Resource cost optimization system, method, and program
View Patent ↗Apparatus and method use a Markov decision process (MDP) to reduce the cost of variations in electric power usage. The user notifies a power company of a predicted value for a period. The period is divided into subsections. For each subsection, on the basis of a MDP including a state that depends on an electric power usage amount error, charge amount, and set target, the amount of charging and discharging of a storage battery as an action at any given time is optimally decided depending on the electric power usage amount error, charge amount, time, and set target at that time. A predetermined time in a subsection is a target setting time, at which a future target is further set as the action. The action includes deciding the charging and discharging amount in that subsection and deciding a future target in a subsection whose target should be set.
1. A computer implemented method for generating a policy for optimizing a cost of a resource under a predetermined cost structure, the method comprising:
storing, in a computer-readable medium, an error distribution that indicates a deviation of an amount of usage from a predicted value referred to as usage amount error, a characteristic of a storage battery configured to store or release the resource, wherein the characteristic includes an amount of the resource in the storage battery, and the cost structure;
calculating, using a computer processor processing a Markov decision process, an expected cost and a parameter that includes a transition probability on the basis of the error distribution, the characteristic of the storage battery, and the cost structure, the Markov decision process including a state defined by at least the usage amount error, the amount of the resource in the storage battery, a specification of a subsection within a section of usage interval, and a set target for a next section; and
deciding, using the computer processor, and implementing an optimal policy for the next section that includes an action of storing or releasing the resource in the storage battery for the state of the Markov decision process using the expected cost in the Markov decision process and the parameter including the transition probability, wherein the resource comprises electric power.
2. The method according to claim 1 , wherein the cost structure comprises a piecewise linear function.
3. The method according to claim 1 , wherein said deciding an optimal policy includes solving using linear programming.
4. The method according to claim 1 , wherein said deciding an optimal policy includes solving using value iteration.
5. The method according to claim 1 , wherein said deciding an optimal policy includes solving using policy iteration.
6. A computer program product for generating a policy for optimizing a cost of a resource under a predetermined cost structure, the program product comprising a computer-readable storage medium having program code embodied therewith, the program code being executable by a processor to perform a method comprising:
a step of storing an error distribution that indicates a deviation of an amount of usage from a predicted value referred to as usage amount error, a characteristic of a storage battery configured to store or release the resource the resource, wherein the characteristic includes an amount of the resource in the storage battery, and the cost structure in a computer-readable form;
a step of calculating, using a Markov decision process, an expected cost and a parameter that includes a transition probability on the basis of the error distribution, the characteristic of the storage battery, and the cost structure, the Markov decision process including a state defined by at least the usage amount error, the amount of resource in the storage battery, a specification of a subsection within a section of usage interval, and a set target for a next section; and
a step of deciding an optimal policy for implementation, the optimal policy includes an action of storing or releasing the resource in the storage battery for the state of the Markov decision process using the expected cost in the Markov decision process and the parameter including the transition probability, wherein the resource comprises electric power.
7. The program product according to claim 6 , wherein the cost structure comprises a piecewise linear function.
8. The program product according to claim 6 , wherein the step of deciding the optimal policy is solved using linear programming.
9. The program product according to claim 6 , wherein the step of deciding the optimal policy is solved using value iteration.
10. The program product according to claim 6 , wherein the step of deciding the optimal policy is solved using policy iteration.