A machine learning-based energy management system, the system comprising:
- a power load module comprising a propulsion load and a service electrical load and configured to consume power supplied from outside;
- a power source module comprising a fuel cell, a battery, and a generator and configured to supply power to the power load module;
- a load management control module being connected to the power source module to perform status monitoring and operation control thereof, with the load management control module being configured to induce an overall power generation efficiency increase and energy saving while controlling charging and discharging of the battery according to a load size value of the generator;
- and
- an energy management control module that is connected to the power load module to control power supply to the propulsion load and the service electrical load, and controls an operation of the power source module in conjunction with the load management control module, wherein the energy management control module allows a power supply control signal of the power source module corresponding to a required load of the power load module to be calculated in real time by a control signal generation energy management system (EMS) algorithm to which machine learning is applied, wherein the control signal generation EMS algorithm performs machine learning of load data of the power load module and power data of the power source module, and calculates the power supply control signal to increase output efficiency and energy saving efficiency of the power source module according to a load characteristic, the required load, and a load change of the power load module,
- wherein the energy management control module comprises:
- a data collection unit being connected to an electronic chart system (ECS) module, the power load module, and the power source module to cumulatively collect initial data for machine learning comprising ship operation data, required load data, actual operation load data, and status information of the power load module, and supply power data and status information of the power source module;
- a valid data calculation unit being connected to the data collection unit to receive the initial data for machine learning, wherein the valid data calculation unit calculates valid data for machine learning by extracting only the initial data for machine learning that fall within a valid data range region set through a filtering algorithm,
- wherein the filtering algorithm of the valid data calculation unit sets a numerical value region between a valid upper data limit (UDL) calculated from Equation 1 and a valid lower data limit (LDL) calculated from Equation 2 as the valid data range region,
- wherein
- UDL
- =
- m
- 0
- +
- W
-
- d
-
- s
- 2
- -
- s
- [
- 1
- -
- (
- 1
- -
- s
- )
- 2
-
- k
- ]
- LDL
- =
- m
- 0
- -
- W
-
- d
-
- s
- 2
- -
- s
- [
- 1
- -
- (
- 1
- -
- s
- )
- 2
-
- k
- ]
-
- and
- m0 is an input data average value, d is a standard deviation, W is a control width, and s is a smoothing parameter.