WO2024050409A1 - System and methods for retrofit energy prediction - Google Patents

Abstract

A system for generating retrofit energy predictions may include a plurality of models. One or more of the models may generate a classification for a building. Classifications may include aspects related to the energy consumption of the building, the cost of a retrofit, the feasibility of a retrofit, the predicted benefit of a retrofit, and the like. Classifications generated by the models may be used, along with other data, to generate a score for the suitability of a building for an energy retrofit.

Description

SYSTEM AND METHODS FOR RETROFIT ENERGY PREDICTION

CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of priority to U.S. Provisional Patent Application No. 63/403,084, filed September 1, 2022, entitled "SYSTEM AND METHODS FOR RETROFIT ENERGY PREDICTION "

[0002] The foregoing application is incorporated herein by reference in its entirety for all purposes.

BACKGROUND

[0003] Many structures use outdated and inefficient heating and cooling systems. Inefficient systems cause unnecessary loads on a power grid, cause excess pollution, increase operating costs, and the like. Reducing the number of structures with inefficient heating and cooling system can lead to buildings that are greener, healthier, more cost-efficient, and smarter.

SUMMARY

[0004] In some aspects, the techniques described herein relate to a system for generating retrofit energy predictions, including: a user interface circuit configured to receive selections of a geographic area; file consumption circuit structured to ingest building data, environment data, and energy data for a set of buildings in the geographic area: predictive model circuit structured to generate energy retrofit predictions for the geographic area, wherein the predictive model circuit includes: an energy profile model configured to generate energy predictions for the set of buildings, a first mathematical model configured to classify each building in the set of buildings based on a technical feasibility of a retrofit, a second mathematical model configured to predict a cost of the retrofit for each building in the set of buildings, a third mathematical model configured to classify each building in the set of buildings based on predicted energy savings, a prediction model configured to generate a score as a function of the models, wherein the score represents a suitability of each building for the retrofit; and a reporting circuit configured to generate an output for display on the user interface circuit identifying buildings with the score above a threshold score.

[0005] In some aspects, the techniques described herein relate to a computer-implemented method of generating retrofit energy predictions, including: obtaining a first mathematical model trained on installation complexity data of retrofits; obtaining a second mathematical model trained on cost data for retrofits; obtaining a third mathematical model trained on energy usage data of buildings; receiving a selection of a geographical area, wherein the geographical area includes a set of buildings; computing, using the first mathematical model, a first classification corresponding to a predicted technical feasibility of a retrofit of each building in the set of buildings; computing, using the second mathematical model, a second classification corresponding to a predicted cost of a retrofit of each building in the set of buildings; computing, using the third mathematical model, a third classification corresponding to a predicted energy savings of each building in the set of buildings; generating a score as a function of the classifications from the models, wherein the score represents a suitability of each building for the retrofit; and generating, for presentation at a user interface, a report of the suitability of the set of buildings based on the score.

BRIEF DESCRIPTION OF THE FIGURES

[0006] The invention and the following detailed description of certain embodiments thereof may be understood by reference to the following figures:

[0007] Figure 1 is a schematic depiction of a system for energy prediction according to an embodiment of the present disclosure.

[0008] Figure 2 is a schematic depiction of the predictive system of FIG. 1 according to an embodiment of the present disclosure.

[0009] Figure 3 is a schematic depiction of the predictive circuit of FIG. 2 according to an embodiment of the present disclosure.

[0010] Figure 4 is a flowchart of an example method for generating retrofit energy predictions.

DETAILED DESCRIPTION

[0011] Retrofit of existing structures with modem heating and/or cooling systems may drastically reduce the energy usage of the structure and may result in benefits such as fewer greenhouse emissions, less pollution, cost savings, more reliable operations, and the like. For example, replacing gas or oil-heating furnaces with electric heat pumps may result in significant heating cost savings while further reducing greenhouse emissions.

[0012] As used herein, a structure is any building, such as a residential or commercial building, and may include single-family homes, apartment complexes, multi-story buildings, underground structures, public infrastructure buildings such as train stations, and the like.

[0013] In many cases, structure owners or managers may not be aware that their structures may be eligible for the benefits associated with a retrofit. Performing an analysis of energy dynamics, markets, structure details, and the like to evaluate the benefits of a retrofit is typically outside of the capabilities of most building owners and managers. Traditionally, analysis of a structure has been performed with an energy audit by a specially trained team that analyzes specifics of a structure and calculates potential benefits. However, such manual time analysis may be extremely timeconsuming, costly, and, therefore, disruptive to the building owner or occupants. The complexity, time, cost, scarcity of trained professionals, and other factors have led to few structures being upgraded with new heating and cooling technology. The lack of widespread adoption of new technologies has led to missed opportunities to improve the health and efficiency of communities. [0014] Systems and methods described herein provide for a widespread identification of structures that are eligible for retrofit or upgrade of heating and/or cooling systems. The methods and systems described herein leverage data and advanced modeling to accelerate the decarbonization of building stock. The methods provide automated analysis and structure identification for decarbonization and energy efficiency services with minimal cost and disruption to owners and occupants and may be particularly beneficial to those in frontline communities that are underserved by traditional companies. Systems and methods described herein provide for automated analysis that spans upfront program design, customer identification, engineering analysis financing, and project implementation at large scales such as a whole city, neighborhood, block, and the like.

[0015] In embodiments, structures and areas that are eligible for system upgrades are automatically identified. The systems and methods include map visualizations. In embodiments, features of the systems and methods help design and simulate city-wide building electrification and energy efficiency programs. The methods and system facilitate the modeling of carbon and energy impact and the cost of applying various electrification and energy efficiency measures to buildings.

[0016] In embodiments, visualization enables users to explore block maps to understand the landscape of investment opportunities to target structures for a program. Users can layer various core building characteristics, including building type, building age, energy usage, intensity, and heating fuel type. Users can leverage predictive insights from machine learning models that predict the feasibility of retrofit for a structure (i.e. retrofit to use a heat pump) and how likely that building is to benefit from energy efficiency updates. In embodiments, predictive models automatically provide insights into the degree of need for environmental justice in a building's area.

[0017] In embodiments, dynamic analysis using predictive models is used to simulate and explore the effects of upgrades in an area. Predictive models identify what percentage of the buildings in an area are technically feasible for that equipment, how much it costs to upgrade, and what aggregate carbon and energy usage reductions you can expect at a given adoption rate. To calculate the cost of a program, methods incorporate potential program subsidies and external subsidies. Users can design various program scenarios with different configurations of buildings' equipment subsidies and export those reports to compare the cost and impact across different program scenarios.

[0018] With reference to the example of FIG. I, an example system 100 is depicted schematically for performing predictive analysis, including technical feasibility, cost estimates, a rate, and energy reduction associated with heating and cooling system retrofit. The example system 100 includes and/or is in communication with a user device 110, which may be operated by a user 112. The user device 110 may be, for example, a personal computer, a mobile device such as a phone, and the like that may enable a graphical user interface via an application, web interface, or the like. In certain embodiments, a device 110 may be embodied as a first physical device (e.g., a particular computer, mobile device, tablet, etc.) at a first time and as a second physical device at a second time. For example, user 112 may access system 100 using a user device 110, which is a laptop, at a first time or for a first interaction, and a mobile phone at a second time or for a second interaction.

[0019] The example system includes the predictive system 140, which may communicate with one or more data 130, 120, 150, and user device 110. The predictive system 140 and user device 110 are depicted as a single device for clarity of the present description, but each of the predictive system and/or devices may additionally or alternatively be a distributed device and/or may be embodied as distinct hardware devices at different times or during certain operations. The data 120, 130, and 150 are depicted as separate data sources for clarity, but each data source may additionally or alternatively be a distributed data source, or all sources may be provided by one source. The predictive system may provide and receive data for a graphical user interface at the user device 110. In one example, the interface on the user device 110 may include a map of an area. The map may allow a user to make selections of an area. The selections may be identified to the predictive system 140. The predictive system 140 may query data 120, 130, and 150 for data related to the selection and generate predictions for the structures in the selection. The predictive system may generate predictions with respect to one more technical feasibility of upgrades, cost estimates, adoption rate, and energy reduction for the structures. The prediction system may process the prediction data and provide the predictions to the interface of the user device 110. The user device 110 may display the predictions as tables, color-coded maps, reports, and other formats.

[0020] With reference to the example of FIG. 2, an example predictive system 140 includes a number of circuits structured to functionally execute one or more operations of the predictive system 140. The examples throughout the present disclosure depict the controller predictive system as a single device, including circuits thereon, but the predictive system 140 and/or circuits thereof may be distributed, in whole or part, across a number of hardware devices, and/or may be embodied at least in part in one or more hardware devices. The terms controller, circuit, processor, computing device, and similar terms as used throughout the present disclosure, should be understood broadly. Embodiments of these terms include, without limitation: computing devices and/or elements thereof configured to perform one or more operations of the associated circuit and/or controller; logic circuits configured to perform one or more operations thereof; hardware devices configured to be responsive to commands and/or user interactions to perform one or more operations thereof (e g., an I/O device, screen, keyboard, mouse); a server hosting one or more aspects of a circuit and/or controller and/or storing data related to the controller and/or circuit; computer-readable instructions configured to, when executed by a processor, perform one or more operations of the circuit and/or controller. In certain embodiments, a circuit may be distributed across a number of devices, such as a portion of a circuit embodied in the predictive system 140 and another portion of the circuit embodied on a user device 110, and the distribution of the circuit may vary based on the specific configuration of a system, at different times for a given system, and/or for distinct interactions or predictions for a given system.

[0021] An example predictive system 140 includes a file consumption circuit 202 structured to ingest data files 220, which may include building data 222, environment data 224, and/or energy data 226. In certain embodiments, the files may be separate files or may be large contiguous data sources. The file consumption circuit 202 may digest the data files 220 and identify relevant data for the selected map area or user search values 230 received from the user interface 266. The file consumption circuit 202 may identify selected map coordinates and identify relevant data from the data files. The circuit 202 may map addresses from the search values to specific structure data, environment data, and the like.

[0022] An example predictive system 140 includes a predictive model circuit 204 structured to generate energy retrofit predictions for the selected areas based on the data files 220. In embodiments, the predictive model circuit may include a trained model, such as a neural network that may be trained to identify or classify structures based on their suitability for retrofit. In embodiments, the trained neural network model may identify suitable structures. Another computation model may be configured to perform additional calculations using data from the data files 220 to determine predictive energy savings, cost savings, and the like. In embodiments, the predictive model circuit 204 may be a model that may be executed on distributed hardware such as a cloud computing platform, server farm, and the like. In embodiments, multiple instances of the predictive model circuit 204 may be instantiated and executed for parallelized prediction for structures.

[0023] An example predictive system 140 includes a computation distribution circuit 206. The computation distribution circuit 206 may access one or more remote computation services 250, such as cloud computing platforms, server farms, and the like, and instantiate multiple copies of the predictive model circuits 204 for parallel processing. In embodiments, the computation distribution circuit 206 may utilize services such as AWS services, AWS Batch, Elastic Container Services 252, and the like to instantiate multiple copies of models for the prediction and classification of structures. In embodiments, the computation distribution circuit 206 may dynamically adjust the number of containers and copies of the predictive model circuit are instantiated based on parameters such as the number of structures selected, computation budget, latency requirements, and the like. [0024] In embodiments, a predictive model circuit 204 may include a plurality of models that may process and classify structures according to different criteria. In one example, one model may be a model that may be trained to classify structures based on the predicted technical feasibility of retrofitting the structure. In another example, another model may be a trained model that may be trained to classify structures based on the predicted cost of retrofitting the structure. In another example, another model may be a trained model that may be trained to classify structures based on the predicted energy savings. In embodiments, the computation distribution circuit may be configured to determine an appropriate model for the structure based on initial parameters of the structure, geographical area, known energy usage, and the like.

[0025] In embodiments, the computation distribution circuit 206 may be further configured to assemble and merge the results of computations using different models and different containers. The computation distribution circuit may generate a data structure of the results of the predictions for display by the user interface 266.

[0026] With reference to the example of FIG. 3, a predictive model circuit 204 may include a plurality of models for making predictions and/or calculations regarding the heating and cooling systems of a structure. In embodiments, a predictive model circuit 204 may include energy profile model (s) 310. The energy profile model(s) 310 may receive building data (such as type of construction, insulation type, facade, exposure, roof type, number of windows, and the like), environment data (such as location, the yearly temperature in the environment, weather data, and the like), and/or energy data (such as energy bills, heating and cooling days, gas consumption, and the like). The energy profile model(s) 310 may generate an energy model of the structure. Energy profile models may be locally managed and/or curated by a third party and may be accessed via an application programming interface (API). The output of the energy model may include energy predictions such as heating and cooling requirements (for example, amount of British thermal units (BTUs) or joules (J) to heat a structure) of the structure as a function of temperature, weather, and other conditions.

[0027] The predictive model circuit 204 may further include one or more models 320, 321, 322. In embodiments, the models 320, 321, and 322 may be one or more classification models, analytic models, heuristic models, and the like. Models 320, 321, and 322 may each receive the output of the energy profile models 310 and/or the data 120, 130, and 150. Models 320, 321, and 322 may be machine learning models trained to classify the structures according to one or more criteria such as technical feasibility, cost, energy savings, and the like. The models may rate or bin the structures in one or more bins (such as low, medium, high), provide a confidence measure (such as a value between 0 and 1), and/or other classification output. Although three models are shown in Fig. 3, it is to be understood that any number of models may be used. In embodiments, different sets of models may be used based on the output of the energy profile models 310 and/or data 120, 130, and 150. In embodiments, the models may include random forest models, bagging models, logic regression models, and the like.

[0028] In one example, model 320 may be configured to calculate the predicted technical feasibility of a retrofit of a structure. Model 320 may be trained to generate an output that is indicative of the feasibility of installing a heat pump for a building. In one example, model 320 may be configured to generate a continuous probability score (i.e., a value between 0 and 1) that is indicative of the feasibility of the retrofit based on the characteristics of each building. In one example, model 320 may be configured to generate classification labels for each building (such as a “low," "medium," and "high" feasibility), and the classification labels may be based on ranges of values of a predicted probability output of the model.

[0029] Model 320 may be a trained regression model such as a gradient boosting regression model. Regression models are a set of statistical processes for estimating the relationships among variables. A gradient-boosting regression model is a machine learning model that produces a prediction model in the form of an ensemble of weak prediction models. In some implementations, the prediction models may be decision trees. In one implementation, model 320 may receive building data such as: building type, heating fuel type, the total energy usage of the building, area of, year built, income level in area, and the like. Model 320 may generate a numerical value that is representative of the predicted technical feasibility.

[0030] In some implementations, inputs to model 320 may be converted to numerical or categorical features. Categorical features may be processed using techniques like one-hot encoding or ordinal encoding. Numerical features may be used directly by the model, or they may be transformed in various ways (e.g., scaling, normalization) depending on the specifics of the problem. In some implementations, input data may include text data such as descriptions or words. Text may be converted into a numerical format using tokenization followed by vectorization, such as TF-IDF (term frequency -inverse document frequency) or a pre-trained word embedding like Word2Vec or GloVe.

[0031] Model 320 may be trained based on the outputs obtained from a plurality of data sets. In some implementations, model 320 may be trained based on outputs from rule-based or heuristic models from one or more geographic areas. The rule-based models may use local datasets, such as permit data, that can be used as a proxy to estimate the feasibility of retrofit. A rule-based model may be used to generate labeled training data sets. For example, building permit data may include parameters of a building, and a label of feasibility may be generated based on a rule associated with an indication of completion or closing of the building permit. In some implementations, training data sets generated from local data, such as building permit data, may be used to generate regional or national training data sets by filtering or removing parameters or data that is only particular to a small locality or region.

[0032] In some implementations, training data sets may be further processed using one or more filtering and/or post-processing operations. In one example, a training data set may be processed to remove outliers in the data set. In some implementations, data elements in the data set may be considered outliers if they are beyond 3 interquartile range (IQR). In some implementations, training data samples that include outliers may be removed or filtered before training a model.

[0033] In some implementations, outliner elements may be treated as missing values such that they can be flagged for revision by a user, synthesized from other data, and/or imputed. In one example, missing elements (that may have been outliers) may be imputed from other data of the training sample using iterative imputation techniques. Iterative imputation models each feature with missing values as a function of other predictors and uses that estimate for imputation. Imputation treats each variable with missing data as the dependent variable in a regression, with other variables as predictors, and may use a regression model to estimate missing values. In implementations, any appropriate type of imputation may be used, such as mean/median/mode imputation, hot deck imputation, cold deck imputation, and the like.

[0034] Training of model 320 may include training using gradient boosting techniques. Gradient boosting may start by fitting a simple model, such as a decision tree, to the training data. Residuals (differences between the model's predictions and the true values) are calculated. A new model is then fit to these residuals, and the predictions from this new model are then added to the predictions of the original model. This has the effect of correcting the original model's errors, thereby boosting its performance. The process is iteratively repeated to produce a sequence of models that are each fit to the residuals of the previous model. The final prediction is obtained by summing up the predictions from all these models.

[0035] In some implementations, model 322 may be configured to calculate an energy efficiency potential (EEP) score. The EEP score may quantify a prediction of how much a building could benefit from a retrofit. The EEP score may be indicative or a proxy for predicted energy savings or energy cost savings. In some implementations, model 322 may be a trained regression model, such as a gradient-boosting regression model.

[0036] In one implementation, model 322 may be a trained model that estimates the energy code release of each building. International Code Council (ICC), releases revisions of the International Energy Conservation Code (IECC, herein referred to as Energy Code) every three years starting in 1927. The IECC provides the minimum requirements for energy-efficient buildings and covers things like insulation, windows, air sealing, HVAC systems, and lighting. The IECC code provides specifications for design and construction to achieve energy' efficiency in buildings.

[0037] Model 322 may be trained to estimate the year of the energy code or the version of the energy code based on location (i.e., latitude, longitude), building type, and energy use intensity (i.e., kilowatt-hour per square foot) of a building. After the model is trained, residuals between the actual Energy Code of a building and the predicted energy code are calculated. The actual Energy Code used for a building may be determined from the year the building was constructed, from building permit data, and the like. The residual is a positive value when the predicted Energy Code version or year is less than the actual energy code version or year. A positive residual may indicate that the building is less efficient than expected, given the efficiency standards of the year it was built. A positive residual may indicate that there may be an opportunity to improve the building's energy efficiency.

[0038] In some implementations, the energy' efficiency potential category may be calculated for each building based on the value of residuals for each building. In one example, if the residual has a value of less than 1, the building has a low potential for energy improvement and may be classified in the "low" category. If the residual has a value that is more significant than one (i.e., 2+) and within the 95 percentile of residuals for the building class, then the building may be classified in the "medium" category. Beyond these thresholds, a "high" category may be assigned to the building. In embodiments, any appropriate threshold values of the residual value may be used for determining EEP categories.

[0039] In some implementations, model 321 may be configured to calculate the cost of a retrofit. In some implementations, model 321 may be an analytical model that determines a predicted cost based on current market costs and/or historical prices. In some implementations, model 321 may be a classification model that predicts a category of the cost of a retrofit. A category of the cost may correspond to a range of predicted costs. In some implementations, model 321 may be a trained neural network configured to estimate the cost of a retrofit and may be trained on historical cost data for retrofits.

[0040] In some implementations, calculating energy usage, energy efficiency potential, and/or feasibility of retrofit may be based on energy usage, construction costs, and energy savings of similar buildings and retrofits of similar buildings. Similar buildings may be determined by comparing parameters and characteristics of buildings in the same geographic area (i.e., same city, county, state) or another geographic area with similar characteristics (i.e., similar demographics, income level, housing, etc.). In one example, similarity may be computed according to a similarity function. A

similarity function may be calculated as:

[0041] where yt is the variable corresponding to a characteristic of a reference building and y is the corresponding variable of the building that it compared. The total similarity between a building and

other buildings may be calculated as:

[0042] In reference to Fig. 3, the predictive model circuit 204 may further include a prediction and/or classification model 323. The model 323 is configured to determine the overall suitability of a structure for a heating and/or cooling system retrofit. The model 323 may take as input outputs of the models 320,321,322, profile models 310, and/or data 120, 130, 150, and generate a rating or prediction for a structure. In embodiments, the classification model 323 may generate a score based on a function of the outputs of the models 320, 321, 322 (such as a weighted sum). In some embodiments, model 323 may be a trained neural network. In embodiments, model 323 may generate the predictions and/or ratings 340 and/or signal a numerical estimation model 324 to generate numerical predictions of the cost, feasibility, savings, carbon reduction, and the like for a structure. In embodiments, the estimation model 324 may be triggered when the prediction 340 from the model 323 exceeds a prediction threshold. The triggering of estimation model 324 based on a prediction threshold may reduce computational requirements as the estimations may only be computed for structures where a retrofit is feasible.

[0043] In embodiments, the system may include a plurality of arrangements of the predictive model circuit 204. Each version of the predictive model circuit 204 may include different versions and/or combinations of the models (such as models 310, 320, 321, 323, 324). Each version of the predictive model circuit 204 may be tailored so specific geography, building type, climate, and the like. In embodiments, a plurality of predictive models may be preconfigured with different models and associated with an indication as to the performance of the version for different geographies, climates, building types, building sizes, and the like. In embodiments, at least a subset of the versions may be trained on a subset of the data that is specific to the desired parameters (geography, climate, building type, etc.).

[0044] In embodiments, at least a subset of the versions of the predictive model circuit 204 may be trained on all of the available data for all parameters (geography, climate, building type, etc.). The performance of each version may then be evaluated for each version on different parameters using available data such as historical data of installations. During the evaluation of the models, the system may generate a matrix of versions and the performance of the versions for the parameters. During the system's operation, a model control circuit 302 may identify the parameters of the data and select the appropriate version of the models. In embodiments, predictive model circuit 204 may be dynamically arranged based on the detected in data parameters.

[0045] In embodiments, models described herein may use various training techniques, including supervised and unsupervised training. Predictive models may be trained using historical data from installations that include data about the building and data after the installation. In embodiments, models may be refined using data from on-the-ground observations such as consultations, appraisals, and estimates from professionals. Estimate data from the models may be identified as accurate or inaccurate by professionals on the ground and the feedback may be used to refine and/or retrain the trained models.

[0046] In embodiments, data (such as data 120, 130, 150) may be nonuniform across regions and or times. For example, different regions may have different rules regarding energy usage reporting. In some areas, energy usage data may be fine-grained data and may include energy usage data from residential and commercial structures. In some areas, reporting requirements may only require reporting for commercial structures. Nonuniform data may results in unwanted bias for models trained based on the data in some regions or times. For example, in regions where energy usage data is received only for commercial structures may cause an unwanted bias towards higher energy usage assumptions since many commercial structures use more energy than residential structures. In embodiments, data may be processed to identify bias in data. In embodiments, data may be processed to normalize and correct data across areas and times. In embodiments, data may be normalized before the training of models. In some cases, models may be trained on unprocessed data and classifications using the models may be performed on normalized data.

[0047] In embodiments, processing of data may include normalizing the data and may include finding or generating missing data. In some cases, data that is missing from one region may be inferred or calculated from other related data. In embodiments, data processing may include data abstracting. Data values may be converted to abstract units to provide a more uniform prediction between regions. For example, in some regions, gas usage may be common while wind energy may be more common in other regions. The energy usage estimates may be converted to an abstract data value and may allow comparison and more consistent model training for the data of the two areas. [0048] In embodiments, the systems and methods described here may be used to provide a dynamic analysis of a selection of buildings. The analysis models can provide a summary of the percentage of the selected buildings that are technically feasible for upgrade equipment, the cost of upgrades, and the expected aggregate carbon and energy usage reductions. [0049] In embodiments, model data results may be used to generate instant building reports. The instant building reports may provide building owners with data about what equipment is the best fit for their building, the monthly savings, and payback. The reports may include local energy program information, the performance of historical projects in the vicinity, and the like.

[0050] In embodiments, the systems and methods described here may be used to provide a buildings dashboard. A buildings dashboard may be a visual analytic space dashboard to help teams analyze the buildings. Data may be aggregated from different sources and combined into a refined data set. The dashboard may provide a view of the distribution based on the heating fuel type, distribution based on the cooling type, distribution based on the square footage, the distribution based on the various different kinds of rooms or the zones, and the like.

[0051] Fig. 4 is a flowchart of an example method 400 for generating retrofit energy predictions. In one example, method 700 may be implemented by the systems and models described with respect to Figs. 1-3. At step 410, a mathematical model that is trained on installation complexity data of retrofits is obtained. In one example, the model may be model 320 as described herein. Step 410 may further include obtaining a second mathematical model trained on cost data for retrofits. In one example, the model may be a model 321 as described herein. Step 410 may further include obtaining a third mathematical model trained on energy usage data of buildings. In one example, the model may be model 322 as described herein. As used herein, obtaining a model may include installing a model, initializing or enabling a communication channel with the model, training the model, and/or the like.

[0052] At step 420, a selection of a geographical area may be received. The selection may be a geographical area that includes a set of buildings. The selection may be received from a user interface. In one example, the selection may be received from a selection on a map. In another example, selection may be specified using a zip code, tow n, address, distance radius around a location, latitude and longitude coordinates, and the like.

[0053] At step 430, using the first mathematical model, a first classification corresponding to the predicted technical feasibility of a retrofit of each building in the set of buildings may be computed. A classification may be a category such as low, medium, or high. In some implementations, a classification may be a numerical value. At step 440, using the second mathematical model, a second classification corresponding to a predicted cost of a retrofit of each building in the set of buildings may be computed. The model used in step 440 may calculate an estimated cost that may be a numerical value and further assign a classification to the cost based on the value of the cost, the relative value of the cost compared to the value of the building, and the like. At step 450, using the third mathematical model, a third classification corresponding to a predicted energy savings of each building in the set of buildings may be computed. In implementations, the classification may be determined from historical data of similar buildings that were retrofitted. In some implementations, the classification may be determined using a regression model trained to compare predicted energy code versions as described herein.

[0054] At step 460, a score that represents the suitability of each building for a retrofit may be generated. The score may be a function of the classifications generated by the models. In one example, the score may be a weighted function of the classifications. In some implementations, the score may be generated in response to a selection and may be calculated in real time. In some implementations, the score may be pre-generated and stored generating a score may include retrieving the score from a database.

[0055] At step 470, a report of the suitability of the set of buildings based on the score may be generated. The report may be generated for display on a graphical user interface. The report may include a summary of the number of buildings in a selected area that are predicted to be suitable for retrofit. The report may include an option to drill down on each building to identify details of each building, classifications, outputs of models, and other data.

[0056] The methods and systems described herein may be deployed in part or in whole through network infrastructures. The network infrastructure may include elements such as computing devices, servers, routers, hubs, firewalls, clients, personal computers, communication devices, routing devices and other active and passive devices, modules and/or components as known in the art. The computing and/or non-computing device(s) associated with the network infrastructure may include, apart from other components, a storage medium such as flash memory, buffer, stack, RAM, ROM and the like. The processes, methods, program codes, instructions described herein and elsewhere may be executed by one or more of the network infrastructural elements.

[0057] The methods, program codes, and instructions described herein and elsewhere may be implemented on a cellular network having multiple cells. The cellular network may either be a frequency division multiple access (FDMA) network or a code division multiple access (CDMA) network. The cellular network may include mobile devices, cell sites, base stations, repeaters, antennas, towers, and the like. The cell network may be a GSM, GPRS, 3G, EVDO, mesh, or other network types.

[0058] The methods, programs codes, and instructions described herein and elsewhere may be implemented on or through mobile devices. The mobile devices may include navigation devices, cell phones, mobile phones, mobile personal digital assistants, laptops, palmtops, netbooks, pagers, electronic book readers, music players and the like. These devices may include, apart from other components, a storage medium such as a flash memory, buffer, RAM, ROM and one or more computing devices. The computing devices associated with mobile devices may be enabled to execute program codes, methods, and instructions stored thereon. Alternatively, the mobile devices may be configured to execute instructions in collaboration with other devices. The mobile devices may communicate with base stations interfaced with servers and configured to execute program codes. The mobile devices may communicate on a peer-to-peer network, mesh network, or other communications network. The program code may be stored on the storage medium associated with the server and executed by a computing device embedded within the server. The base station may include a computing device and a storage medium. The storage device may store program codes and instructions executed by the computing devices associated with the base station.

[0059] The computer software, program codes, and/or instructions may be stored and/or accessed on machine readable media that may include: computer components, devices, and recording media that retain digital data used for computing for some interval of time; semiconductor storage known as random access memory (RAM); mass storage typically for more permanent storage, such as optical discs, forms of magnetic storage like hard disks, tapes, drums, cards and other types; processor registers, cache memory, volatile memory, non-volatile memory; optical storage such as CD, DVD; removable media such as flash memory (e.g. USB sticks or keys), floppy disks, magnetic tape, paper tape, punch cards, standalone RAM disks, Zip drives, removable mass storage, off-line, and the like; other computer memory such as dynamic memory, static memory, read/write storage, mutable storage, read only, random access, sequential access, location addressable, file addressable, content addressable, network attached storage, storage area network, bar codes, magnetic ink, and the like.

[0060] The methods and systems described herein may transform physical and/or intangible items from one state to another. The methods and systems described herein may also transform data representing physical and/or intangible items from one state to another.

[0061] The elements described and depicted herein, including in flow charts and block diagrams throughout the figures, imply logical boundaries between the elements. However, according to software or hardware engineering practices, the depicted elements and the functions thereof may be implemented on machines through computer executable media having a processor capable of executing program instructions stored thereon as a monolithic software structure, as standalone software modules, or as modules that employ external routines, code, services, and so forth, or any combination of these, and all such implementations may be within the scope of the present disclosure. Examples of such machines may include, but may not be limited to, personal digital assistants, laptops, personal computers, mobile phones, other handheld computing devices, medical equipment, wired or wireless communication devices, transducers, chips, calculators, satellites, tablet PCs, electronic books, gadgets, electronic devices, devices having artificial intelligence, computing devices, networking equipment, servers, routers and the like. Furthermore, the elements depicted in the flow chart and block diagrams or any other logical component may be implemented on a machine capable of executing program instructions. Thus, while the foregoing drawings and descriptions set forth functional aspects of the disclosed systems, no particular arrangement of software for implementing these functional aspects should be inferred from these descriptions unless explicitly stated or otherwise clear from the context. Similarly , it will be appreciated that the various steps identified and described above may be varied, and that the order of steps may be adapted to particular applications of the techniques disclosed herein. All such variations and modifications are intended to fall within the scope of this disclosure. As such, the depiction and/or description of an order for various steps should not be understood to require a particular order of execution for those steps, unless required by a particular application, or explicitly stated or otherwise clear from the context.

[0062] The methods and/or processes described above, and steps thereof, may be realized in hardware, software or any combination of hardware and software suitable for a particular application. The hardware may include a general-purpose computer and/or dedicated computing device or specific computing device or particular aspect or component of a specific computing device. The processes may be realized in one or more microprocessors, microcontrollers, embedded microcontrollers, programmable digital signal processors or other programmable device, along with internal and/or external memory. The processes may also, or instead, be embodied in an application specific integrated circuit, a programmable gate array, programmable array logic, or any other device or combination of devices that may be configured to process electronic signals. It will further be appreciated that one or more of the processes may be realized as a computer executable code capable of being executed on a machine-readable medium.

[0063] The computer executable code may be created using a structured programming language such as C, an object oriented programming languages such as C++, or any other high-level or low-level programming language (including assembly languages, hardware description languages, and database programming languages and technologies) that may be stored, compiled or interpreted to run on one of the above devices, as well as heterogeneous combinations of processors, processor architectures, or combinations of different hardware and software, or any other machine capable of executing program instructions.

[0064] Thus, in one aspect, each method described above and combinations thereof may be embodied in computer executable code that, when executing on one or more computing devices, performs the steps thereof. In another aspect, the methods may be embodied in systems that perform the steps thereof, and may be distributed across devices in a number of ways, or all of the functionality may be integrated into a dedicated, standalone device or other hardware. In another aspect, the means for performing the steps associated with the processes described above may include any of the hardware and/or software described above. All such permutations and combinations are intended to fall within the scope of the present disclosure.

[0065] While the invention has been disclosed in connection with the preferred embodiments shown and described in detail, various modifications and improvements thereon will become readily apparent to those skilled in the art. Accordingly, the spirit and scope of the present invention is not to be limited by the foregoing examples, but is to be understood in the broadest sense allowable by law;

[0066] All documents referenced herein are hereby incorporated by reference in their entirety.

Claims

CLAIMS What is claimed is:

1. A system for generating retrofit energy predictions, comprising: a user interface configured to receive selections of a geographic area; file consumption circuit structured to ingest building data, environment data, and energy data for a set of buildings in the geographic area; predictive model circuit structured to generate energy retrofit predictions for the geographic area, wherein the predictive model circuit comprises: an energy profile model configured to generate energy predictions for the set of buildings, a first mathematical model configured to classify each building in the set of buildings based on a technical feasibility of a retrofit, a second mathematical model configured to predict a cost of the retrofit for each building in the set of buildings, a third mathematical model configured to classify each building in the set of buildings based on predicted energy savings, and a prediction model configured to generate a score as a function of the models, wherein the score represents a suitability of each building for the retrofit; and a reporting circuit configured to generate an output for display on the user interface identifying buildings with the score above a threshold score.

2. The system of claim 1, wherein the energy predictions comprise heating and cooling requirements for each building.

3. The system of claim 1, wherein the energy profile model generates predictions as a function of temperature and weather.

4. The system of claim 1, wherein the predictive model circuit comprises a computation distribution circuit

5. The system of claim 1, wherein the prediction model is further configured to generate numerical energy predictions and determine structures within a threshold value.

6. The system of claim 1, wherein the prediction model is configured for a specific geography, climate, or building size.

7. The system of claim 1, further comprising a missing data component configured to generate data abstractions for missing data using iterative imputation.

8. The system of claim 1, wherein the first mathematical model is a gradient-boosting regression model.

9. The system of claim 1, wherein the third mathematical model is a gradient-boosting regression model.

10. The system of claim 9, wherein the third mathematical model is configured to determine an energy efficiency potential classification based on a residual value between a predicted energy code of the building and an actual energy code of the building.

11. A computer-implemented method of generating retrofit energy predictions, comprising: obtaining a first mathematical model trained on installation complexity data of retrofits; obtaining a second mathematical model trained on cost data for retrofits; obtaining a third mathematical model trained on energy usage data of buildings; receiving a selection of a geographical area, wherein the geographical area includes a set of buildings; computing, using the first mathematical model, a first classification corresponding to a predicted technical feasibility of a retrofit of each building in the set of buildings; computing, using the second mathematical model, a second classification corresponding to a predicted cost of a retrofit of each building in the set of buildings; computing, using the third mathematical model, a third classification corresponding to a predicted energy savings of each building in the set of buildings; generating a score as a function of the classifications from the models, wherein the score represents a suitability of each building for the retrofit; and generating, for presentation at a user interface, a report of the suitability of the set of buildings based on the score.

12. The method of claim 11, wherein the first classification is based on predicted heating and cooling requirements for each building.

13. The method of claim 11 , wherein the first classification is based on a predicted temperature and weather.

14. The method of claim 11, wherein generating the score comprises processing the first classification, the second classification, and the third classification with a trained prediction model.

15. The method of claim 14, wherein the trained prediction model is further configured to generate numerical energy predictions and determine structures that are within a threshold value.

16. The method of claim 14, wherein the prediction model is configured for a specific geography, climate, and building size.

17. The method of claim 11 , further comprising generating data abstractions for missing data using iterative imputation.

18. The method of claim 11, wherein the first mathematical model is a gradient-boosting regression model.

19. The method of claim 11 , wherein the third mathematical model is a gradient-boosting regression model.

20. The method of claim 11, wherein the third mathematical model is configured to determine an energy efficiency potential classification based on a residual value between a predicted energy code of the building and an actual energy code of the building.