An Adaptive Large Neighbourhood Search for asset protection during escaped wildfires

The asset protection problem is encountered where an uncontrollable fire is sweeping across a landscape comprising important infrastructure assets. Protective activities by teams of firefighters can reduce the risk of losing a particular asset. These activities must be performed during a time-window for each asset determined by the progression of the fire. The nature of some assets is such that they require the simultaneous presence of more than one fire vehicle and its capabilities must meet the requirements of each asset visited. The objective is then to maximise the value of the assets protected subject to constraints on the number and type of fire trucks available. The solution times to this problem using commercial solvers preclude their use for operational purposes. In this work we develop an Adaptive Large Neighbourhood Search algorithm (ALNS) based on problem-specific attributes. Several removal and insertion heuristics, including some new algorithms, are applied. A new benchmark set is generated by considering the problem attributes. In tests with small instances the ALNS is shown to achieve optimal, or near optimal, results in a fraction of the time required by CPLEX. In a second set of experiments comprising larger instances the ALNS was able to produce solutions in times suitable for operational purposes. These solutions mean that significantly more assets can be protected than would be the case otherwise.