# US 11,909,207 B2 · Hierarchical method for prediction of loads with high variance > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,909,207 B2 - Document ID: US11909207B2 - Status: Granted - Kind: B2 - Application number: 17/197,625 - Filed: 2021-03-10 - Published: 2022-09-15 - Granted: 2024-02-20 - Earliest claimed: 2021-03-10 - Assignees: ABB Schweiz AG - Applicants: ABB Schweiz AG - Inventors: Joseph Carr; Alexander Brissette - Web page: https://patentgrep.com/patent/US11909207B2 ## CPC - H02J 3/003: {Load forecast, e.g. methods or systems for forecasting future load demand} - F24F 11/63: Electronic processing - G05B 13/027: {using neural networks only} - G05B 13/048: {using a predictor} - G06N 3/044: Recurrent networks, e.g. Hopfield networks - G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] - G06N 3/09: Supervised learning - H02J 3/14: by switching loads on to, or off from, the networks, e.g. progressively balanced loading - H02J 2103/30: Simulating, planning, modelling, reliability check or computer assisted design [CAD] of electric power networks - H02J 2105/52: for limitation of the power consumption in the networks or in one section of the networks, e.g. load shedding or peak shaving - H02J: ELECTRIC POWER NETWORKS; CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY - F24F: AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING (removing dirt or fumes from areas where they are produced B08B15/00; vertical ducts for carrying away waste gases from buildings E04F17/02; tops for chimneys or ventilating shafts, terminals for flues F23L17/02) - G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A method and system are provided for improving predictions of electrical power usage. In the method and system, load and/or environmental data is classified into data sets that correspond to different modes of operation of an electrical load. Different predictive models are also provided for each set of classified data. The predictive models may provide more efficient and/or more accurate predictions of power usage since each model is limited to a particular mode of operation. ## Claims ### Claim 1 A method of compensating electrical power distribution, comprising: collecting input data comprising load and/or environmental data; classifying the input data into a first set of data and a second set of data, the first set of data defining one level of power consumption and the second set of data defining another level of power consumption; predicting a first power consumption for a first period corresponding to the first set of data; predicting a second power consumption for a second period corresponding to the second set of data; determining an expected power consumption based on the first power consumption or the second power consumption; and shedding an electrical load or increasing an electrical supply based on the expected power consumption; wherein the first set of data defines a valley of power consumption based on a first operating mode and the second set of data defines a peak of power consumption based on a second operating mode. ### Claim 2 (depends on 1) The method according to claim 1, wherein the expected power consumption is determined at different times based on the first power consumption and the second power consumption. ### Claim 3 (depends on 1) The method according to claim 1, wherein the first set of data is not used to predict the second power consumption and the second set of data is not used to predict the first power consumption. ### Claim 4 (depends on 1) The method according to claim 1, wherein the first set of data is defined by an electrical device being at a lower power level and the second set of data is defined by the electrical device being at a higher power level. ### Claim 5 (depends on 4) The method according to claim 4, wherein electrical device is off at the lower power level and is on at the higher power level. ### Claim 6 (depends on 4) The method according to claim 4, wherein the electrical device is an HVAC system. ### Claim 7 (depends on 1) The method according to claim 1, further comprising predicting whether an electrical device will be off or on at a future time, wherein the first power consumption is used to determine the expected power consumption at the future time when the electrical device is predicted to be off, and the second power consumption is used to determine the expected power consumption at the future time when the electrical device is predicted to be on. ### Claim 8 (depends on 1) The method according to claim 1, wherein the electrical load is an HVAC system, and the HVAC system is shed in response to determining the expected power consumption. ### Claim 9 (depends on 1) The method according to claim 1, wherein the expected power consumption is determined for a future time of at least five minutes. ### Claim 10 (depends on 1) The method according to claim 1, wherein the expected power consumption is determined for a future time of less than 10 hours. ### Claim 11 (depends on 1) The method according to claim 1, wherein the load and/or environmental data comprises outdoor air temperature, a user setpoint and time of day. ### Claim 12 (depends on 1) The method according to claim 1, wherein classifying the input data comprises training a classifier model using historical load and/or environmental data and operating the classifier model after training using current load and/or environmental data to predict the first and second power consumptions. ### Claim 13 (depends on 12) The method according to claim 12, wherein training the classifier model comprises supervised learning where the historical load and/or environmental data comprises defined modes of operation. ### Claim 14 (depends on 12) The method according to claim 12, wherein training the classifier model comprises unsupervised learning where the historical load and/or environmental data does not include defined modes of operation and the unsupervised learning determines modes of operation. ### Claim 15 (depends on 12) The method according to claim 12, wherein the historical load and/or environmental data comprises power usage, outdoor air temperature, a user setpoint and time of day. ### Claim 16 (depends on 1) The method according to claim 1, wherein classifying the input data comprises a machine learning model. ### Claim 17 (depends on 1) The method according to claim 1, wherein classifying the input data comprises a neural network algorithm. ### Claim 18 (depends on 17) The method according to claim 17, wherein the neural network algorithm comprises Perceptron, a decision tree or a support vector machine (SVM). ### Claim 19 (depends on 1) The method according to claim 1, wherein predicting the first power consumption and predicting the second power consumption comprises different regression models. ### Claim 20 (depends on 19) The method according to claim 19, wherein the different regression models each comprises a recurrent neural network (RNN). ### Claim 21 (depends on 1) The method according to claim 1, wherein classifying the input data comprises classifying the input data into more than two sets of data, and further comprising predicting a power consumption corresponding to each of the more than two sets of data. ### Claim 22 (depends on 1) The method according to claim 1, further comprising classifying the input data for at least two different electrical loads, determining the expected power consumption for each of the two different electrical loads based on the first and second power consumptions, respectively, and summing the expected power consumptions for the two different electrical loads to determine an aggregated load. ### Claim 23 (depends on 22) The method according to claim 22, further comprising subtracting the aggregated load from a total load to determine a residual load.