Load Prediction for Electrical Equipment Using Machine Learning

Abstract

An embodiment discloses predicting, by a processor circuit, a load parameter value of an electric device at a future time based on at least one machine learning model and a plurality of load parameter values ​​including a set of a predetermined number of load parameter values ​​extracted from a time-series data stream of load parameter values ​​obtained for the electric device, then calculating, by the processor circuit, an overload capacity for the future time based on the predicted load parameter values, and modifying, by the processor circuit, at least one parameter associated with the electric device at a current time based on the calculated overload capacity for the future time.

Description

[Background technology]

[0001] background The present disclosure relates to analysis of electrical equipment such as high voltage transformers. In particular, the present disclosure relates to load forecasting for electrical equipment using machine learning. [Technical field]

[0002] Traditional load forecasting techniques for electrical equipment, such as transformers, typically use complex statistical tools to estimate short-term and long-term forecasts for load, hotspot temperature, overload capacity, and other parameters. These tools, such as Auto Regressive Integrated Moving Average (ARIMA), can split the time series of historical load data into several components, such as trend component, seasonal variation component, random variation component, and arrive at a complex model that aggregates these several components into a final forecast. To reduce the forecast error, these tools can also incorporate additional correlated parameters, such as daily average temperature, seasonality, festivals and sporting events, which may add additional complexity and require specialized tools and may require specialized data for different times and regions, such as countries or regions with different cultures, holidays, climates, etc.

[0003] In the case of transformers, it may be desirable to obtain accurate estimates of the load for a future time, such as one or two hours into the future, to plan for load shedding, the need for overloading for contingencies, and the eventual removal of the transformer from operation based on the transformer's life and remaining life or the changing needs of the power system. However, due to the complexity of existing solutions, many types of electrical equipment lack the computational power required to accurately estimate the future load. Summary of the Invention [Means for solving the problem]

[0004] overview According to some embodiments, the method includes predicting, by a processor circuit, a load parameter value of the electric equipment at a future time based on at least one machine learning model and a plurality of load parameter values ​​including a set of a predetermined number of load parameter values ​​extracted from a time series data stream of load parameter values ​​obtained for the electric equipment. The method further includes calculating, by the processor circuit, an overload capacity for the future time based on the predicted load parameter values. The method further includes modifying, by the processor circuit, at least one parameter associated with the electric equipment at a current time based on the calculated overload capacity for the future time.

[0005] According to some embodiments, at least one machine learning model is trained based on a plurality of determined relationships between a predetermined number of load parameter values ​​and at least one subsequent load parameter value from the time series data stream acquired over a predetermined period of time.

[0006] According to some embodiments, the plurality of determined relationships are verified based on a comparison of at least one expected load parameter value derived from a predetermined number of load parameter values ​​to at least one subsequent load parameter value.

[0007] According to some embodiments, the prediction of future time load parameter values ​​is performed continuously using a predetermined number of load parameter values ​​as an input set to at least one machine learning model, the input set being continuously generated from a time series data stream of load parameter values ​​using a moving window technique.

[0008] According to some embodiments, the plurality of load parameter values ​​includes a set of at least five load parameter values ​​iteratively extracted from a stream of load parameter values ​​obtained from the electrical equipment.

[0009] According to some embodiments, the future time for the predicted load parameter value is at least one hour after the prediction.

[0010] According to some embodiments, the at least one machine learning model predicts a load parameter value of the electrical equipment at a future time based solely on the plurality of load parameter values.

[0011] According to some embodiments, the at least one machine learning model predicts a load parameter value of the electrical equipment at a future time based on the plurality of load parameter values ​​and at least one temperature parameter associated with the electrical equipment.

[0012] According to some embodiments, the method further includes predicting, by the processor circuitry, at least one hot spot temperature value for the component of the electrical equipment at a future time based on the predicted load parameter value.

[0013] According to some embodiments, the method further includes predicting, by the processor circuitry, at least one overload capacity value for the electrical equipment based on the predicted load parameter values, the at least one overload capacity value being associated with at least one time period following the future time.

[0014] According to some embodiments, the electrical device includes a transformer, and the method further includes operating the transformer based at least in part on the at least one changed parameter.

[0015] According to some embodiments, operating the transformer includes operating at least one cooling component of the transformer in response to the predicted load parameter value to change a temperature of at least one component of the electrical equipment prior to the future time.

[0016] According to some embodiments, a monitoring device includes a processor circuit and a memory including machine-readable instructions. When executed by the processor circuit, the machine-readable instructions cause the processor circuit to predict a load parameter value of the electric equipment at a future time based on at least one machine learning model and a plurality of load parameter values ​​including a set of a predetermined number of load parameter values ​​extracted from a time series data stream of load parameter values ​​obtained for the electric equipment. The machine-readable instructions further cause the processor circuit to calculate an overload capacity for the future time based on the predicted load parameter values. The machine-readable instructions further cause the processor circuit to modify at least one parameter associated with the electric equipment at a current time based on the calculated overload capacity for the future time.

[0017] According to some embodiments, at least one machine learning model is trained based on a plurality of determined relationships between a predetermined number of load parameter values ​​and at least one subsequent load parameter value from the time series data stream acquired over a predetermined period of time.

[0018] According to some embodiments, the plurality of load parameter values ​​includes a set of at least five load parameter values ​​iteratively extracted from a stream of load parameter values ​​obtained from the electrical equipment.

[0019] According to some embodiments, the future time for the predicted load parameter value is at least one hour after the prediction.

[0020] According to some embodiments, a non-transitory computer-readable medium includes instructions that, when executed by a processor circuit, cause the processor circuit to predict a load parameter value for an electric equipment at a future time based on at least one machine learning model and a plurality of load parameter values ​​including a set of a predetermined number of load parameter values ​​extracted from a time series data stream of load parameter values ​​obtained for the electric equipment. The instructions further cause the processor circuit to calculate an overload capacity for the future time based on the predicted load parameter values. The instructions further cause the processor circuit to modify at least one parameter associated with the electric equipment at a current time based on the calculated overload capacity for the future time.

[0021] According to some embodiments, at least one machine learning model is trained based on a plurality of determined relationships between a predetermined number of load parameter values ​​and at least one subsequent load parameter value from the time series data stream acquired over a predetermined period of time.

[0022] According to some embodiments, the plurality of load parameter values ​​includes a set of at least five load parameter values ​​iteratively extracted from a stream of load parameter values ​​obtained from the electrical equipment.

[0023] According to some embodiments, the future time for the predicted load parameter value is at least one hour after the prediction.

[0024] BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this application, illustrate certain non-limiting embodiments of the inventive concepts. [Brief description of the drawings]

[0025] [Figure 1] FIG. 1 illustrates a flat file used by a machine learning application for the iterative application of a multivariate algorithm, according to some embodiments. [Figure 2A]FIG. 1 illustrates the conversion of a single-variable dataset representing historical load data into a flattened dataset for use with a machine learning algorithm in predicting future load on an electrical device, according to some embodiments. [Figure 2B] FIG. 1 illustrates the conversion of a single-variable dataset representing historical load data into a flattened dataset for use with a machine learning algorithm in predicting future load on an electrical device, according to some embodiments. [Diagram 3] FIG. 1 illustrates operations for training and selecting a machine learning model for predicting future load of an electric device based on historical load data, according to some embodiments. [Figure 4] FIG. 4 illustrates a visualization of the operations of FIG. 3 on training and validation datasets for a machine learning model, according to some embodiments. [Figure 5A] 1 is a graph plot illustrating a comparison of actual load over time and predicted load over time for an electrical device using a machine learning model, according to some embodiments. [Figure 5B] 1 is a graph plot illustrating a comparison of actual load over time and predicted load over time for an electrical device using a machine learning model, according to some embodiments. [Figure 5C] 1 is a graph plot illustrating a comparison of actual load over time and predicted load over time for an electrical device using a machine learning model, according to some embodiments. [Figure 6] 1 is a flowchart of operations for predicting a load parameter value for a future time based on historical load data, according to some embodiments. [Figure 7] 7 is a graph plot illustrating past load over time, predicted load for a future time, and determined future overload capacity using the operations of FIG. 6 in accordance with some embodiments. [Figure 8] 1 is a block diagram illustrating a load forecasting system for performing operations according to some embodiments. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0026] Detailed Description The inventive concept will be described more fully below with reference to the accompanying drawings, in which examples of embodiments of the inventive concept are shown. However, the inventive concept may be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the inventive concept to those skilled in the art. It should also be noted that these embodiments are not mutually exclusive. An element from one embodiment may be implicitly assumed to be present / used in another embodiment.

[0027] The following description presents various embodiments of the disclosed subject matter. These embodiments are presented as instructional examples and should not be construed as limiting the scope of the disclosed subject matter. For example, certain details of the described embodiments can be modified, omitted, or expanded without departing from the scope of the described subject matter.

[0028] Transformer overload capacity can be estimated for the current time based on existing dynamic thermal models and current ambient temperature and load conditions, allowing for continuous calculation of hot spot temperature and resulting life consumption. To calculate overload capacity for a given future time interval, the model can assume that the hot spot temperature will not exceed a given limit (e.g., 110C for thermally improved kraft paper and 98C for regular kraft paper) to avoid unnecessarily shortening the transformer's operating life. Based on the current state of the transformer, an optimal "k" load factor can be calculated that allows the transformer to be overloaded at different levels for different corresponding times without sacrificing life.

[0029] Accurate prediction of future load values ​​can aid in effective load management of power equipment and transmission / distribution lines in power systems, such as power grids, by load redistribution when load-bearing equipment has overload capacity in response to equipment failures on parallel distribution lines, and can also be used to prepare electrical equipment to effectively deal with future conditions, for example by providing sufficient cooling in advance of an anticipated overload condition to extend the time that the electrical equipment can be safely overloaded, or by reducing cooling in anticipation of reduced demand. This is useful in large transformers where cooling can be a time-consuming process.

[0030] According to embodiments of the present disclosure, the future load of a transformer can be accurately estimated using a relatively small set of historical load data to train machine learning models and efficiently select the machine learning models with the highest accuracy for different electrical equipment. In some embodiments, the future load can be accurately estimated without other external information such as holidays, sporting events, time of year, etc. One advantage of this approach is that the complexity of the model is significantly reduced, allowing a greater variety of electrical and / or computing equipment to use the model and reducing dependency on advanced computing equipment that may be remotely located and difficult to access.

[0031] Many machine learning applications (e.g., supervised learning problems) operate on flat files or streams of sampled data. In this regard, Figure 1 shows an example of a flat file 100 that is used by a machine learning application to repeatedly apply a multivariate regression algorithm to each row 1-m, using columns x1-xn as inputs and the final column as the target output, to determine and refine the algorithm over time.

[0032] As shown in Figures 2A and 2B, a single variable data set 200 (a time series of load vs. time in this example) can be converted into a multivariate flattened data set 202. In this simplified example, the single variable data set 200 provides 14 consecutive load values ​​204 at times t1-t14, and each row 206 of the flattened data set 202 uses a sequence of five consecutive load values ​​204 as input 208 for modeling the next load value in the sequence, i.e., the target value 210, such that each row 206 of the flattened data set 202 uses a sequence of six consecutive load values ​​204 (e.g., a "moving window"). However, it should be understood that the choice of the number of variables (i.e., predictors) depends on each individual problem. The number of variables in each row 206 can be determined and optimized based on additional testing of, for example, sensitivity, model accuracy, hardware and software constraints, and other parameters.

[0033] To convert the single-variable dataset 200 into a flattened dataset 202, a first row 206 of the flattened dataset is populated with the loading values ​​204 from the single-variable dataset 200 at times t1-t6, a second row 206 is populated with the loading values ​​204 at times t2-t7, and so on for nine rows 206 of the flattened dataset, with the ninth row 206 being populated with the last six loading values ​​204 at times t9-t14. In this way, the vertical single-variable dataset 200 is converted into a tabular multivariable flattened dataset suitable for use in many types of machine learning models.

[0034] One advantage of this data transformation technique is that it transforms a single-variable data set (e.g., loads over time) into a multivariate problem, facilitating the use of many machine learning models suitable for regression or classification applications. The strength of such machine learning models is the fact that they can "learn" from large data sets that contain a large number of cases (or examples) and a large number of features (or predictors, or independent variables). In this example, the machine learning model performs a regression-type prediction to predict a real number as the target value 210, while in other examples, it may perform a classification-type prediction to predict a categorical target such as "good or bad," "yes or no," levels 1, 2, 3, etc.

[0035] One advantage of using these and other machine learning techniques with transformer load data is that these techniques provide very accurate predictions of future load based on a relatively small, single-variable data set of historical load over time, without the need for other external parameters such as temperature, holidays, events, etc.

[0036] In some embodiments, many different machine learning models (e.g., linear and non-linear algorithms) are trained using the flattened data and the results are compared to determine the machine learning model with the highest accuracy. Many different criteria can be used to determine accuracy, such as root mean square error (RMSE), mean absolute error (MAE), etc. Examples of suitable linear machine learning models may include general linear regression, logistic regression (e.g., for classification), linear discriminant analysis, etc. Examples of suitable non-linear machine learning models may include classification and regression trees, naive Bayesian, K-nearest neighbors, support vector machines, etc. Examples of suitable ensemble machine learning models may include random forests, tree bagging, extreme gradient boosting machines, artificial neural networks, etc.

[0037] In one illustrative example, a large dataset of approximately 120,000 load data points was used to train and test different machine learning models. The dataset represented hourly load data collected over several years. In this example, the load data was split into a set of training data (containing 80% of the data) and validation data (containing the remaining 20% ​​of the data). The training data was converted into a flat file with 25 columns (24 input variables and 1 target variable). The training data was then used to train various machine learning models. After training, the validation data was similarly converted into a 25 column flat file, with the target column removed. After predicting the target values ​​for the validation data, the predicted values ​​were compared to the actual target values. The process was then repeated so that the results could incorporate variance in how the machine learning models processed the data, allowing for short-term load predictions over the next 1-2 hours with high accuracy.

[0038] In one example, Fig. 3 illustrates operations 300 for training and selecting a machine learning model for predicting future loads of electrical equipment based on historical load data. See also Fig. 4, which illustrates a visualization of the operations 300 of Fig. 3 for a training and validation dataset 402 of the machine learning model.

[0039] The operations 300 of Figure 3 include splitting a full dataset into training data and visualization data (block 302). For example, in Figure 4, the dataset 402 is separated into training data 404 (e.g., 80%) and validation data 406 (e.g., 20%). The operations further include splitting the training data into K random pieces (block 304). For example, the training data 404 of Figure 4 is split into K equal-sized pieces 408 (e.g., 10 for purposes of this example).

[0040] Next, the machine learning models are trained with the K-1 pieces of training data (block 306) and validated using the remaining pieces of training data (block 308). The accuracy of the models is annotated (block 310) and the process is repeated for multiple different machine learning models (block 312). For example, in FIG. 4, nine of the ten pieces 408 (e.g., pieces 1-9) are used to train each machine learning model, and the last piece 408 (e.g., piece 10) is used to test the accuracy of each trained machine learning model, with each piece 408 of the training data 404 reserved for testing once per machine learning model.

[0041] Next, operations 306-310 are repeated for each of the K-1 sets (block 312) for a total of K training and testing runs for each machine learning model (e.g., 10 times for each machine learning model for each combination of the nine pieces 408 in FIG. 4). Operations 304-314 are then repeated for a different set of K random pieces of training data (block 316) to obtain independent results. For example, operations 306-310 are repeated a total of three times for the training data 404 in FIG. 4, resulting in a total of 30 training runs in this example. Operations 300 further include using validation data (e.g., validation data 406 in FIG. 4) to validate the trained machine learning models and selecting the most accurate machine learning model for use in forecasting future loads.

[0042] These operations 300 in FIG. 3 have the advantage of minimizing "overtraining," which occurs when a given model processes the same data many times and overweights the training data, making its output nearly 100% correct when it faces the training data again. This means that the model is almost "unbiased" with respect to the training set and is considered to be "overtrained." When faced with new unseen data, an overtrained model may perform poorly because it can perfectly represent the training data but cannot represent the new unseen data. In general, it is often more desirable for the trained machine learning to have strong generalization capabilities, even at the expense of perfect accuracy with respect to the training data.

[0043] 5A-5C, graph plots 500A-500C show a comparison of actual load values ​​502A-502C (solid lines) over time and predicted load values ​​504A-504C (dashed lines) over time for an electrical device using a machine learning model trained and selected using operation 300 of FIG. 3. In this example, a data set of actual load values ​​502A-502C from three different 1000 hour time periods was selected from a large validation data set covering thousands of hours of load data for a transformer. Each data set of actual load values ​​502A-502C was used to train a machine learning model to predict predicted load values ​​504A-504C over time for each set of actual load values ​​502A-502C with high accuracy. Notably, each set of predicted load values ​​504A-504C in this example was derived independently from each set of actual load values ​​502A-502C without retraining or incorporating results from other training examples. Despite the wide variation between the different sets of actual load values ​​502A-502C, all three sets of predicted load values ​​504A-504C demonstrated high accuracy. For example, the actual load values ​​502A in FIG. 5A vary between approximately 11,250 MW and 21,250 MW, the actual load values ​​502A in FIG. 5B vary between approximately 10,000 MW and 17,000 MW, and the actual load values ​​502C in FIG. 5C vary between approximately 12,500 MW and 22,500 MW. Despite the wide variance between the different sets of actual load values ​​502A-502C, each predicted load 504A all provide an average prediction error between 1.13% and 1.26%, with 90% of all predicted load values ​​504A-504C providing an error rate of less than 2.55%. As a result, the machine learning models trained and selected using the embodiments disclosed herein, including, for example, operation 300 of FIG. 3, can provide predicted load values ​​504A-504C, particularly short-term predicted loads 506A-506C, with high accuracy.

[0044]

[0033] Referring now to Figure 6, there is shown a flow chart of operations 600 for predicting load parameter values ​​for future times based on past load data. Reference is also made to Figure 7, which is a graphical plot 700 illustrating load over time 702, predicted load parameter values ​​for future times 706, and future overload capacities 710A-710C determined using the operations of Figure 6.

[0045] The operations 600 include predicting (block 602) a load parameter value of an electric equipment at a future time based on at least one machine learning model and a plurality of load parameter values. In this example, the plurality of load parameter values ​​are obtained by sampling values ​​in a continuous moving window on a time series data stream of load parameter values. Thus, the plurality of load parameter values ​​includes a set of a predetermined number of load parameter values ​​(e.g., a predetermined window size, such as a moving window of five consecutive load values ​​204 as described above with respect to FIGS. 2A and 2B) extracted from the time series data stream of load parameter values ​​obtained for the electric equipment. For example, as shown in FIG. 7, one or more machine learning models, e.g., models trained and selected using the operations of FIG. 3 above, are used to predict a predicted load at a current time t0 for a future time t1, e.g., 1-2 hours after the current time t0. The prediction in this example is based on a set 704 of past load values. In some examples, the load parameter values ​​may be electrical and / or thermal load values ​​of the electric equipment. The electric load may be measured by a load current value. Additionally, in a transformer, the secondary current can be measured and the electrical load of the transformer can be derived from the measured secondary current.

[0046] The thermal load can be measured by temperature values ​​corresponding to heat caused by electrical losses in the electrical equipment. In a transformer, temperature values ​​can be obtained from different components, e.g., maximum oil temperature, which, together with the current load and thermal fingerprint of the individual transformer, can be used to estimate future electrical load and / or future associated hot spot temperature, which can indicate the contribution of excess heat to the aging of the transformer. Other parameters, including output power, primary current, etc., can also be used to estimate the operating temperature. Thus, in some embodiments, the operations 600 may further include predicting at least one hot spot temperature value for a component of the electrical equipment at a future time based on the predicted load parameter value (block 604).

[0047] The operations 600 further include calculating an overload capacity for a future time based on the predicted load parameter value (block 606). In some examples, calculating the overload capacity includes predicting at least one overload capacity value for the electric equipment based on the predicted load parameter value for a period following the future time (block 608). For example, as shown in FIG. 7, different overload capacities 710A-710C can be calculated based on the load parameter value and / or other factors, such as ambient temperature, maximum oil level, thermal fingerprint, etc. For example, based on the predicted load parameter value 706 for the future time t1, the transformer can safely operate at more than full capacity 712 (i.e., 100%) for different amounts of time. For example, based on the predicted load parameter value 706, the transformer may have 135% overload capacity 710A for 30 minutes (e.g., until time t2), 120% overload capacity 710B for 1 hour (e.g., until time t3), 120% capacity for 1 hour, or 105% overload capacity 710C for 2 hours (e.g., until time t3). In this manner, overload capacity may be determined as a percentage indicative of the amount of additional load that a particular electrical device can withstand over a particular period of time without abnormally and / or significantly aging or damaging the electrical equipment.

[0048] Unlike some conventional overload capacity calculations where the overload capacity is calculated based on a directly measured load at the present time (e.g., current load 708 at t0 in FIG. 7), these and other embodiments allow future overload capacities (e.g., overload capacities 710A-710C) to be calculated based on highly accurate predicted loads (e.g., predicted load parameter values ​​706 at a future time t1) before the actual load at the future time is known. This in turn provides operators and asset managers more time to plan and respond to contingencies.

[0049] The operations 600 further include modifying at least one parameter associated with the electric equipment at the current time based on the calculated overload capacity for the future time (block 610). The operations 600 may also include operating the electric equipment based at least in part on the at least one modified parameter (block 612). For example, as described above, a cooling component of a transformer may operate in response to the predicted load parameter value to change a temperature of at least one component of the electric equipment prior to the future time. Modifying a parameter and / or operating the electric equipment may also include communicating with and / or providing instructions to the electric equipment.

[0050] In some examples, the at least one machine learning model is trained based on a plurality of determined relationships between a predetermined number of load parameter values ​​and at least one subsequent load parameter value from the time series data stream acquired over a predetermined period of time. In this example, the training is performed prior to deploying the machine learning model and may also be performed continuously during operation.

[0051] In some examples, the plurality of determined relationships are validated based on a comparison of at least one expected load parameter value derived from the predetermined number of load parameter values ​​to at least one subsequent load parameter value. For example, as described above with respect to FIGS. 2A-2B, the prediction of the future time load parameter values ​​may be performed continuously using the predetermined number of load parameter values ​​as an input set to at least one machine learning model. The input set may be continuously generated using a moving window technique from a time series data stream of the load parameter values. For example, the plurality of load parameter values ​​may include a set of at least five load parameter values ​​iteratively extracted from a stream of load parameter values ​​obtained from the electrical equipment. The future time of the predicted load parameter value may also be at least one hour after the prediction.

[0052] In some examples, the at least one machine learning model predicts a load parameter value for the electric equipment at a future time based solely on the multiple load parameter values. For example, as described above, the machine learning model may use a relatively small set of historical load data to accurately predict future loads without additional input. In some examples, the machine learning model may also take into account known changes to the electric equipment, such as performing cooling actions in anticipation of an overload condition. For example, the at least one machine learning model predicts a load parameter value for the electric equipment at a future time based solely on the multiple load parameter values ​​and at least one temperature parameter associated with the electric equipment.

[0053] Thus, the at least one machine learning model is trained with data stream values ​​of load parameter values, which may include actual measured load values, temperature values, etc., for prediction. The at least one machine learning model may be included in a device, such as a controller or relay associated with the electrical equipment, that receives the data stream values ​​from a sensor connected to the electrical equipment and may control or adjust other devices to perform at least one operation related to the electrical equipment, or may be included in a central monitoring system located in a substation or distribution management system used to manage the power system. For example, based on the prediction of future load parameter values ​​and based on the processing performed to determine an overload condition at a future time, the device (e.g., controller / relay) may predict a future overload condition and perform / adjust a cooling operation.

[0054] FIG. 8 is a block diagram of a transformer load forecasting system 800 configured to perform operations disclosed herein, such as, for example, operation 300 of FIG. 3 and / or operation 600 of FIG. 6. In the embodiment of FIG. 8, the transformer monitoring system 30 of the load forecasting system 800 can monitor one or more transformers 10A, 10B. In some embodiments, the transformer monitoring system 30 can be integrated within the transformer 10A provided as a device for monitoring and load forecasting, allowing only the transformer 10A to be monitored, while in other embodiments, the transformer monitoring system 30 can be integrated with the transformer 10A to monitor and optionally receive data from one or more adjacent electrical devices (e.g., the transformer 10B or another transformer or current transformer or circuit breaker) or connected transmission / distribution lines. In yet another embodiment, the transformer monitoring system 30 is separate from the transformers 10A, 10B being monitored.

[0055] The transformer monitoring system 30 includes a processor circuit 34, a communications interface 32 coupled to the processor circuit 34, and a memory 36 coupled to the processor circuit 34. The memory 36 includes machine-readable computer program instructions that, when executed by the processor circuit 34, cause the processor circuit 34 to perform some of the operations illustrated and described herein, such as, for example, operation 600 as described above with respect to FIG.

[0056] As shown, the load forecasting system 800 includes a communication interface 32 (also referred to as a network interface) configured to provide communication with other devices, such as sensors 20 in the transformers 10A, 10B, via wired or wireless communication channels 14. The transformer monitoring system 30 may receive signals from the sensors 20 indicative of physical parameters of the transformers 10A, 10B, such as voltages, currents, oil temperature, ambient temperature, etc. associated with the transformers 10A, 10B. One advantage of some embodiments is that the transformer monitoring system 30 may be a resource-constrained device since less data and / or parameters are required for machine learning algorithms to make accurate load forecasting and predictions.

[0057] In this example, the transformer monitoring system 30 is shown as a separate monitoring device communicating with the transformer 10A, 10B circuits over the communication channel 14, for example, in a server-client model, a cloud-based platform, a substation automation system used in a substation, a distribution management system used in power system management, or other network configuration. One advantage of the client-server configuration is that an overall optimization of the load of the power system and / or substation can be achieved based on load predictions and overload capacity calculations of multiple individual pieces of equipment, such as the transformers 10A, 10B. For example, load management in a power system can include redistributing loads across different electrical equipment based on the predicted load carrying capacity of the different electrical equipment. However, it should also be understood that in other embodiments, the transformer monitoring system 30 may be part of the transformers 10A, 10B or other electrical equipment as desired. It should also be understood that a client-server model may also be operated, where the client is associated with the electrical equipment making measurements, and the server may run machine learning algorithms and calculate overload carrying capacity.

[0058] In another embodiment of the server-client model, a transformer monitoring system may have devices (e.g., clients) associated with the transformers being monitored, the devices comprising machine learning models for load forecasting, and a central system (e.g., server) configured to monitor a plurality of electrical equipment / transformers. The server may also include an instance of the machine learning model included in the device associated with the transformer. The machine learning model in the server may be continuously trained for load forecasting with data received from the transformer or / and a plurality of electrical equipment, and the server may provide information / data (e.g., model coefficients) for tuning / adapting the machine learning model in the device derived from the continuous learning machine learning model in the server. The server may also perform simulations or advanced processing to predict / simulate conditions in the transformer (e.g., hotspot determination based on electrical / thermal load information made available by devices or sensors connected to the transformer) and provide information regarding such determination to devices (e.g., clients) connected to the transformer to modify at least one parameter (e.g., cooling) associated with the transformer (or other electrical equipment) by the device. According to various embodiments, the transformer monitoring system 30 may include electronic, computing, and communications hardware and software for predicting load parameter values ​​and performing at least one activity related to the transformer.

[0059] The transformer monitoring system 30 also includes a processor circuit 34 (also referred to as a processor) and a memory circuit 36 ​​(also referred to as a memory) coupled to the processor circuit 34. According to other embodiments, the processor circuit 34 may be defined to include a memory such that a separate memory circuit is not required.

[0060] As described herein, operations of the transformer monitoring system 30 and other aspects of the load forecasting system 800 may be performed by the processor circuitry 34 and / or the communication interface 32. For example, the processor circuitry 34 may control the communication interface 32 to send communications to one or more other devices via the communication interface 32 and / or receive communications from one or more other devices via the network interface. Additionally, modules may be stored in the memory 36, and these modules may provide instructions for the processor circuitry 34 to perform respective operations (e.g., operations described herein with respect to the exemplary embodiments) when the instructions of the modules are executed by the processor circuitry 34. For example, the modules may be further configured to manage fault detection, generate updated probabilities of different nodes, provide interfaces (e.g., application programming interfaces (APIs)) for managing, configuring and / or modifying the fault tree structure by a customer or other user, etc., as necessary.

[0061] The transformers 10A, 10B, which may be high voltage transformers, for example, include sensors 20 that measure various quantities associated with the transformers 10A, 10B, such as operating load, ambient temperature, moisture and / or oxygen content, and transmit the measurements to the transformer monitoring system 30 via a communication channel 14. The transformers 10A, 10B may also include subsystems, such as an active section 22 coupled to a power line 26 (e.g., an overhead power line), a cooling system 24 (e.g., for a transformer or reactor), which may operate by or in response to instructions from, for example, a processor circuit 34. In some examples, similar monitoring systems may be associated with the power line 26 or other components of the load forecasting system 800. In this and other examples, for ease of explanation, it should be understood that many other types of electrical equipment, such as reactors, transmission lines, equipment transformers, generators, and the like, that are exposed to load conditions may benefit from the embodiments described herein, and all such electrical equipment should also be considered within the scope of the present disclosure.

[0062] These measurements may be used by the transformer monitoring system 30 to detect and / or determine the presence of faults in various components or subsystems of the transformers 10A, 10B and / or general fault conditions of the transformers 10A, 10B. The communication channel 14 may include a wired or wireless link, and in some embodiments may include a wireless local area network (WLAN) or a cellular communication network, such as a 4G or 5G communication network.

[0063] The load prediction system 800 can receive online or offline measurements of operating load, temperature, moisture, oxygen content, etc. from the transformers 10A, 10B and process the measurements to detect and / or determine the presence of faults. The load prediction system 800 can be implemented on a server, a server cluster, a cloud-based remote server system, and / or a standalone device. Sensor data can be obtained by the load prediction system 800 from one transformer and / or multiple transformers.

[0064] The load forecasting system 700 described herein can be implemented in many different ways. For example, the transformer monitoring system 30 according to some embodiments can receive online / offline data, which is used by machine learning techniques configured in the device for learning and classification to identify different patterns that can be considered for estimation / simulation as described in various embodiments. The device may be connectable to one or more transformers 10 to receive measurement data.

[0065] In the above description of various embodiments of the inventive concept, it should be understood that the terms used herein are only for describing specific embodiments and are not intended to limit the inventive concept.Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which the inventive concept belongs.It should be further understood that terms such as those defined in commonly used dictionaries should be interpreted to have a meaning consistent with their meaning in the context of this specification and related art.

[0066] When an element is referred to as being "connected," "coupled," "responsive," or variations thereof to another element, it may be directly connected, coupled, or responsive to the other element, or there may be intervening elements. In contrast, when an element is referred to as being "directly connected," "directly coupled," "directly responsive," or variations thereof to another element, there are no intervening elements. Like numbers refer to like elements throughout. Additionally, as used herein, "coupled," "connected," "responsive," or variations thereof may include wirelessly coupled, connected, or responsive. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise. Well-known features or configurations may not be described in detail for brevity and / or clarity. The term "and / or" includes any and all combinations of one or more of the associated listed items. The phrase "at least one of A and B" means "A or B" or "A and B."

[0067] In this specification, terms such as first, second, third, etc. may be used to describe various elements / operations, but it is understood that these elements / operations should not be limited by these terms. These terms are used only to distinguish one element / operation from another. Thus, a first element / operation in some embodiments can be referred to as a second element / operation in other embodiments without departing from the teachings of the inventive concept. The same reference numbers or the same reference characters refer to the same or similar elements throughout this specification.

[0068] As used in this specification, the terms "comprise," "comprising," "comprises," "include," "including," "includes," "have," "has," "having," or variations thereof are open-ended and refer to the inclusion of one or more stated features, integers, elements, steps, components, or functions but do not exclude the presence or addition of one or more other features, integers, elements, steps, components, functions, or groups thereof.

[0069] Exemplary embodiments are described herein with reference to block diagrams and / or flowchart illustrations of computer-implemented methods, apparatus (systems and / or devices) and / or computer program products. It will be understood that blocks of the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can be implemented by computer program instructions executed by one or more computer circuits. These computer program instructions can be provided to general-purpose computer circuits, special-purpose computer circuits, and / or processor circuits of other programmable data processing circuits to manufacture machines such that the instructions, executed via a processor of a computer and / or other programmable data processing apparatus, transform and control transistors, values ​​stored in memory locations, and other hardware components in such circuits to perform the function / operation specified in the block diagram and / or flowchart block or blocks, thereby creating means (functions) and / or structure for performing the function / operation specified in the block diagram and / or flowchart block or blocks.

[0070] These computer program instructions may also be stored on a tangible computer readable medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, and the instructions stored on the computer readable medium produce an article of manufacture that includes instructions that implement the functions / operations specified in the block diagram and / or flowchart block or blocks. Thus, embodiments of the inventive concepts may be implemented in hardware and / or software (including firmware, resident software, microcode, etc.) executing on a processor, such as a digital signal processor, which may collectively be referred to as a "circuit," "module," or variations thereof.

[0071] It should also be noted that in some alternative implementations, the functions / acts noted in the blocks may occur in a different order than that noted in the flowcharts. For example, two blocks shown in succession may in fact be executed substantially simultaneously, or the blocks may sometimes be executed in the reverse order, depending on the functions / acts involved. Furthermore, the functionality of a given block of the flowcharts and / or block diagrams may be separated into multiple blocks, and / or the functionality of two or more blocks of the flowcharts and / or block diagrams may be at least partially integrated. Finally, other blocks may be added / inserted between the illustrated blocks, and / or blocks / acts may be omitted, without departing from the scope of the inventive concept. Furthermore, while some of the figures include arrows on communication paths to indicate a primary direction of communication, it should be understood that communication may occur in a direction opposite to that of the depicted arrows.

[0072] Many variations and modifications can be made to the embodiments without substantially departing from the principles of the inventive concept. All such variations and modifications are intended to be included within the scope of the inventive concept. Therefore, the subject matter disclosed above should be considered as illustrative rather than limiting, and the example embodiments are intended to cover all such modifications, enhancements, and other embodiments that fall within the spirit and scope of the inventive concept. Therefore, to the maximum extent permitted by law, the scope of the inventive concept should be determined by the broadest permissible interpretation of the present disclosure, including examples of embodiments and their equivalents, and should not be limited or restricted by the foregoing detailed description.

Claims

1. a processor circuit predicting a load parameter value of the electric equipment at a future time based on at least one machine learning model and a plurality of load parameter values ​​including a plurality of sets of a predetermined number of load parameter values ​​extracted by successive moving windows from a time series data stream of load parameter values ​​obtained for the electric equipment; the processor circuit calculating an overload capacity for the future time based on the predicted load parameter value; the processor circuit altering at least one parameter associated with the electrical device at a current time based on the calculated overload capacity at the future time; A method comprising:

2. 2. The method of claim 1 , wherein the at least one machine learning model is trained based on a plurality of determined relationships between a predetermined number of load parameter values ​​and at least one subsequent load parameter value from a time series data stream acquired over a predetermined period of time.

3. 3. The method of claim 2, wherein the plurality of determined relationships are verified based on a comparison of at least one expected load parameter value derived from the predetermined number of load parameter values ​​to the at least one subsequent load parameter value.

4. predicting the load parameter value at the future time occurs continuously using the predetermined number of load parameter values ​​as an input set to the at least one machine learning model; A method according to any one of claims 1 to 3, wherein the input set is generated continuously from the time series data stream of load parameter values ​​using a moving window technique.

5. 5. The method of claim 1, wherein the plurality of load parameter values ​​comprises a set of at least five load parameter values ​​iteratively extracted from a stream of load parameter values ​​obtained from the electrical equipment.

6. A method according to any preceding claim, wherein the future time of the predicted load parameter value is at least one hour after the prediction.

7. The method of any one of claims 1 to 6, wherein the at least one machine learning model predicts the load parameter value of the electrical equipment at the future time based solely on the plurality of load parameter values.

8. 8. The method of claim 1, wherein the at least one machine learning model predicts the load parameter value of the electrical equipment at the future time based on the plurality of load parameter values ​​and at least one temperature parameter associated with the electrical equipment.

9. 9. The method of claim 1, further comprising: the processor circuit predicting at least one hot spot temperature value for a component of the electrical equipment at the future time based on the predicted load parameter value.

10. 10. The method of claim 1, further comprising: the processor circuit predicting at least one overload capacity value for the electrical equipment based on the predicted load parameter value, the at least one overload capacity value being associated with at least one time period following the future time.

11. the electrical equipment comprises a transformer, and the method further comprises: The method of any one of claims 1 to 10, further comprising operating the transformer based at least in part on the at least one changed parameter.

12. 12. The method of claim 11, wherein operating the transformer includes operating at least one cooling component of the transformer in response to the predicted load parameter value to change a temperature of at least one component of the electrical equipment prior to the future time.

13. A processor circuit; When executed by the processor circuit, the processor circuit predicting a load parameter value of the electrical equipment at a future time based on at least one machine learning model and a plurality of load parameter values, the plurality of sets including a predetermined number of load parameter values ​​extracted by successive moving windows from a time series data stream of load parameter values ​​obtained for the electrical equipment; Calculating an overload capacity for the future time based on the predicted load parameter value; changing at least one parameter associated with the electrical device at a current time based on the calculated overload capacity at the future time; a memory containing machine-readable instructions; A monitoring device comprising:

14. 14. The monitoring device of claim 13, wherein the at least one machine learning model is trained based on a plurality of determined relationships between a predetermined number of load parameter values ​​and at least one subsequent load parameter value from a time series data stream acquired over a predetermined period of time.

15. 15. A monitoring device according to claim 13 or 14, wherein the plurality of load parameter values ​​comprises a set of at least five load parameter values ​​iteratively extracted from a stream of load parameter values ​​obtained from the electrical equipment.

16. A monitoring device according to any one of claims 13 to 15, wherein the future time of the predicted load parameter value is at least one hour after the prediction.

17. When executed by the processor circuit, the processor circuit: predicting a load parameter value of the electrical equipment at a future time based on at least one machine learning model and a plurality of load parameter values, the plurality of sets including a predetermined number of load parameter values ​​extracted by successive moving windows from a time series data stream of load parameter values ​​obtained for the electrical equipment; Calculating an overload capacity for the future time based on the predicted load parameter value; changing at least one parameter associated with the electrical device at a current time based on the calculated overload capacity at the future time; A non-transitory computer-readable medium containing instructions.

18. 20. The computer-readable medium of claim 17, wherein the at least one machine learning model is trained based on a plurality of determined relationships between a predetermined number of load parameter values ​​and at least one subsequent load parameter value from a time series data stream acquired over a predetermined period of time.

19. 19. The computer-readable medium of claim 17 or 18, wherein the plurality of load parameter values ​​comprises a set of at least five load parameter values ​​iteratively extracted from a stream of load parameter values ​​obtained from the electrical equipment.

20. The computer readable medium of any one of claims 17 to 19, wherein the future time of the predicted load parameter value is at least one hour after the prediction.

21. At least one machine learning model is trained and selected using an operation (300) based on historical load data, said operation (300) comprising: (302) splitting the historical load data into training data (404) and validation data (406); (304) Dividing the training data (404) into K random pieces (408); (306) training a machine learning model using the K-1 pieces of the training data (404); (308) validating a machine learning model using the remaining pieces of the training data (404); (310) Annotating the accuracy of the model; (312) repeating operations (306) through (310) for a plurality of different machine learning models; (314) repeating operations (306) through (310) for each set of K−1 pieces, performing a total of K training and testing operations for each machine learning model; (316) repeating operations (304) through (314) for sets of K different random pieces of the training data; (318) validating the trained machine learning model using the validation data (406); (320) selecting at least one most accurate machine learning model to use for the prediction.

Images