Method and apparatus for dynamic network configuration and optimisation using artificial life

An apparatus for recommending configuration changes in a communications network, the apparatus comprising a processing circuitry and a memory, the memory containing instructions executable by the processing circuitry such that the apparatus is operative to:

• maintain a plurality of machine-learning processes, wherein an individual machine-learning process operates based on a data model and decision-making rules, and the plurality of machine-learning processes operate based on a plurality of different data models and a plurality of different decision-making rules;
• obtain values of Key Performance Indicators, KPIs, from network elements of the communications network;
• obtain a goal defining at least one KPI value characterising wanted operation of the communications network;
• produce by the plurality of machine-learning processes, based on the received values of KPIs and using the data models and decision-making rules, a plurality of individual recommendations, wherein an individual recommendation gives value of at least one parameter, which needs to be configured to achieve the goal;
• produce an output recommendation based on the produced individual recommendations, and
• wherein each one of the individual machine-learning processes has a parameter indicative of viability of said individual machine-learning process and wherein the apparatus is operative to periodically carry out the operations leading to producing an output recommendation and, in each cycle, the apparatus is operative to compare the obtained KPI values with the goal obtained in a preceding cycle, and if the output recommendation has been implemented the apparatus is operative to:
• for each individual machine-learning process which produced individual recommendation falling within a defined range around the output recommendation:
• increase a value of the parameter indicative of viability of an individual machine-learning process if the goal obtained in the preceding cycle has been achieved;
• reduce the value of the parameter indicative of viability of an individual machine-learning processes by a value corresponding to energy cost required for carrying out the operation of producing an individual recommendation if the goal obtained in the preceding cycle has not been achieved.