating to one or to a plurality of registry-type features.

[0153] The meaning of a registry-type feature will be defined in greater detail below, and generally refers to physical features that define the type of supply and / or geographic positioning relating to each of the units, and which are common both to the units of the type with consumption and also to the registry-type units.

[0154] Steps 700 to 703 define a preparatory step to the training and execution of a clustering algorithm, for example an algorithm called k-means provided for a preferred embodiment of the present invention.

[0155] The choice of the optimal number of clusters has important consequences for the convergence toward a definition of stable centroids that characterize the different clusters in a supervised clustering process, in which the iteration of the clustering steps is terminated when, for successive steps, the variation of the centroids in a space of the features is minimal, that is, below a pre-established threshold value.

[0156] In step 700, the procedure is started to determine the optimal number of clusters on which to perform the subsequent step of clustering the units of the type with consumption and those of the registry type which are provided in the network, or in a part thereof. In the state of the art, there are several methods for determining the optimal number of clusters. A description of these methods is published in the document “Review on Determining of Cluster in K-means Clustering” Trupti m Kodinariya, Prashant Makwana published at the web address https: / / www.researchgate.net / publication / 313554124_Rev iew_on_Determining_of_Cluster_in_K-means_Clustering.

[0157] According to a characteristic of the method of the present invention, two known techniques are used in parallel to determine the optimal number of clusters, which are called the Elbow Method and the Silhouette Method, respectively, and which are performed in steps 701 and 702. The outcome of the determination of the optimal number of clusters according to these two technologies is compared in step 703. If the values determined in steps 701 and 702 differ to an extent that goes beyond a pre-established tolerance, it is possible to repeat the process as shown in figure 7, possibly and optionally replacing one of the two methods with another different method, or it is possible to determine an optimal number of clusters in the form of the integer value of the average between the results of the two technologies.

[0158] Figures 9 and 10 graphically show how to determine the optimal number of clusters according to the two methods called “Elbow” and “Silhouette”.

[0159] The above methods provide to evaluate the effect of the number of clusters relative to a cost function to be minimized, which cost function relates to a measure of the global variation in the position of the centroids. As can be seen from the graph in figure 9, the method called “Elbow” provides to select, as the optimal number of clusters, the number of clusters for which the curve of the minimums of the cost function as a function of the number of clusters passes from a steeper inclination to a less steep inclination, that is, to a zone in which, as the number of clusters varies, the minimum of the cost function remains substantially stable or only varies slightly. The arrow indicates the value of the point corresponding to the “elbow” and the optimal value of the number of clusters.

[0160] Since the trend of the curve in figure 8 can be very progressive as it passes between the two branches, steeper and less steep, the determination of the “elbow point” can be very inaccurate.

[0161] As figure 10 shows, in accordance with step 702, the “Silhouette” method is based on a different approach for determining the optimal number of clusters according to a different cost function, therefore it offers an alternative to the Elbow method thanks to which the inaccuracies of this method can be resolved.

[0162] Figures 9 and 10 show that, for an example embodiment, the optimal number of clusters is 8.

[0163] Once the optimal number of clusters has been defined, the method provides a step of defining and selecting a plurality of features representative of the energy withdrawal and / or generation units of the registry type and of the type with consumption, and of defining corresponding parametric variables descriptive of such features, as indicated with 704. In particular, these features are registry-type features, as defined above, and are common both to the units of the type with consumption and also to the registry-type units.

[0164] Figure 11 shows a dataset of registry features for both the units with consumption and also the registry units, which registry features specifically comprise position features such as ZIP code, administrative features such as tariff type, support for daily closure, type of supply, district number, forfeit supply, geographical range code, technical features of the supply such as supply voltage, power limitation, available power, power reduction percentage, power limitation profile, climatic features such as description of the duration of the day, seasonal conditions. This example is to be considered as non-limiting, but exemplary, and other features can be provided in addition to those shown. However, the registry features defined in step 704 have a different influence as regards the effectiveness in differentiating; therefore, as shown in figure 7, there is provided a step 705 of determining one or more features of high discriminatory degree for the allocation of the units to the different clusters. Figure 12 shows a diagram in which the relevance of the different registry features, for the purposes of discriminating between clusters, is reported for each of the features as defined according to the example of figure 11.

[0165] The various units present on the network, both of the type with consumption and also of the registry type, are therefore divided into clusters, and in particular into the 8 clusters, on the basis of the values of the parameter relating to at least one of the most relevant discriminatory features. According to the example in figure 12, the registry feature with the greatest relevance for the purposes of discriminating and therefore defining the criteria for assigning units to the clusters is the ZIP code, that is, an indication of geographical positioning with reference to the territory over which the electricity network extends. This distribution of units performed across eight different clusters on the basis of the feature relating to the location, defined on the basis of a ZIP code, constitutes the starting configuration of the iterative process of the clustering algorithm, and in particular of the preferred algorithm called k-means, and in step 706 the algorithm is executed, generating eight different clusters characterized by a specific geographical position and which each contain the units present in the geographical zones coinciding with the clusters.

[0166] The process of generating the synthetic data relating to the time trend of the energy withdrawn and / or that generated or fed into the network by a registry unit, provides that for each cluster the registry-type units present in the cluster are selected, as indicated in step 708, while following the definition of a metric for the distance from the units of the type with consumption present in the same cluster defined in step 709, a certain number of units of the type with consumption that are closer to a specific registry-type unit is selected, while in step 710 a time window located in a past time period is defined and the curves relating to the time trends of the energy withdrawn and / or of the energy produced or fed into the network are retrieved from the memories of each of the units with consumption selected, as indicated in step 710.

[0167] On the basis of this real data obtained for the units with consumption selected, in step 711 the synthetic data of the time trend of the energy withdrawn and / or of the energy produced or fed for the time window are then calculated as a function of the curves relating to the time trends of the energy withdrawn and / or of the energy produced or fed into the network from the memories of each of the units with consumption, selected as per steps 709 and 710.

[0168] Once the synthetic data for a registry unit have been obtained, step 712 provides that these are used to generate a virtual unit with consumption corresponding to the registry unit, integrating the registry features thereof with the synthetic data of the time trend of the energy withdrawn and / or of the energy produced or fed for the time window.

[0169] In the example of figure 7, the process of selecting a registry-type unit in the various clusters is followed sequentially for each registry unit of a cluster and for the clusters, whereby step 713 is provided which provides to verify whether all the registry units of all the clusters have undergone the process of generating the corresponding synthetic data and of integrating them with the registry data for the generation of a virtual unit of the type with consumption. If the result of the verification is negative, the cycle is repeated by searching each of the clusters for an additional registry unit that has not yet been processed, then returning to step 708; while if it positive, it is possible to provide a step 714 of generating a database for the prediction of the energy withdrawn and / or the energy generated or fed into the network both of the real units of the type with consumption and also of the registry-type units in the form of the virtual units of the type with consumption.

[0170] Figure 8 shows a variant of the process, in which steps 800, 801, 802, 803 correspond, albeit with different notations, essentially to the steps that precede the execution of the clustering algorithm.

[0171] Unlike the embodiment of figure 7, the process of selecting the registry units for selecting the pre-established number of units with consumption that are closest to the individual registry units present in the cluster, and therefore the process of calculating the synthetic data for the respective registry units of the individual clusters, are performed in parallel for each of the eight clusters shown.

[0172] In this embodiment, moreover, the number of units with consumption to be considered for each registry unit is established at the value of 5, while the determination of the five units closest to a registry unit is performed using an algorithm of the type KNN, (K- nearest neighbor algorithm) as shown with the blocks 804.

[0173] Furthermore, for each registry unit, as indicated in blocks 805, the synthetic data are calculated as the average of the time trends in the pre-established time window of the energy withdrawal and / or of the energy generated or fed into the network of the 5 nearest units with consumption determined in step 804.

[0174] The synthetic data therefore represents an estimate of the history of the trend of the energy withdrawn or of that generated or fed into the network for each of the registry units present in the clusters, as shown in block 806. These data are then integrated with the registry features of the registry-type units in order to generate the virtual units with consumption and to be used in the algorithm model that generates the prediction at the future instants of the trend of the energy withdrawn and / or that generated or fed into the network, both for the units with consumption and also for the registry-type units. This is summarized in blocks 807 and 808. Using intuitive graphics, figure 13 shows the result of the clustering process, by depicting, on the left part, the 8 clusters on which both the units with consumption and also the registry-type units are distributed.

[0175] While the right part graphically shows the step of determining the five units closest to a target registry unit in a same cluster by applying a KNN type algorithm. Figure 14 instead shows an example of the combination of registry features and the features related to the synthetic data that make up the combination of features of a registry unit transformed into a virtual unit of the type with consumption to be used in the process of predicting the load curves referred to, for example, in the embodiment described with reference to figure 6.

Claims

CLAIMS1. Control method of an electricity distribution network, which network comprises one or more electrical energy generation units which feed the said network and a plurality of separate and independent withdrawal points of the said electrical energy, in particular distributed over a pre-established geographical area and wherein at least a part of the withdrawal points is provided with a storage unit for the electricity withdrawal curve measured as a function of time from the corresponding withdrawal unit and with a port for reading said withdrawal curve and wherein the withdrawal curves stored in at least a part of the withdrawal units of the said plurality of withdrawal units are supplied to a central control unit, and wherein at least a further part of the said withdrawal units is not provided with a storage unit for the electricity withdrawal curve measured as a function of time from the corresponding withdrawal unit, which central control unit is further connected to measurement units for the electricity generation curve of each of the said electrical energy generation units and for the operating conditions of the said electrical energy units, a control program being stored and executable in said central control unit for controlling the electricity distribution network as a function of the withdrawal curves of the individual withdrawal units and of the electricity generation curves of the individual generation units, which method further provides steps of generating synthetic data of the time trend of the energy withdrawn and / or the energy generated or fed into the network for each or at least part of the withdrawal units and / or generation units that are not provided with the function of detecting and storing the time trend of the energy withdrawn and / or the energy generated or fed into the network and which steps of generating said synthetic data comprise the generation of said synthetic data by means of an algorithm for estimating said synthetic data based on the data relating to the time trend of the energy withdrawn and / or that produced or fed into the network which are stored in each or a part of a plurality of different withdrawal units and / or different energy generation units and which units are provided in a pre-established boundary with respect to the individual withdrawal units and / or the respective generation units which are without said detection and storage functions, the steps of generating said synthetic data being carried out before calculating the prediction for each or a part of the energy withdrawal units and / or the energy generationor feed into the network units and said synthetic data generated being associated respectively with each or a pre- established part of corresponding energy withdrawal units and / or generation or feed into the network units and said synthetic data being used for the calculation of the prediction on the trend of the energy withdrawn from the network relating to the said withdrawal units and / or generation or feed into the network units which are without the functionality for detecting and storing the time trend of the withdrawal of energy and / or the generation or feed of energy into the network, respectively, wherein the registry-type units and the units of the type with consumption are divided into a pre-established number of clusters by means of a clustering algorithm, the computation of the synthetic data, i.e. of the synthetic values, being executed for the registry and with consumption units present in each cluster and selecting from these a pre- established number of units with consumption closest to a registry unit present in the same cluster, these steps being carried out for each cluster.

2. Method according to claim 1, wherein said pre-established boundary within which to select the units of the type with consumption to be used for the purposes of determining the synthetic data relating to the trend of the energy withdrawn and / or the energy generated or fed into the network for the withdrawal units and / or the generation units of the so-called registry type is carried out by means of the following steps comprising: the definition of one or more characterization variables of the said withdrawal and / or generation units relating to corresponding description parameters of the said characterization variables; the definition of metrics that define a distance value as a function of the said characterization variables between each registry-type withdrawal and / or generation unit and each or a part of said withdrawal and / or generation units provided in the network and the definition of threshold values of the said distance values which threshold values define the boundaries of the said boundary and the selection of the withdrawal units and / or generation units that fall within the said boundary as units whose time trends are stored and related to the withdrawal of electricity and / or the generation and / or feed of electricity into the network are used to estimate the corresponding time trends for a pre-established registry-type unit; the application to said data of the trend of energy withdrawal and / or generation or feed of energy into the electricity network of an algorithm for estimating said synthetic datafor the corresponding registry-type withdrawal unit and / or generation or feed unit of electrical energy into the network and the unique association of said synthetic data to the corresponding registry unit for the generation of a corresponding virtual unit that simulates a unit of the type with consumption.

3. Method according to claims 1 or 2, wherein the clustering algorithm is a k-means algorithm, a combination of steps being provided which consist in: defining the optimal number of clusters through analysis of the minimum total variation of the clusters represented by the trend of a cost function as a function of the number of clusters, at least two different methods of determining the optimal number of clusters as a function of the analysis of the trend of a cost function, in particular the so- called Elbow Method and the so-called Silhouette Method, being used in parallel; the step of generating a training database of the algorithm and the training of the k- means algorithm itself, which step provides to define one or more identifying registry features of the individual energy withdrawal and / or generation units which characterize both the so-called registry units and also the so-called units with consumption; associating each or at least part of the said features with corresponding variables and defining metrics for calculating the distance in the multidimensional space defined by the said variables representing respectively each or at least part of the said features between the individual withdrawal and / or generation units of the registry type and of the type with consumption, the variables that describe the said features as well as the distance values defined by the said metrics constitute the records of the training database on the basis of which the k-means algorithm is trained; determining the effectiveness in relation to the discrimination of the individual withdrawal and / or generation units from each other, comprising both the units with consumption and also the registry units, of at least some or all of the different registry features, defined in the previous step, this identification being performed on the basis of an analysis technique called PCA (Principal Component Analysis).

4. Method according to one or more of the previous claims, wherein the result of the clustering process operated on the registry data of the individual withdrawal and / or generation units generates a distribution of the said withdrawal and / or generation units on the said optimal number of clusters, and the determination of the synthetic data of the time trend of the energy withdrawal and / or the generation or feed of energy into the network for a registry-type withdrawal and / or generation unit is performed using the realdata of the time trends of the withdrawal of energy and / or the generation of energy or the feed of energy into the network of a pre-established number of energy withdrawal and / or generation units of the type with consumption which are present in the same cluster comprising the said registry-type withdrawal and / or generation unit and which present the smallest distance from the said registry-type unit with reference to the variables representing one or more features with discriminatory effectiveness of the said energy withdrawal and / or generation units and with reference to the specific embodiment with reference to the geographical position.

5. Method according to claim 4, wherein the units with consumption that are closest to a registry unit present in the same cluster are determined by applying an algorithm called KNN.

6. Method for the generation of virtual energy withdrawal and / or generation or feed units that simulate corresponding units of the type with consumption, associating with each of the said registry-type energy withdrawal and / or generation units synthetic data consisting of the estimate of the time trend in a pre-established time window of the energy withdrawn and / or the energy generated or fed into the network for the corresponding registry-type units as a function of the real data of the time trend in a time window which are detected and stored in a plurality of energy withdrawal units and / or energy generation or feed into the network units of the type with consumption, the set, comprising a pre-established number of withdrawal units and / or generation units of the type with consumption and a pre-established number of registry-type withdrawal units and / or generation units, being subjected to a clustering process in which a certain number of the said units of the type with consumption and a certain number of the said registry-type units present in the said set is attributed to one of a predefined number of clusters and the said synthetic data for a registry withdrawal unit and / or energy generation or feed into the network unit provided in one of the clusters, calculated as a function of the real data of the time trend in a time window, being detected and stored in a certain number of withdrawal units and / or energy generation or feed into the network units of the type with consumption, which units with consumption have, within the same cluster, smaller distances from the corresponding said registry unit with reference to a mono or multidimensional space of representation of the variables relating to one or more representative features of the said units, in particular one or more registry features.

7. Method according to claim 6, wherein it is provided to use a supervised algorithm of the type called k-means to carry out the clustering process.

8. Method according to claim 7, which provides the preventive execution of a step of determining the optimal number of clusters for the execution of the clustering algorithm and wherein the aforementioned step of determining the optimal number of clusters provides to determine the said optimal number by means of two different methodologies for determining the optimal number of clusters for the execution of a clustering algorithm, said number being considered optimal when the optimal number determined with the two different methods is to be considered identical within the scope of pre-established tolerances.

9. Method according to one or more of the previous claims 6 to 8, wherein further steps are provided for defining and / or selecting a plurality of registry features which are common to the registry units and to those with consumption, and a step of identifying and selecting, from the said registry features, those registry features that most determine a differentiation between the different units provided in the network, both of the type with consumption and also of the registry type, the distribution of the said units, both those with consumption and also those of the registry type, on the different clusters, being determined on the basis of the or of one or more of the said registry features that most determine a differentiation between the said units and wherein the said distribution of the units, both those with consumption and also those of the registry type, on the different clusters constitutes the initial starting distribution of the iterative clustering process of the clustering algorithm, i.e. the k-means algorithm.

10. Method according to one or more of the previous claims 6 to 9, wherein there is provided a step for validating the synthetic data relating to the time trend of the withdrawal and / or generation or feed of energy for a registry-type unit, which step provides the comparison of the said synthetic data determined for pre-established time instants with the real measured data and the quantification of a distance metric of the said synthetic data from the said real data on the basis of which to determine the accuracy of the synthetic data in order to integrate for each registry unit, or for a part thereof, the registry data of the registry-type unit with the said synthetic data on the time trend of the energy withdrawn and / or that generated or fed into the network simulating that the said registry unit is a unit of the type with consumption.

11. Method according to one or more of the previous claims 6 to 10, wherein thesynthetic data relating to the estimated trend for a registry-type withdrawal unit and / or generation unit are calculated by determining the average of the time trends of the withdrawal of energy and / or the generation or feed of energy into the network stored in the corresponding units with consumption selected as closest and within a pre-established time window, thus generating a fictitious time trend of the energy withdrawn and / or the energy generated and / or fed into the network relating to the registry-type withdrawal unit and / or generation unit.

12. Method according to claim 11, wherein the step of determining the units with consumption closest to a registry unit present in the same cluster is performed using an algorithm called KNN (k-nearest neighbors).

13. Method according to claim 12, wherein the step of determining the units of the type with consumption which, within the same cluster, are closest to a registry-type unit provides in combination to determine a maximum threshold of the difference between a possible historical value of consumption recorded for the said registry unit referring to a pre-established past instant in time, and the consumption measured by the units of the type with consumption at the same instant in time, when said historical value of consumption is available, as a further condition to determine the proximity, the value of the said difference is used and a unit of the type with consumption is considered to be among the closest to a registry-type unit when the said difference is less than the said maximum threshold defined for the said difference.

14. System for implementing the aforementioned method according to one or more of the previous claims 1 to 13, which system comprises: an electricity distribution network, which network in turn comprises one or more electrical energy generation units which feed the said network and a plurality of separate and independent withdrawal points of the said electrical energy, in particular distributed over a pre-established geographical area and wherein at least a part of said withdrawal points are provided with a storage unit for the electricity withdrawal curve measured as a function of time from the corresponding withdrawal unit, i.e. the time trend of the electricity withdrawn from the network, and with a port for reading said withdrawal curve and wherein said withdrawal curves are stored in at least a part of the withdrawal units of the said plurality of withdrawal units and are supplied to, or can be retrieved from, a central controlunit, at least a part of said withdrawal units being instead without the functionalities for detecting and storing the time trend of the electricity withdrawal for a pre-established time window, which central control unit is further connected to measurement units for the curve as a function of the electrical energy generation time of at least part or each of the said electrical energy generation units and for the operating conditions of the said electrical energy units, i.e. of the time trend of the electricity generated, in said central control unit a program is stored and executable comprising the instructions that make said control unit capable of carrying out the processing steps of an algorithm for generating synthetic data of the time trend of the energy withdrawn and / or the energy generated or fed into the network for each or at least part of the withdrawal units and / or of the generation units that are not provided with the function of detecting and storing the time trend of the energy withdrawn and / or the energy generated or fed into the network and which steps of generating the said synthetic data comprise the generation of the said synthetic data by means of an algorithm for estimating said synthetic data as a function of the data relating to the time trend of the energy withdrawn and / or that produced or fed into the network which are stored in each or a part of a plurality of different withdrawal units and / or different energy generation units and which units are provided in a pre- established boundary with respect to the individual withdrawal units and / or the respective generation units which are without the said detection and storage functions, the steps of generating the said synthetic data being carried out before calculating the prediction for each or a part of the withdrawal units and / or the energy generation or feed into the network units, and the said synthetic data generated being associated respectively with each or with a pre- established part of corresponding energy withdrawal units and / or energy generation or feed into the network units, and said synthetic data being used for the calculation of the prediction on the trend of the energy withdrawn from the network relating to the said withdrawal and / or generation or feed into the network units which are without the functionalities for detecting and storing the time trend of the withdrawal of energy and / or the generation or feed of energy into the network, respectively.

15. System according to claim 14, wherein a program is further storable in a memoryof said central control unit and executable for predicting the time trend of the energy consumption of at least part, preferably of each of the withdrawal units within a pre- established future time period and / or for predicting the time trend of the energy generated by at least part or by each of the said electrical energy generation units connected to said network in said pre-established future time period, which prediction program comprises the instructions for the processor of said central control unit, which are related to the execution steps of an algorithm predictive of the said time trend of the consumption relating to at least part or to each of the said withdrawal units and / or predictive of said time trend of the quantity of energy generated by at least part or by each of the said generation units in the said pre-established future time period as a function of the said curves as a function of the time respectively of withdrawal and / or generation of the electrical energy by at least said part or by each of the said withdrawal units and / or said generation units stored therein and relating to a time period prior to said future time period, said predictions being carried out for each or at least part of the units that are provided with the functionalities for detecting and storing the time trend of the energy withdrawn and / or that generated or fed into the network and also for the units that are not provided with the functionalities for detecting and storing the time trend of the energy withdrawn and / or that generated or fed into the network, on the basis of said predictions there being activated units for compensating for the transient peaks of electricity withdrawal and / or electricity generation synchronized with the time instant in which said transient peaks are expected.

16. Fixed or portable or cloud-resident memory medium, on which a software is stored comprising the instructions for a processor or a combination of processors to carry out the steps of the method according to one or more of the previous claims 1 to 13.

Images