Sensor design support apparatus, sensor design support method and non-transitory computer readable medium
According to one embodiment, a sensor design support apparatus includes: an inter-data feature calculator and a model constructor. The inter-data feature calculator combines any two pieces of data based on first to k-th data acquired from each of a plurality of sensors in response to first to k-th conditions being satisfied to generate a plurality of data pairs for each of sensors, the sensors monitoring a target; and calculates a plurality of inter-data features based on the plurality of data pairs. The model constructor calculates a plurality of coefficients corresponding to the plurality of inter-data features in a classification model based on state data indicating states of the monitoring target at times which the first to k-th conditions are satisfied, wherein the classification model associates the plurality of coefficients and the plurality of inter-data features with a value identifying a state of the monitoring target.
1. An information processing apparatus comprising:
communication circuitry configured to receive first to k-th data detected by each of a plurality of sensors at first to k-th timings, the sensors monitoring a target, and receive actual values of a state of the target at the first to k-th timings;
a hardware storage configured to store the first to k-th data to be associated with the first to k-th timings for each of the sensors, and the state values; and
processing circuitry configured to:
read the first to k-th data associated with the first to k-th timings for each of the sensors from the hardware storage,
combine any two pieces of the first to k-th data for each of the sensors to generate a plurality of data pairs for each of the sensors,
perform data processing on each pair of the plurality of data pairs to obtain a plurality of inter-data features corresponding to the plurality of data pairs for each of the sensors,
generate a classification model providing a predicted value of a state of the target by performing regression of a state of the target on the plurality of inter-data features of the sensors weighted by a plurality of coefficients corresponding to the sensors,
generate a first function including a loss function of differences between predicted values of the classification model and the actual values of the state of the target and including a penalty term including the plurality of coefficients,
solve the first function to find values of the plurality of coefficients, and
select at least one sensor from among the plurality of sensors on the basis of the values of the plurality of coefficients,
wherein
the communication circuitry is configured to receive data detected by the selected sensor and not receive data from a non-selected sensor, and
the processing circuitry is configured to calculate a predicted value of a state of the target based on the data received from the selected sensor and the classification model and diagnose the target based on the predicted value.
2. The information processing apparatus according to claim 1 , wherein
the plurality of coefficients further correspond to the first to k-th timings,
the processing circuitry selects at least one timing from among the first to k-th timings on the basis of the values of the plurality of coefficients, and
the communication circuitry is configured to receive data at the selected timing from the selected sensor and not receive data at non-selected timing from the selected sensor.
3. The information processing apparatus according to claim 1 , wherein
the processing circuitry selects a nonzero coefficient from among the plurality of coefficients,
selects a sensor corresponding to the nonzero coefficient, and
does not select a sensor corresponding to a coefficient of zero.
4. The information processing apparatus according to claim 1 , wherein the processing circuitry minimizes the first function to find the plurality of coefficients.
5. The information processing apparatus according to claim 1 , wherein
the penalty term includes constants corresponding to the plurality of sensors,
the sensors have priority values, and
sensors having a higher priority value have a smaller constraint value.
6. The information processing apparatus according to claim 2 , wherein
the processing circuitry selects a coefficient of nonzero from among the plurality of coefficients, selects a timing corresponding to the coefficient of nonzero, and does not select a timing corresponding to a coefficient of zero.
7. The information processing apparatus according to claim 1 , wherein at least one another target different from the target is to be installed with the selected sensor for monitoring the at least another target, and not to be installed with the non-selected sensor.
8. An information processing method comprising:
receiving first to k-th data detected by each of a plurality of sensors at first to k-th timings, the sensors monitoring a target, and receiving actual values of a state of the target at the first to k-th timings;
storing, in a hardware storage, the first to k-th data to be associated with the first to k-th timings for each of the sensors, and the state values;
reading the first to k-th data associated with the first to k-th timings for each of the sensors from the hardware storage;
combining any two pieces of the first to k-th data for each of the sensors to generate a plurality of data pairs for each of the sensors;
performing data processing on each pair of the plurality of data pairs to obtain a plurality of inter-data features corresponding to the plurality of data pairs for each of the sensors;
generating a classification model providing a predicted value of a state of the target by performing regression of a state of the target on the plurality of inter-data features of the sensors weighted by a plurality of coefficients corresponding to the sensors;
generating a first function including a loss function of differences between predicted values of the classification model and the actual values of the state of the target and including a penalty term including the plurality of coefficients;
solving the first function to find values of the plurality of coefficients;
selecting at least one sensor from among the plurality of sensors on the basis of the values of the plurality of coefficients; and
receiving data detected by the selected sensor and not receiving data from a non-selected sensor; and
calculating a predicted value of a state of the target based on the data received and the classification model and diagnosing the target based on the predicted value.
9. A non-transitory computer readable medium having a program stored therein which causes, when executed by a computer, the computer to execute processing comprising:
receiving first to k-th data detected by each of a plurality of sensors at first to k-th timings, the sensors monitoring a target, and receiving actual values of a state of the target at the first to k-th timings;
storing, in a hardware storage, the first to k-th data to be associated with the first to k-th timings for each of the sensors, and the state values;
reading the first to k-th data associated with the first to k-th timings for each of the sensors from the hardware storage;
combining any two pieces of the first to k-th data for each of the sensors to generate a plurality of data pairs for each of the sensors;
performing data processing on each pair of the plurality of data pairs to obtain a plurality of inter-data features corresponding to the plurality of data pairs for each of the sensors;
generating a classification model providing a predicted value of a state of the target by performing regression of a state of the target on the plurality of inter-data features of the sensors weighted by a plurality of coefficients corresponding to the sensors;
generating a first function including a loss function of differences between predicted values of the classification model and the actual values of the state of the target and including a penalty term including the plurality of coefficients;
solving the first function to find values of the plurality of coefficients;
selecting at least one sensor from among the plurality of sensors on the basis of the values of the plurality of coefficients;
receiving data detected by the selected sensor and not receiving data from a non-selected sensor; and
calculating a predicted value of a state of the target based on the data received and the classification and diagnosing the target based on the predicted value.