Maintenance range optimization apparatus, maintenance range optimization method, and computer-readable recording medium
A maintenance range optimization apparatus 10 optimizes a range of maintenance on an object that requires maintenance at a plurality of places. The maintenance range optimization apparatus 10 includes a learning processing unit 20 that executes machine learning, using, as learning data, information from when maintenance was previously executed, including a pre-maintenance state, a maintenance cost and a movement cost of a place subjected to maintenance, and constructs a model indicating a relationship between the range of maintenance and an overall cost incurred in maintenance, and a maintenance range setting unit 30 that sets the range of maintenance using the model.
1. A maintenance range optimization apparatus for optimizing a range of maintenance on an object that requires the maintenance at a plurality of places, the apparatus comprising:
a processor; and
a memory storing program code executable by the processor to:
execute machine learning, using, as learning data, information from when the maintenance was previously executed, including a pre-maintenance state, a maintenance cost and a movement cost of each place subjected to the maintenance, to construct a machine learning model indicating a relationship between the range of the maintenance and an overall end cost including a risk resulting from deteriorated objects incurred in the maintenance; and
set the range of the maintenance as a vector indicating whether each place has been maintained or not, using the model,
wherein when the machine learning model is constructed, Q-learning as the machine learning is executed that employs a Q function as the machine learning model is constructed, where data specifying the pre-maintenance state of each place is input into the Q function, where the Q function into which the data is input is applied to the vector, and where the Q function is expressed as a decomposition of a current maintenance cost and a maintenance benefit at each place,
and the range of the maintenance is set as the vector at which the value of the Q function is maximized.
2. A maintenance range optimization method for optimizing a range of maintenance on an object that requires the maintenance at a plurality of places, the method comprising:
executing machine learning, using, as learning data, information from when the maintenance was previously executed, including a pre-maintenance state, a maintenance cost and a movement cost of each place subjected to the maintenance, to constructing a machine learning model indicating a relationship between the range of the maintenance and an overall end cost including a risk resulting from deteriorated objects incurred in the maintenance; and
setting the range of the maintenance as a vector indicating whether each place has been maintained or not, using the machine learning model,
wherein when the machine learning model is constructed, Q-learning as the machine learning is executed that employs a Q function as the machine learning model is constructed, where data specifying the pre-maintenance state of each place is input into the Q function, where the Q function into which the data is input is applied to the vector, and where the Q function is expressed as a decomposition of a current maintenance cost and a maintenance benefit at each place,
and the range of the maintenance is set as the vector at which the value of the Q function is maximized.
3. A non-transitory computer-readable recording medium that includes a program recorded thereon for optimizing, by computer, a range of maintenance on an object that requires the maintenance at a plurality of places, the program including instructions that cause the computer to carry out:
executing machine learning, using, as learning data, information from when the maintenance was previously executed, including a pre-maintenance state, a maintenance cost and a movement cost of each place subjected to the maintenance, to constructing a machine learning model indicating a relationship between the range of the maintenance and an overall end cost including a risk resulting from deteriorated objects incurred in the maintenance; and
setting the range of the maintenance as a vector indicating whether each place has been maintained or not, using the machine learning model,
wherein when the machine learning model is constructed, Q-learning as the machine learning is executed that employs a Q function as the machine learning model is constructed, where data specifying the pre-maintenance state of each place is input into the Q function, where the Q function into which the data is input is applied to the vector, and where the Q function is expressed as a decomposition of a current maintenance cost and a maintenance benefit at each place,
and the range of the maintenance is set as the vector at which the value of the Q function is maximized.