Cognitive framework for improving responsivity in demand response programs
Methods, computer program products, and systems are presented. The methods include, for instance: extracting from historical data and demand response agreements, attributes relevant to responsivities of demand response programs; and training a demand-response (DR) user pooling model as a machine learning model with training datasets including the attributes from the extracting.
1 . A computer implemented method comprising:
obtaining historical data of demand response programs and demand response agreements via a demand response interconnection application programming interface (API);
extracting from the historical data and the demand response agreements, structured attributes comprising numerical offtake obligations, response-time windows, and pool size values relevant to responsivities of the demand response programs;
training a demand-response (DR) user pooling model as a machine learning model with training datasets including the attributes from the extracting, wherein the DR user pooling model identifies two or more users amongst a plurality of users as a DR user pool;
predicting a new set of values corresponding to the attributes and the responsivities of the demand response programs as being responded to by the DR user pool, the predicting including generating probability vectors and comparing the vectors to threshold ranges derived from the historical data; and
adjusting a configuration of the DR user pool according to the new set of values from the predicting, including reallocating offtake load among users of the DR user pool using a queue algorithm constrained by the defined response-time windows.
2 . The computer implemented method of claim 1 , wherein the obtaining includes obtaining the historical data of demand response programs between one or more provider of a subject energy and a plurality of users and demand response agreements of respective ones of the users.
3 . The computer implemented method of claim 1 , wherein the training includes training the demand-response (DR) user pooling model as a machine learning model with training datasets including the attributes from the extracting and values corresponding to respective ones of the attributes.
4 . The computer implemented method of claim 1 , wherein one or more demands of the demand response programs are responded together by the two or more users in the DR user pool.
5 . The computer implemented method of claim 1 , wherein the adjusting includes adjusting the configuration of the DR user pool according to the new set of values from the predicting, upon ascertaining that improved responsivities of the demand response programs had been predicted.
6 . The computer implemented method of claim 1 , wherein the adjusting includes adjusting the configuration of the DR user pool according to the new set of values from the predicting, upon ascertaining that improved responsivities of the demand response programs had been predicted with one or more instances from the new set of values.
7 . The computer implemented method of claim 1 , further comprising configuring the DR user pool as a queue system in which the two or more users in the DR user pool share respective capacities for offtake and respond to the one or more demands together by shifting a demand of the one or more demands to any user in the DR user pool having a capacity for offtake available within an offtake window corresponding to the demand within which the demand is to be responded to that is being shifted within the DR user pool.
8 . The computer implemented method of claim 1 , further comprising testing, prior to the predicting, the DR user pooling model with the attributes relevant to the responsivities of the demand response programs and another set of values corresponding to the attributes, as obtained from a test dataset.
9 . The computer implemented method of claim 1 , further comprising validating the new set of values corresponding to the attributes and the responsivities of the demand response programs as being responded to by the DR user pool, based on a determination that each of the new set of values are within respective threshold ranges for fitting with the historical data of the demand response programs.
10 . The computer implemented method of claim 1 , wherein the historical data of the demand response programs and the demand response agreements are collected from a DR interconnection application program interface (API) coupling one or more provider and the plurality of users subject to the demand response agreements.
11 . The computer implemented method of claim 1 , wherein a subject energy for the demand response programs is electricity, and wherein the attributes of the training datasets comprise: an offtake obligation for each demand, peak hours, a number of users in the DR user pool, and a size of an offtake window within which the demand is to be responded.
12 . The computer implemented method of claim 1 , further comprising iterating the training of the DR user pooling model base based on a determination that new training datasets have been collected from a DR interconnection API coupling one or more provider and the plurality of users subject to the demand response agreements, upon ascertaining that the new set of values corresponding to the attributes and the responsivities of the demand response programs as being responded to by the DR user pool fits the historical data of the demand response programs as falling within respective threshold ranges for each of the attributes.
13 . A computer program product comprising:
a computer readable storage medium readable by one or more processors and storing instructions for execution by the one or more processors for performing a method comprising:
obtaining historical data of demand response programs and demand response agreements via a demand response interconnection application programming interface (API);
extracting from the historical data and the demand response agreements, structured attributes comprising numerical offtake obligations, response-time windows, and pool size values relevant to responsivities of the demand response programs;
training a demand-response (DR) user pooling model as a machine learning model with training datasets including the attributes from the extracting, wherein the DR user pooling model identifies two or more users amongst a plurality of users as a DR user pool;
predicting a new set of values corresponding to the attributes and the responsivities of the demand response programs as being responded to by the DR user pool, the predicting including generating probability vectors and comparing the vectors to threshold ranges derived from the historical data; and
adjusting a configuration of the DR user pool according to the new set of values from the predicting, including reallocating offtake load among users of the DR user pool using a queue algorithm constrained by the defined response-time windows.
14 . The computer implemented method of claim 1 , wherein the extracted attributes include numerical offtake obligations and response-time window values encoded in the demand response agreements.
15 . The computer implemented method of claim 1 , wherein the demand response interconnection API provides the demand response agreements in a structured, machine-readable format usable as training input for the demand-response user pooling model.
16 . The computer implemented method of claim 1 , wherein the predicting comprises generating probability vectors representing predicted responsiveness of respective candidate DR user pools and comparing the vectors to threshold ranges derived from historical datasets.
17 . The computer implemented method of claim 1 , further comprising writing prediction validation results to a log data structure accessible for subsequent verification.
18 . The computer implemented method of claim 1 , wherein the adjusting comprises executing the queue algorithm that reallocates offtake load among the users of the DR user pool during the defined response-time windows based on current capacity values.
19 . The computer implemented method of claim 1 , wherein the obtaining includes obtaining the historical data of demand response programs between one or more provider of a subject energy and a plurality of users and demand response agreements.
20 . A system comprising:
a memory;
one or more processors in communication with the memory; and
program instructions executable by the one or more processors via the memory to perform a method comprising:
obtaining historical data of demand response programs and demand response agreements via a demand response interconnection application programming interface (API);
extracting from the historical data and the demand response agreements, structured attributes comprising numerical offtake obligations, response-time windows, and pool size values relevant to responsivities of the demand response programs;
training a demand-response (DR) user pooling model as a machine learning model with training datasets including the attributes from the extracting, wherein the DR user pooling model identifies two or more users amongst a plurality of users as a DR user pool;
predicting a new set of values corresponding to the attributes and the responsivities of the demand response programs as being responded to by the DR user pool, the predicting including generating probability vectors and comparing the vectors to threshold ranges derived from the historical data; and
adjusting a configuration of the DR user pool according to the new set of values from the predicting, including reallocating offtake load among users of the DR user pool using a queue algorithm constrained by the defined response-time windows.