Method and apparatus for forecasting future values of a time series

Software in machine readable form on a storage medium for performing a method of predicting at least one future data value of a time series of data using a neural network so as to manage resources in a communications network, the method comprising the steps of:

• (i) inputting a plurality of data values of the time series into the neural network;
• (ii) inputting, with said data values, time information relating to those values;
• (iii) obtaining outputs from the neural network said outputs comprising predicted future values of the time series based on said data values and on said time information;
• and
• (iv) allocating network resources corresponding to said predicted future values;
• wherein said step (i) of inputting a plurality of values of the time series into the neural network comprises the step of determining a number of values of the time series that are required to be input to the neural network, said step of determining a number of values comprising the steps of:
• (v) forming a set of first vectors wherein each first vector is the same size and each first vector comprises a number of sequential values of the time series;
• (vi) forming a set of second vectors, wherein each second vector is the same size, and each second vector comprises a number of sequential values of the time series, and wherein the first and second vectors are different sizes;
• (vii) for each first vector selecting another of the first vectors as a first neighbour where a first measure of similarity between each first vector and its first neighbour is less than a threshold value;
• (viii) for each second vector selecting another of the second vectors as a second neighbour where a second measure of similarity between each first vector and its second neighbour is less than a threshold value, and wherein each second neighbour corresponds to a first neighbour;
• (ix) determining the number of false neighbours by comparing each first neighbour with its corresponding second neighbour;
• and
• (x) determining the number of values that are required to be input to the neural network according to a first vector size for which a threshold number of false neighbours are obtained.

A method for predicting at least one future data value of a time series of data relating to a communications network, said communications network incorporating a communications network management system arranged to allocate resources, said method comprising the steps of:

• (i) automatically inputting one or more data values of the time series into the neural network from the communications network management system, said data values having associated therewith time information relating to those values;
• and
• (ii) obtaining outputs from the neural network said outputs comprising predicted future value(s) of the time series based on said data values and on said time information, and automatically providing said outputs to the communications network management system so as to enable control of said resource allocation;
• wherein said step (i) of inputting a plurality of values of the time series into the neural network comprises the step of determining a number of values of the time series that are required to be input to the neural network, said step of determining a number of values comprising the steps of:
• (iii) forming a set of first vectors wherein each first vector is the same size and each first vector comprises a number of sequential values of the time series;
• (iv) forming a set of second vectors, wherein each second vector is the same size, and each second vector comprises a number of sequential values of the time series, and wherein the first and second vectors are different sizes;
• (v) for each first vector selecting another of the first vectors as a first neighbour where a first measure of similarity between each first vector and its first neighbour is less than a threshold value;
• (vi) for each second vector selecting another of the second vectors as a second neighbour where a second measure of similarity between each first vector and its second neighbour is less than a threshold value, and wherein each second neighbour corresponds to a first neighbour;
• (vii) determining the number of false neighbours by comparing each first neighbour with its corresponding second neighbour;
• and
• (viii) determining the number of values that are required to be input to the neural network according to a first vector size for which a threshold number of false neighbours are obtained.

A communications network comprising:

• a communications network management system arranged to allocate resources;
• a computer system for predicting at least one future data value of a time series of data relating to the communications network;
• said computer system comprising:
• a neural network;
• an input to the neural network arranged to automatically accept values of the time series from the communications network management system, said data values having associated therewith time information relating to those values;
• and
• an output from the neural network arranged to provide predicted future value(s) of the time series based on said data values and on said time information to the communications network management system;
• wherein said computer system is arranged to compute the values of the time series input to the neural network by determining a number of values of the time series that are required to be input to the neural network, said step of determining a number of values comprising the steps of:
• (i) forming a set of first vectors wherein each first vector is the same size and each first vector comprises a number of sequential values of the time series;
• (ii) forming a set of second vectors, wherein each second vector is the same size, and each second vector comprises a number of sequential values of the time series, and wherein the first and second vectors are different sizes;
• (iii) for each first vector selecting another of the first vectors as a first neighbour where a first measure of similarity between each first vector and its first neighbour is less than a threshold value;
• (iv) for each second vector selecting another of the second vectors as a second neighbour where a second measure of similarity between each first vector and its second neighbour is less than a threshold value, and wherein each second neighbour corresponds to a first neighbour;
• (vi) determining the number of false neighbours by comparing each first neighbour with its corresponding second neighbour;
• and
• (vii) determining the number of values that are required to be input to the neural network according to a first vector size for which a threshold number of false neighbours are obtained.