Building equipment control system with modular models
A method includes obtaining a fault prediction model for building equipment, predicting, with the fault prediction model, both (i) whether a fault will occur during a first prediction bin and (ii) whether a fault will occur during a second prediction bin, performing a first mitigating action for the building equipment if the fault is predicted to occur during the first prediction bin, and performing a second mitigating action for the building equipment if the fault is predicted to occur during the second prediction bin.
1. A method for mitigating faults in building equipment, comprising:
training a fault prediction model on training data comprising (1) a plurality of rolling sequences of timeseries data corresponding to overlapping segments of time and (2) for each of the plurality of rolling sequences of timeseries data, a first fault label for a first time bin after the corresponding segment of time and second fault label for a second time bin after the first time bin, wherein the method comprises preprocessing the training data by calculating synthetic features comprising the first fault label and the second fault label based on raw timeseries data and fault rules;
predicting, with the fault prediction model, whether a fault will occur for a first prediction bin and whether the fault will occur for a second prediction bin by generating, by the fault prediction model, labels for both the first prediction bin and the second prediction bin;
altering operation of the building equipment in a first manner if the fault is predicted to occur for the first prediction bin; and
altering operation of the building equipment in a second manner if the fault is predicted to occur for the second prediction bin.
2. The method of claim 1 , wherein preprocessing the training data further comprises resampling and interpolating to achieve a constant sample rate for a plurality of variables represented in the training data.
3. The method of claim 1 , wherein preprocessing the training data further comprises splitting the training data into the plurality of rolling sequences of the time series data corresponding to the overlapping segments of time, the overlapping segments of time having a same duration.
4. The method of claim 1 , wherein the fault prediction model comprises a classifier and an autoencoder.
5. The method of claim 4 , further comprising providing composite training of the classifier and the autoencoder.
6. The method of claim 1 , further comprising modularly constructing the fault prediction model by combining (1) an autoencoder selected from a first group comprising a principle component analysis model, a long short-term memory network, a feedforward network, and a convolutional network with (2) a classifier selected from a second group comprising a Gaussian classifier, a feedforward model, and an XGBoost model; and
further comprising constructing a second fault prediction model using different selections from the first group and the second group as compared to the fault prediction model.
7. The method of claim 6 , wherein the building equipment is a chiller.
8. One or more non-transitory computer-readable media storing program instructions, that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
training a fault prediction model on training data comprising (1) a plurality of rolling sequences of timeseries data corresponding to overlapping segments of time and (2) for each of the plurality of rolling sequences of timeseries data, a first fault label for a first time bin after the corresponding segment of time and second fault label for a second time bin after the first time bin, wherein the operations comprise preprocessing the training data by calculating synthetic features comprising the first fault label and the second fault label based on raw timeseries data and fault rules;
predicting, with the fault prediction model for building equipment, whether a first fault will occur for a first prediction bin and whether a second fault will occur for a second prediction bin by generating, by the fault prediction model, labels for both the first prediction bin and the second prediction bin;
altering operation of the building equipment in a first manner if the first fault is predicted to occur for the first prediction bin; and
altering operation of the building equipment in a second manner if the second fault is predicted to occur for the second prediction bin.
9. The one or more non-transitory computer-readable media of claim 8 , wherein the operations comprise processing the raw timeseries data by resampling and interpolating to achieve a constant sample rate for a plurality of variables represented in the training data.
10. The one or more non-transitory computer-readable media of claim 8 , wherein the operations comprise processing the raw timeseries data by splitting the raw timeseries data into the plurality of rolling sequences corresponding to the overlapping segments of time, the overlapping segments of time have a same duration.
11. The one or more non-transitory computer-readable media of claim 8 , wherein:
the fault prediction model comprises a classifier and an autoencoder; and
training the fault prediction model comprises providing composite training of the classifier and the autoencoder.
12. The one or more non-transitory computer-readable media of claim 8 , the operations further comprising modularly constructing the fault prediction model by combining (1) an autoencoder selected from a first group comprising a principle component analysis model, a long short-term memory network, a feedforward network, and a convolutional network with (2) a classifier selected from a second group comprising a Gaussian classifier, a feedforward model, and an XGBoost model.
13. The one or more non-transitory computer-readable media of claim 12 , the operations further comprising constructing a second fault prediction model using different selections from the first group and the second group as compared to the fault prediction model.
14. A unit of building equipment, comprising:
a heating, ventilation, or cooling component;
one or more processors; and
one or more non-transitory computer-readable media storing program instructions, that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
training a fault prediction model on training data comprising (1) a plurality of rolling sequences of timeseries data corresponding to overlapping segments of time and (2) for each of the plurality of rolling sequences of timeseries data, a first fault label for a first time bin after the corresponding segment of time and second fault label for a second time bin after the first time bin, wherein the operations comprise preprocessing the training data by calculating synthetic features comprising the first fault label and the second fault label based on raw timeseries data and fault rules;
predicting, with the fault prediction model at a current time, both (i) whether a fault of the unit of building equipment will occur for a first prediction bin extending from the current time to a first future time and (ii) whether the fault of the unit of building equipment will occur for a second prediction bin extending from the first future time to a second future time, by generating, by the fault prediction model, labels for both the first prediction bin and the second prediction bin as solutions to a multi-label sequence classification problem;
altering operation of the unit of building equipment in a first manner if the fault is predicted to occur for the first prediction bin; and
altering operation of the unit of building equipment in a second manner if the fault is predicted to occur for the second prediction bin.
15. The unit of building equipment of claim 14 , wherein altering the operation of the unit of building equipment in the first manner comprises operating the building equipment using a changed first operating parameter of the unit of building equipment and altering the operation of the unit of building equipment in the second manner comprises operating the building equipment using a changed second operating parameter of the unit of building equipment.
16. The unit of building equipment of claim 14 , the operations further comprising modularly constructing the fault prediction model by combining (1) an autoencoder selected from a first group comprising a principle component analysis model, a long short-term memory network, a feedforward network, and a convolutional network with (2) a classifier selected from a second group comprising a Gaussian classifier, a feedforward model, and an XGBoost model.