Power grid optimization using pinns

Abstract

The disclosure pertains to a system for optimizing a power distribution grid. The system includes a grid optimizer for real-time capacity assessment of power distribution grid components and a physics-informed neural network (PINNs) designed to model and predict power system dynamics in the power distribution grid. The input data received by the physics-informed neural network (PINNs) comprises power distribution grid component data and training data. The grid optimizer collects real-time data from power components using existing sensors and data acquisition systems within the power distribution grid. The system is also configured to integrate weather data with power component data.

Description

[0001] Power Grid Optimization Using PINNs

[0002] Technical Field

[0003] The technology pertains to the field of power distribution and grid management, specifically focusing on the real-time assessment and optimization of power distribution grid components. It also involves the application of artificial intelligence and machine learning, particularly physics-informed neural networks (PINNs), in modeling and predicting power system dynamics.

[0004] Background

[0005] The power distribution grid is a complex network of interconnected components, including transformers and power lines, that work together to distribute electrical power to con- sumers. The efficient operation of this grid is crucial for ensuring reliable power supply and minimizing energy losses. However, managing the power grid and its components in real-time poses significant challenges.

[0006] One of the key challenges is the real-time capacity assessment of power components. Traditional methods for assessing the capacity of transformers and power lines often rely on static ratings based on conservative assumptions and worst-case scenarios. These methods do not account for real-time conditions, such as weather variations and load fluctuations, which can significantly affect the actual capacity of power components. As a result, these methods often underestimate the available capacity, leading to inefficient utilization of power components and potential overloads. Another challenge is the accurate temperature monitoring of power components. Overheating is a common cause of failures in transformers and power lines, and it can also accelerate the degradation of these components, reducing their lifespan. However, traditional temperature monitoring methods often rely on indirect measurements or simulations, which may not accurately reflect the actual temperature of power components. Inaccurate temperature readings can lead to incorrect capacity assessments and potential overheating, which can cause failures or reduce the lifespan of these components.

[0007] The integration of weather data with power component data is also a challenge. Weather conditions, such as ambient temperature and wind speed, can significantly affect the performance and capacity of power components. However, traditional methods often fail to effectively integrate weather data with power component data, limiting the accuracy of capacity assessments and predictions. Predicting the performance of power components is another challenge. Short-term predictions on the performance and capacity of power components are crucial for better planning and dispatching of power. However, traditional prediction methods often rely on historical data and statistical models, which may not accurately capture the complex dynamics of power systems and the impact of real-time conditions. Furthermore, tracking the loss of life of power components is important for asset management and maintenance planning. However, traditional methods often lack the capability to monitor the degradation and remaining lifespan of power components in real-time, making it difficult to plan for replacements and avoid unexpected failures.

[0008] In addition, managing large-scale electric vehicle chargers, especially in high-demand areas like airports and industrial zones, is essential for balancing power use peaks. However, traditional methods often lack the flexibility and scalability to effectively manage these chargers, leading to inefficiencies and potential power shortages.

[0009] Moreover, providing dual-use applications that support logistics and grid operation during wartime enhances the resilience and flexibility of the power system. However, traditional methods often lack the capability to support these applications, compromising the stability and reliability of the power grid in critical situations.

[0010] Finally, opening up capacity and flexibility in the power grid supports the integration of more renewable energy sources. However, traditional methods often lack the capability to effectively manage the integration of renewable energy, limiting the adoption of renewable energy and hindering efforts to reduce carbon emissions.

[0011] In summary, the prior art in power grid management faces significant challenges in realtime capacity assessment, temperature monitoring, integration of weather data, prediction of power component performance, tracking of component lifespan, management of electric vehicle chargers, support for wartime operations, and integration of renewable energy. These challenges limit the efficiency, reliability, and flexibility of power grid operations, leading to increased energy losses, potential component failures, and reduced adoption of renewable energy. Summary

[0012] According to a first aspect of the disclosure, a system for optimizing a power distribution grid is provided. The system comprises a grid optimizer for real-time capacity assessment of power distribution grid components and a physics-informed neural network (PINNs) configured to model and predict power system dynamics in the power distribution grid. The input data received by the physics-informed neural network (PINNs) comprises power grid component data and training data. The grid optimizer is configured to collect real-time data from power components, utilizing existing sensors and data acquisition systems within the power distribution grid. Additionally, the grid optimizer is configured to integrate weather data with power component data. The advantage of this claim is that it allows for realtime and accurate assessment of the power distribution grid's's capacity, enhancing the efficiency and reliability of power delivery.

[0013] Optionally in some examples, the power distribution grid components comprise powerline components and transformer components. This claim's advantage is that it ensures com- prehensive monitoring and optimization of components within the power distribution grid, leading to improved overall system performance.

[0014] Optionally in some examples, the physics-informed neural network (PINNs) performs a dynamic stress analysis of the power grid components. The advantage of this claim is that it enables the system to predict and mitigate potential failures due to dynamic stresses, thereby enhancing the longevity and reliability of the power grid components.

[0015] Optionally in some examples, the training data of the physics-informed neural network (PINNs) comprises simulated data and real-time data. The advantage of this claim is that it allows the neural network to be trained on a diverse set of scenarios, improving its robustness and accuracy in predicting power system dynamics under various conditions. Optionally in some examples, the simulated data comprises scenarios representing extreme weather conditions, including heatwaves, and corresponding electricity demand surges. The advantage of this claim is that it prepares the system to handle extreme conditions effectively, ensuring reliable power distribution even during adverse weather events and high demand periods. Optionally in some examples, the real-time data comprises weather data and conditions. The advantage of this claim is that it enables the system to make accurate predictions and adjustments based on current and forecasted weather conditions, improving the overall efficiency and reliability of the power distribution grid. Optionally in some examples, receiving real-time data from the power distribution grid components comprises receiving the power grid component data including historical load and / or weather data. The advantage of this claim is that it provides a comprehensive dataset for the neural network, enhancing its ability to make accurate predictions and assessments of the power system's performance.

[0016] Optionally in some examples, the weather data provides information about current and forecasted weather conditions. The advantage of this claim is that it allows the system to incorporate real-time and forecasted weather information into its predictions, leading to more accurate and reliable power grid management. According to a first aspect of the disclosure, a system for optimizing a power distribution grid is provided. The system comprises a grid optimizer for real-time capacity assessment of Power distribution grid components and a physics-informed neural network (PINNs) configured to model and predict power system dynamics in the power distribution grid. The input data received by the physics-informed neural network (PINNs) comprises power grid component data and training data. The grid optimizer is configured to collect real-time data from power components, utilizing existing sensors and data acquisition systems within the power distribution grid. Additionally, the grid optimizer is configured to integrate weather data with power component data. The advantage of this claim is that it allows for realtime and accurate assessment of the power distribution grid's's capacity, enhancing the efficiency and reliability of power delivery.

[0017] Optionally in some examples, the Power distribution grid components comprise powerline components and transformer components. This claim's advantage is that it ensures comprehensive monitoring and optimization of components within the power distribution grid, leading to improved overall system performance. Optionally in some examples, the physics-informed neural network (PINNs) performs a dynamic stress analysis of the Power distribution grid components. The advantage of this claim is that it enables the system to predict and mitigate potential failures due to dynamic stresses, thereby enhancing the longevity and reliability of the Power distribution grid components. Optionally in some examples, the training data of the physics-informed neural network (PINNs) comprises simulated data and real-time data. The advantage of this claim is that it allows the neural network to be trained on a diverse set of scenarios, improving its robustness and accuracy in predicting power system dynamics under various conditions.

[0018] Optionally in some examples, the simulated data comprises scenarios representing ex- treme weather conditions, including heatwaves, and corresponding electricity demand surges. The advantage of this claim is that it prepares the system to handle extreme conditions effectively, ensuring reliable power distribution even during adverse weather events and high demand periods. Optionally in some examples, the real-time data comprises weather data and conditions. The advantage of this claim is that it enables the system to make accurate predictions and adjustments based on current and forecasted weather conditions, improving the overall efficiency and reliability of the power distribution grid.

[0019] Optionally in some examples, receiving real-time data from the Power distribution grid components comprises receiving the power grid component data including historical load and / or weather data. The advantage of this claim is that it provides a comprehensive dataset for the neural network, enhancing its ability to make accurate predictions and assessments of the power system's performance.

[0020] Optionally in some examples, the weather data provides information about current and forecasted weather conditions. The advantage of this claim is that it allows the system to incorporate real-time and forecasted weather information into its predictions, leading to more accurate and reliable power grid management.

[0021] Optionally in some examples, the physics-informed neural network (PINNs) include an input layer. The advantage of this claim is that it ensures the neural network can effectively receive and process the necessary input data for accurate modelling and predictions.

[0022] Optionally in some examples, the input layer receives input data comprising of power grid component data. The advantage of this claim is that it allows the neural network to directly utilize relevant data from the Power distribution grid components, enhancing the accuracy of its predictions. Optionally in some examples, the physics-informed neural network (PINNs) include hidden layers. The advantage of this claim is that it enables the neural network to capture complex relationships and nonlinearities within the data, improving the accuracy of its predictions.

[0023] Optionally in some examples, the hidden layers have multiple layers with Rectified Linear Unit or Tanh activations to capture nonlinearities. The advantage of this claim is that it allows the neural network to effectively model complex and nonlinear relationships within the power system data, leading to more accurate predictions.

[0024] Optionally in some examples, the physics-informed neural network (PINNs) include an output layer. The advantage of this claim is that it ensures the neural network can generate useful predictions based on the processed data from the hidden layers.

[0025] Optionally in some examples, the output layer generates predictions based on processed data from the hidden layers including loss of life and / or maximum capacity. The advantage of this claim is that it provides valuable insights into the condition and capacity of power 5 grid components, aiding in effective grid management and maintenance planning.

[0026] Optionally in some examples, the physics-informed neural network (PINNs) generate predictions. The advantage of this claim is that it allows for real-time and accurate forecasting of power system dynamics, enhancing the overall efficiency and reliability of the power distribution grid.

[0027] .10 Optionally in some examples, the predictions comprise a real-time capacity assessment of Power distribution grid components. The advantage of this claim is that it enables proactive management of the power grid, preventing overloads and ensuring efficient power distribution.

[0028] Optionally in some examples, receiving real-time data from Power distribution grid compels nents comprises gathering temperature readings from transformer and / or powerline using existing temperature sensors. The advantage of this claim is that it leverages existing infrastructure to monitor parameters, reducing the need for additional hardware and associated costs.

[0029] Optionally in some examples, receiving real-time data from Power distribution grid com- 20 ponents comprises collecting load data from transformer and / or powerline using existing current and voltage sensors. The advantage of this claim is that it ensures accurate and timely data collection on power loads, which is for effective grid management and preventing overloads.

[0030] Optionally in some examples, receiving real-time data from Power distribution grid com- 25 ponents comprises acquiring SCADAdata, including grid frequency, voltage levels, and / or switch statuses, from the existing SCADA system. The advantage of this claim is that it integrates comprehensive operational data from the SCADA system, enhancing the overall situational awareness and control of the power distribution grid.

[0031] Optionally in some examples, receiving real-time data from Power distribution grid compose nents comprises obtaining nameplate rating information for transformer and / or powerline from existing databases or documentation. The advantage of this claim is that it provides baseline data for accurate capacity assessments and performance predictions, ensuring the reliability and safety of the power distribution grid. The grid optimization may be performed through a coordinated process where the physics- informed neural network (pinns) generates real-time capacity assessments for individual power distribution grid components, and the grid optimizer uses these assessments to actively manage power flow. In one example, the grid optimizer may be configured to integrate weather data with power component data by incorporating weather variables as input data to the physics-informed neural network (pinns). By way of another example, the physics-informed neural network (pinns) may be configured to generate real-time capacity assessments based on the integrated real-time measurement data and environmental conditions. In yet another example, the grid optimizer may be configured to adjust power flow based on the real-time capacity assessments and proactively divert power away from overloaded components, thereby enhancing the efficiency and reliability of power delivery. In this context, "proactively" may relate to diverting power away from components before their dynamic capacity ratings are exceeded, based on the real-time capacity assessments generated by the physics-informed neural network (pinns). This active optimization process continuously monitors component capacity limits and dynamically redistributes electrical load to prevent overloads while maximizing utilization of available grid capacity. Hence, it may be said that each of these examples may provide real-time and accurate assessment of the power distribution grid's capacity, enhancing the efficiency and reliability of power delivery. Brief Description of the Figures

[0032] Examples are described in more detail below with reference to the appended drawings.

[0033] Figure 1 is a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINN) with input data, simulated data, and predictions.

[0034] Figure 2 is a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINN) with input data, real-time data, and predictions.

[0035] Figure 3 is a flowchart illustrating the process of a physics-informed neural network (PINN) with hidden layers.

[0036] Figure 4 is a flowchart illustrating the process of data handling and model deployment with predictions and edge devices. Figure 5 is a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINN) with training data. P4609WC00

[0037] Figure 6 is a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINN) with predictions.

[0038] Detailed Description of the Figures

[0039] The detailed description set forth below provides information and examples of the diss' closed technology with sufficient detail to enable those skilled in the art to practice the disclosure.

[0040] Figure 1 shows a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINNs). The flowchart begins with three input data sources: 'Real-World Data', 'simulated data', and 'KPIs'. These three data sources0 converge into a single process labeled 'Pre-processing'. From 'Pre-processing', an arrow points to the next process labeled 'physics-informed neural network (PINNs)'. An arrow from 'physics-informed neural network (PINNs)' leads to the next step labeled 'Training and Optimization'. Following 'Training and Optimization', an arrow points to a process labeled 'Validated Model'. Finally, an arrow from 'Validated Model' leads to the last step5 in the flowchart labeled 'predictions'. This figure shows the following Node(s): input data, simulated data, predictions.

[0041] Figure 2 shows a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINNs). The flowchart begins with three input data sources: 'Historical Logs', 'Data Sources', and 'Simulations'. 'Historical Logs' and0 'Data Sources' provide 'Load, Weather' data to the 'real-time data' process. 'Simulations' provide 'Extreme Scenarios' data to the 'real-time data' process. An arrow from 'real-time data' leads to the next process labeled 'Pre-processed Data'. An arrow from 'Preprocessed Data' points to the next step labeled 'PINN Training'. An arrow from 'PINN Training' leads to the next step labeled 'predictions'. An arrow from 'predictions' provides 'Feedback' back to5 the 'real-time data' process. This figure shows the following Node(s): input data, real-time data, predictions.

[0042] Figure 3 shows a flowchart illustrating the process of a physics-informed neural network (PINNs). The flowchart begins with a box labeled 'Inputs'. An arrow points from 'Inputs' to the next box labeled 'hidden layers'. An arrow points from 'hidden layers' to the next0 box labeled 'hidden layers Nth'. An arrow points from 'hidden layers Nth' to the next box labeled 'Physics Constraints Applied'. An arrow points from 'Physics Constraints Applied' to the next box labeled 'hidden layers N + 1'. An arrow points from 'hidden layers N + P4609WC00

[0043] 1' to the next box labeled 'Outputs'. From 'Outputs', three arrows point to three separate boxes. The first box is labeled 'Predicted Max Capacity', the second box is labeled 'Loss of Life', and the third box is labeled 'Hot-Spot Temperature'. This figure shows the following Node(s): hidden layers. Figure 4 shows a flowchart illustrating the process of data handling and model deployment. The flowchart begins with a process labeled 'Field Data - SCADA'. An arrow from 'Field Data - SCADA' points to the next process labeled 'edge devices'. An arrow from 'edge devices' leads to the next step labeled 'Real-Time Pre-processing'. An arrow from 'Real- Time Pre-processing' points to the next step labeled 'Local predictions - Edge'. An arrow from 'Local predictions - Edge' leads to the next step labeled 'Cloud Server'. An arrow from 'Cloud Server' points to the next step labeled 'Centralized Training and Updates'. Finally, an arrow from 'Centralized Training and Updates' leads to the last step in the flowchart labeled 'Updated Model Deployed'. This figure shows the following Node(s): predictions, edge devices. Figure 5 shows a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINNs). The flowchart begins with a process labeled 'training data'. From 'training data', two arrows branch out. One arrow, labeled '80%', points to a process labeled 'Real-World Data'. Another arrow, labeled '20%', points to a process labeled 'simulated data'. An arrow from 'Real-World Data' leads to the next process labeled 'Normal Operations'. An arrow from 'simulated data' leads to the next process labeled 'Extreme Conditions'. Arrows from both 'Normal Operations' and 'Extreme Conditions' converge into a single process labeled 'Combined Dataset'. Finally, an arrow from 'Combined Dataset' leads to the last step in the flowchart labeled 'PINN Training'. This figure shows the following Node(s): training data. Figure 6 shows a flowchart illustrating the process of data handling and model training for a physics-informed neural network (PINNs). The flowchart begins with a process labeled 'SCADA Input'. An arrow from 'SCADA Input' points to the next process labeled 'Data Pre-processing'. An arrow from 'Data Pre-processing' leads to the next step labeled 'Real-Time predictions'. From 'Real-Time predictions', an arrow points to a process labeled 'Operational Decisions'. An arrow from 'Operational Decisions' leads to the next step labeled 'Feedback to SCADA'. Additionally, from 'Real-Time predictions', an arrow points to a process labeled 'Updates to Training'. An arrow from 'Updates to Training' leads to the next step labeled 'Improved Model'. Finally, an arrow from 'Improved Model' loops back to 'Real-Time predictions'. This figure shows the following Node(s): predictions. Features Description Text

[0044] 1 Power Distribution Grid Details

[0045] The power distribution grid serves the function of distributing electrical power to consumers through a network of transmission and distribution lines. The grid includes various Power distribution grid components. In some instances, these components may include transformer, which transform voltage levels within the grid. The components may also include powerliness designed to transmit electrical power over long distances.

[0046] The efficient operation of the power distribution grid depends on the coordinated functioning of its components. Transformer play a role in adjusting voltage levels to match the requirements of different parts of the grid, ensuring efficient power delivery. Powerline facilitate the transmission of electricity across geographical areas, connecting power generation sources to end-users.

[0047] 2 Grid Optimizer Details

[0048] The grid optimizer uses physics-informed neural network (PINNs). In some instances, the grid optimizer may also utilize hardware and software components.

[0049] 2.1 Physics-Informed Neural Network (PINNs)

[0050] The grid optimizer uses physics-informed neural network (PINNs). PINNs model and predict power system dynamics within the power distribution grid, helping to plan power dispatch, distribution, and transmission based on real-time conditions using data such as grid frequency, voltage levels, and weather forecasts. This approach helps manage the complex interactions and variables found in large-scale power systems, including voltage fluctuations, frequency variations, and load imbalances.

[0051] PINNs also address heat and mass transfer issues to maintain the efficiency and longevity of Power distribution grid components . They perform dynamic stress analysis of materials and components, optimizing power delivery efficiency by minimizing transmission losses and preventing failures like transformer overloads and line outages. For system identification and improved decision-making, PINNs use real-time data on grid status, predictive analytics for potential issues, and optimized control strategies, leading to better decisions regarding power dispatch, load balancing, and grid maintenance. PINNs can detect charged particles in electromagnetic fields to identify insulation degra- dation in transformerss or powerliness, improving the monitoring and control of the Power distribution grid components. PINNs enable accurate real-time temperature estimation of Power distribution grid components by incorporating heat transfer equations and boundary conditions directly into the neural network, mitigating the impact of inaccurate or sparse sensor readings. They also seamlessly integrate weather data with power component data by incorporating weather variables (temperature, wind speed, solar radiation) as inputs and using physical equations to model their influence on component temperature and capacity. The PINN includes input data. In some configurations, the PINN may include an input layer to receive input data, hidden layers with multiple layers and Rectified Lin- ear Unit (ReLU) or Tanh activations to capture nonlinearities, an output layer to generate predictions like loss of life and maximum capacity, and predictions.

[0052] 2.1.1 Input Data

[0053] Physics-informed neural network (PINNs) utilize input data. For instance, input data may include historical load, which provides historical data on power consumption patterns. In- put data may also include weather data to provide information about current and forecasted weather conditions, improving the accuracy of predictions and assessments of power system performance. Input data includes training data and power grid component data. Power grid component data provides real-time and historical information about the status and performance of Power distribution grid components. Training Data

[0054] Training data is derived from input data and serves to train machine learning models. It facilitates pre-processing techniques to manage noisy operational data, ensuring the creation of high-quality training sets. The training data may include simulated data, which can represent extreme weather conditions such as heatwaves and demand surges. In some instances, simulated data may constitute 20% of the training data set. Training data incorporates real-time data, which may use real-world data for typical weather conditions. Real-time data may represent 80% of the training data set. Real-world data may serve as an example of real-time data.

[0055] 2.1.2 Predictions Physics-informed neural network (PINNs) may use predictions. Predictions perform temperature monitoring of the Power distribution grid components to prevent overheating and potential failures by automatically adjusting cooling systems or sending alerts to operators when temperature thresholds are exceeded. Predictions also address performance, P4609WC00 including real-time capacity assessment, of Power distribution grid components. This realtime capacity assessment enables efficient power distribution grid management and prevents overloads by adjusting power flow based on real-time capacity assessments and proactively diverting power away from overloaded components. Predicting the Power diss' tribution grid components performance aids in better planning and dispatching of power.

[0056] Predictions also track loss of life of Power distribution grid components, enabling predictive maintenance by identifying components nearing end-of-life, optimizing maintenance schedules, and reducing costs. This tracking also helps in asset management and maintenance planning. Real-time data from predictions is sent back as training data for the0 physics-informed neural network (PIN Ns).

[0057] Predictions offer several advantages. They enable real-time capacity assessment by providing accurate estimates of the current and available capacity of Power distribution grid components, addressing the limitations of traditional methods. They enable accurate temperature monitoring by predicting potential overheating risks based on real-time data and5 weather forecasts, mitigating the impact of inaccurate temperature readings. Finally, predictions enable effective integration of weather data by incorporating weather forecasts into capacity predictions, addressing the issue of insufficient integration of weather data with power grid component data.

[0058] 2.2 Hardware 0 In some instances, the hardware may include edge devices that perform real-time monitoring and analysis locally, ensuring continued operation even in areas with limited connectivity. The hardware may also include servers and databases.

[0059] 2.2.1 Servers and Databases

[0060] Servers and databases store real-time data and historical records, supporting large-scale5 deployments. Cloud platforms may be a component of servers and databases.

[0061] Cloud Platforms

[0062] Cloud platforms may incorporate servers and databases to perform computational tasks, data storage, and scaling, which supports large grids efficiently. Cloud platforms may also include web-based dashboards to provide a user interface, enabling access from0 any device with an internet connection. 2.3 Software Requirements

[0063] The grid optimizer may have software requirements. In some instances, the software requirements may include integration with existing systems, such as grid management tools, to reduce deployment complexity by leveraging existing communication protocols and data formats. For example, integration with SCADA systems, which supervise and control power system operations in real-time, may be a component of the software requirements.

[0064] 3 Method Details

[0065] The method for optimizing a power distribution grid involves a systematic process of data collection, processing, and predictions generation. This process begins with the collection of real-time data from various Power distribution grid components, utilizing existing sensors and data acquisition systems within the grid. The collected data is then processed and used to train a physics-informed neural network (PINNs). The trained PINN is then used to generate predictions about the power distribution grid's's behavior, including real- time capacity assessments and predictions of potential failures. These predictions can then be used to make informed decisions about grid operation and maintenance.

[0066] The method leverages the capabilities of PINNs to model and predict power system dynamics accurately. By incorporating physical laws and constraints into the neural network, the PINN can capture the complex interactions and variables within the power distribution grid. This approach enables more accurate and reliable predictions compared to traditional methods.

[0067] The method also emphasizes the integration of weather data with power grid component data. By incorporating weather variables as inputs to the PINN, the method can account for the impact of weather conditions on grid performance. This integration enhances the accuracy of predictions and enables more effective grid management strategies.

[0068] 3.1 Data Collection Process

[0069] The data collection process involves gathering real-time data from various Power distribution grid components, such as transformers and powerlines. This data includes temperature readings, load data, and SCADA data. The process utilizes existing sensors and data acquisition systems within the power distribution grid, minimizing the need for additional hardware.

[0070] The collected data is then pre-processed to ensure its quality and consistency. This pre- processing may involve filtering noise, handling missing values, and converting data formats. The pre-processed data is then used as input to the physics-informed neural network (PINNs) for training and predictions generation.

[0071] 3.1.1 Utilization of Existing Sensors and Data Acquisition Systems The method utilizes existing sensors and data acquisition systems within the power distribution grid to collect real-time data. This approach minimizes the need for additional hardware and reduces deployment costs. The existing sensors provide data on various parameters, such as temperature, current, voltage, and grid frequency.

[0072] The data acquisition systems gather data from these sensors and transmit it to a central processing unit. The system is designed to be compatible with existing communication protocols and data formats, ensuring seamless integration with the existing grid infrastructure.

[0073] 3.1.2 Gathering Temperature Readings from Transformers and Power Lines

[0074] Temperature readings from transform er(s) and powerline are collected using existing tem- perature sensors. These sensors are placed at locations within the grid to monitor the temperature of the Power distribution grid components. The collected temperature data is used to assess the thermal condition of the components and predict potential overheating risks.

[0075] The temperature data is also used as input to the PINN, which incorporates heat trans- fer equations and boundary conditions to model the thermal behavior of the components.

[0076] This physics-informed approach enables more accurate temperature estimations and predictions) compared to traditional methods.

[0077] 3.1.3 Collection of Load Data from Transformers and Power Lines

[0078] Load data, representing the amount of power being consumed by different parts of the power distribution grid, is collected from transformers and powerlines using existing current and voltage sensors. This data provides insights into power consumption patterns and helps identify areas of high demand.

[0079] The load data is used by the PINN to model and predict power system dynamics. By incorporating load data into its calculations, the PINN can accurately predict the impact of changing demand on grid performance. This information is for making informed decisions about power dispatch and grid management. 3.1.4 Acquisition of SCADA Data

[0080] SCADA (Supervisory Control and Data Acquisition) data, which provides real-time information about the grid's operating conditions, is acquired from the existing SCADA system. This data includes grid frequency, voltage levels, and switch statuses. The SCADA data is used by the PINN to monitor the grid's overall health and stability.

[0081] By incorporating SCADA data into its predictions, the PINN can identify potential issues and trigger appropriate control actions. This real-time monitoring and control capability enhances the grid's reliability and resilience.

[0082] 3.2 Data Processing and Model Training The data processing stage involves preparing the collected data for use in training the PINN. This includes cleaning and transforming the data, handling missing values, and potentially generating synthetic data to augment the training dataset. Feature engineering techniques may be applied to extract relevant features from the raw data.

[0083] The model training process involves feeding the processed data to the PINN. The PINN is trained using a combination of real-time data and simulated data, which includes scenarios representing extreme weather conditions and corresponding electricity demand surges. The training process optimizes the PINN's parameters to minimize the difference between its prediction(s) and the actual grid behavior. The trained PINN is then validated using a separate dataset to ensure its accuracy and reliability. 3.2.1 Use of Physics-Informed Neural Network (PINNs)

[0084] Physics-informed neural network(s) (PINNs) are employed to model and predict the behavior of the power distribution grid. PINNs incorporate physical laws and constraints, such as heat transfer equations and boundary conditions, directly into the neural network architecture. This integration allows the PINN to capture the complex interactions and vari- ables within the power distribution grid, leading to more accurate and reliable prediction(s) compared to traditional methods.

[0085] The use of PINNs enables the system to estimate the temperature of Power distribution grid components in real-time, even with inaccurate or sparse sensor readings. By incorporating weather variables, such as temperature, wind speed, and solar radiation, as inputs to the PINN, the system can seamlessly integrate weather data with power component data. This integration allows the PINN to model the influence of weather conditions on component temperature and capacity, further enhancing prediction(s) accuracy. PINNs are also used for dynamic stress analysis of materials and components, optimizing power delivery efficiency by minimizing transmission losses and preventing failures like transformer overloads and line outages. They contribute to system identification for improved decision-making by using real-time data on grid status, predictive analytics for potential issues, and optimized control strategies. This leads to better decisions regarding power dispatch, load balancing, and grid maintenance.

[0086] 3.2.2 Training and Optimization of Models

[0087] The training and optimization of the PINN models involve a multi-step process. Initially, real-time data is collected from various Power distribution grid components, including tem- perature readings from transformer and powerline load data, and SCADA data. This data is then pre-processed to ensure its quality and consistency, which may involve filtering noise, handling missing values, and converting data formats.

[0088] The pre-processed data is then used to train the PINN. The training process involves feeding the PINN with both real-time data and simulated data, which includes scenarios representing extreme weather conditions and corresponding electricity demand surges.

[0089] This combination of real-world and simulated data allows the PINN to learn the behavior of the power distribution grid under both normal and extreme operating conditions.

[0090] During training, the PINN's parameters are optimized to minimize the difference between its predictions and the actual grid behavior. This optimization process typically involves using optimization algorithms, such as gradient descent, to adjust the PINN's weights and biases. The trained PINN is then validated using a separate dataset to ensure its accuracy and reliability before being deployed for real-time predictions.

[0091] 3.2.3 Integration of Weather Data with Power Component Data

[0092] Weather data plays a role in accurately predicting the behavior of the power distribution grid. The system integrates weather data with power component data by incorporating weather variables, such as temperature, wind speed, and solar radiation, as inputs to the PINN. Physical equations are used to model the influence of these weather variables on component temperature and capacity.

[0093] This integration allows the system to account for the impact of weather conditions on grid performance, leading to more accurate predictions and assessments. For example, the system can predict how a heatwave might affect the capacity of a transformer or how strong winds might impact the stability of powerliness. This information can then be used to make informed decisions about grid operation and maintenance, such as adjusting power flow or scheduling preventative maintenance.

[0094] The integration of weather data is achieved through the use of existing weather data sources and forecasting models. The system can access real-time weather data from weather stations and online services, as well as use weather forecasts to predict future grid conditions. This allows the system to proactively adjust grid operations in anticipation of changing weather patterns.

[0095] 3.3 Prediction Generation

[0096] Once the PINN is trained and validated, it can be used to generate predictions about the power distribution grid's's behavior. These predictions include real-time capacity as- sessments of Power distribution grid components, predictions of potential failures, and estimations of the remaining lifespan of components.

[0097] The generated predictions are used to inform operational decisions, such as adjusting power flow, scheduling maintenance, and implementing preventative measures. The predictions are also fed back into the system to further refine the PINN's training and improve its accuracy over time.

[0098] 3.3.1 Real-Time Capacity Assessment

[0099] The system performs real-time capacity assessment of Power distribution grid components by using the trained PINN to predict the current and available capacity of each component. This assessment is based on real-time data from the grid, including temperature readings, load data, and SCADA data, as well as integrated weather data.

[0100] The real-time capacity assessment enables efficient power distribution grid management and prevents overloads by providing accurate estimates of the current operating capacity of each component. This information can be used to adjust power flow based on real-time capacity assessments and prevent overloads by proactively diverting power away from overloaded components.

[0101] By continuously monitoring the capacity of each component, the system can identify potential bottlenecks and take preventative measures to avoid disruptions. This real-time capacity assessment addresses the limitations of traditional methods, which often rely on static capacity ratings and may not accurately reflect the actual operating conditions of the grid. 3.3.2 Tracking Loss of Life of Power Components

[0102] The system tracks the loss of life of Power distribution grid components by using the PINN to predict the remaining lifespan of components. This prediction is based on various factors, including historical operating data, real-time temperature readings, and stress analysis performed by the PINN.

[0103] By tracking the loss of life of each component, the system enables predictive maintenance by identifying components nearing the end of their lifespan. This information can be used to optimize maintenance schedules, reducing costs and minimizing downtime. Predictive maintenance allows for proactive replacement of components before they fail, preventing unexpected outages and ensuring the reliability of the power distribution grid.

[0104] The tracking of loss of life also helps in asset management and maintenance planning. By knowing the expected lifespan of each component, grid operators can make informed decisions about future investments and upgrades. This information can be used to prioritize maintenance activities and allocate resources effectively. Item 1. A system for optimizing a power distribution grid, the system comprising: a grid optimizer for real-time capacity assessment of power distribution grid components; a physics-informed neural network (pinns) configured to model and predict power system dynamics in the power distribution grid; and wherein an input data received by the physics-informed neural network (pinns) comprises a power grid component data and a training data. wherein the grid optimizer is configured to collect real-time data from power components; and wherein the receiving real-time data from power distribution grid components uses existing sensors and data acquisition systems within the power distribution grid; wherein the grid optimizer is configured to integrate weather data with power component data.

[0105] Item 2. The system of item 1, wherein the power distribution grid components includes powerline components and transformer components.

[0106] Item 3. The system of any of items 1 to 2, wherein the physics-informed neural network (pinns) performs a dynamic stress analysis of the power grid components.

[0107] Item 4. The system of any of items 1 to 3, wherein the training data of the physics- informed neural network (pinns) includes simulated data and real-time data.

[0108] Item 5. The system of item 4, wherein the simulated data comprises scenarios representing extreme weather conditions, including heatwaves, and corresponding electricity demand surges. Item 6. The system of any of items 1 to 5, wherein the real-time data comprises weather data and conditions.

[0109] Item 7. The system of any of items 1 to 6, wherein receiving real-time data from the power distribution grid components includes receiving the power grid component data including historical load and / or weather data. Item 8. The system of any of items 1 to 7, wherein the weather data provides information about current and forecasted weather conditions.

[0110] Item 9. The system of any of items 1 to 8, wherein the physics-informed neural network (pinns) include an input layer.

[0111] Item 10. The system of item 9, wherein the input layer receives input data comprising of power grid component data.

[0112] Item 11. The system of any of items 1 to 10, wherein the physics-informed neural network (pinns) include hidden layers.

[0113] Item 12. The system of item 11, wherein the hidden layers have multiple layers with Rectified Linear Unit or Tanh activations to capture nonlinearities. Item 13. The system of any of items 1 to 12, wherein the physics-informed neural network (pinns) include an output layer.

[0114] Item 14. The system of item 13, wherein the output layer generates predictions based on processed data from the hidden layers including loss of life and / or maximum capacity. Item 15. The system of any of items 1 to 14, wherein the physics-informed neural network (pinns) generate predictions.

[0115] Item 16. The system of item 15, wherein the predictions comprise a real-time capacity assessment of power distribution grid components.

[0116] Item 17. The system of any of items 1 to 16, wherein the receiving real-time data from power distribution grid components includes gathering temperature readings from transformer and / or powerline using existing temperature sensors.

[0117] Item 18. The system of any of items 1 to 17, wherein the receiving real-time data from power distribution grid components includes collecting load data from transformer and / or powerline using existing current and voltage sensors. Item 19. The system of any of items 1 to 18, wherein the receiving real-time data from power distribution grid components includes acquiring Scada data, including grid frequency, voltage levels, and / or switch statuses, from the existing Scada system.

[0118] Item 20. The system of any of items 1 to 19, wherein the receiving real-time data from power distribution grid components includes obtaining nameplate rating information for transformer and / or powerline from existing databases or documentation.

[0119] The terminology used herein is for the purpose of describing particular aspects only and is not intended to be limiting of the disclosure. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items. It will be further understood that the terms "comprises," "comprising," "includes," and / or "including" when used herein specify the presence of stated features, integers, actions, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, actions, steps, operations, elements, components, and / or groups thereof.

[0120] It will be understood that, although the terms first, second, etc., may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element without departing from the scope of the present disclosure.

[0121] Relative terms such as "below" or "above" or "upper" or "lower" or "horizontal" or "vertical" may be used herein to describe a relationship of one element to another element as illustrated in the Figures. It will be understood that these terms and those discussed above are intended to encompass different orientations of the device in addition to the orientation depicted in the Figures. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or intervening elements may be present. In contrast, when an element is referred to as being "directly connected" or "directly coupled" to another element, there are no intervening elements present. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. It will be further understood that terms used herein should be interpreted as having a meaning consistent with their meaning in the context of this specification and the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0122] It is to be understood that the present disclosure is not limited to the aspects described above and illustrated in the drawings; rather, the skilled person will recognize that many changes and modifications may be made within the scope of the present disclosure and appended claims. In the drawings and specification, there have been disclosed aspects for purposes of illustration only and not for purposes of limitation, the scope of the disclosure being set forth in the following claims.

Claims

1. Claims1. A system for optimizing a power distribution grid, the system comprising: a grid optimizer for real-time capacity assessment of Power distribution grid components; a physics-informed neural network (PIN Ns) configured to model and predict power system dynamics in the power distribution grid; and wherein an input data received by the physics-informed neural network (PIN Ns) comprises a power grid component data and a training data. wherein the grid optimizer is configured to collect real-time data from power compo- nents; and wherein the receiving real-time data from Power distribution grid components uses existing sensors and data acquisition systems within the power distribution grid; wherein the grid optimizer is configured to integrate weather data with power component data.

2. The system of claim 1, wherein the Power distribution grid components includes powerline components and transformer components.

3. The system of any of claims 1 to 2, wherein the physics-informed neural network (PINNs) performs a dynamic stress analysis of the power grid components.

4. The system of any of claims 1 to 3, wherein the training data of the physics-informed neural network (PINNs) includes simulated data and real-time data.

5. The system of claim 4, wherein the simulated data comprises scenarios representing extreme weather conditions, including heatwaves, and corresponding electricity demand surges.

6. The system of any of claims 1 to 5, wherein the real-time data comprises weather data and conditions.

7. The system of any of claims 1 to 6, wherein receiving real-time data from the Power distribution grid components includes receiving the power grid component data including historical load and / or weather data.

8. The system of any of claims 1 to 7, wherein the weather data provides information about current and forecasted weather conditions.

9. The system of any of claims 1 to 8, wherein the physics-informed neural network (PINNs) include an input layer.

10. The system of claim 9, wherein the input layer receives input data comprising of power grid component data.

Images