The present disclosure generally relates to photovoltaic array operation, and in particular, to a system and associated method for maximizing power output of a photovoltaic array by determining a photovoltaic array topology that maximizes power output based on irradiance of the photovoltaic array.
Photovoltaic (PV) energy systems have played a major part in meeting the renewable energy requirements over the past decade. However, power production from PV systems faces impediments such as partial shading due to environmental and man-made obstructions. Shading causes voltage and current mismatch losses that can significantly reduce the power supplied to the grid.
It is with these observations in mind, among others, that various aspects of the present disclosure were conceived and developed.
Corresponding reference characters indicate corresponding elements among the view of the drawings. The headings used in the figures do not limit the scope of the claims.
Various embodiments of a system and associated methods for deep neural network-based topology reconfiguration of photovoltaic arrays for output power maximization are disclosed herein.
Reconfiguring photovoltaic (PV) array connections is a powerful strategy to mitigate the impact of shading in energy production. Conventionally, PV arrays rely on fixed connections or topologies to generate power required by the grid. However. under partial shading, alternate topologies such as series-parallel (SP), bridge-link (BL), honeycomb (HC) or total-cross-tied (TCT) can lead to improved power production motivating the need for a systematic approach to perform reconfiguration based on the extent of shading on the panels. Various studies highlight the importance of reconfiguring PV arrays to tackle mismatch losses.
There have been several efforts towards addressing PV topology reconfiguration. One system in particular used irradiance equalization in TCT arrays where panels are reconfigured to receive similar irradiance across every row. Another system utilized auxiliary unshaded panels to mitigate partial shading. Additionally, some have proposed strategies to select the best topology among SP, BL and TCT by comparing the maximum powers generated using a simulator driven, sequential approach for different shading conditions. However, such approaches are not scalable with the size of the array and can incur significant overheads. More recently, with the advent of machine learning (ML), one work proposed a graph clustering-based reconfiguration strategy to combine different panels and reported performance improvements.
In contrast, the system disclosed herein provides novel results using regularized neural networks with various topologies for larger (e.g., 5×5 or greater) arrays modeled with wiring losses. More specifically, key contributions presented herein include: (i) A regularized deep neural network architecture with dropout and batchnorm for PV topology reconfiguration for topology selection based on observed operating data; (ii) Expanding optimization across additional possible PV topologies by including a honeycomb topology in addition to series-parallel, bridge-link and total cross tied; (iii) Analysis on the merit of topology reconfiguration under modeled wiring losses; and (iv) Extensive simulations on a 5×5 array to demonstrate more realistic results. During validation, the system was found to achieve an average test accuracy of 83% and an F1 score of 0.83. The system presented herein incorporates wiring losses in its model and implements an optimized deep neural network architecture that guides dynamic switching across different topologies, resulting in an average power improvement of approximately 11% when switching from a series-parallel topology to other topologies.
It is appreciated that the illustrated devices and structures may include a plurality of the same component referenced by the same number. It is appreciated that depending on the context, the description may interchangeably refer to an individual component or use a plural form of the given component(s) with the corresponding reference number.
Reconfiguring photovoltaic (PV) array connections among different topologies such as series-parallel, bridge-link, honeycomb or total cross tied is a viable strategy to mitigate impediments in power production caused by partial shading. Conventional approaches rely on by-passing shaded modules in an array by connecting auxiliary unshaded panels through complex control mechanisms or utilizing a simulator-driven approach to predict the best topology. However, these solutions are not scalable and incur significant installation costs and computational overhead motivating the need to develop ‘smart’ and automated methods for topology reconfiguration. To this end, the present disclosure outlines the system implementing a regularized neural network that leverages panel-level sensor data from a PV array to reconfigure the PV array to a topology that maximizes the power output under arbitrary shading conditions. Based on extensive simulations that include modeling of wiring losses in different configurations, power improvement is observed through reconfiguration. The system is scalable and can be easily deployed in any reconfigurable PV array.
The PV array 100 can include one or more sensor(s) 210 associated with one or more panels 102 that measure the operating data 212, which can be communicated over wired or wireless connection to the device 202. The one or more sensor(s) 210 are operable for capturing operating data including an irradiance value resultant of light on the panel 102. The device 202 can be onboard the PV array 100 or can be positioned elsewhere for communication with the PV array 100.
Once the device 202 determines a topology selection 214 based on the operating data 212, the device 202 can communicate the topology selection 214 to the PV array 100 to re-configure the topology of the PV array 100. In some embodiments, the PV array 100 may be operable for re-configuring the linkages 104 based on the topology selection 214. In other embodiments, the device 202 can be operable for generating and sending control signals to the linkages 104 or other hardware of the PV array 100 to enable the PV array 100 to assume a new topology based on the topology selection 214. One or more linkages 104 of the plurality of linkages 104 can be selectively configurable according to a topology configuration of a plurality of topology configurations, including but not limited to: SP configuration, BL configuration, HC configuration, and TCT configuration. Importantly, system 200 outlined herein does not require additional (secondary panels) for reconfiguration. In other words, the system 200 can be used in scenarios where there is support for static reconfiguration.
Importantly, the structure of the neural network model 220 enables the neural network model 220 to be well-regularized and avoid overfitting during the training process. Each hidden layer of the neural network model 220 applies an affine transformation followed by a non-linear activation function (Rectified Linear Unit—ReLU), with application of a dropout policy and batchnorm on the features from the previous layer during training. The dropout policy randomly sets the weights and gradients of p % of the neurons in each hidden to “zero” during every training iteration, thereby regularizing the neural network model 220 and preventing overfitting. Further, batchnorm at the output of each respective hidden layer handles internal covariate shifts between every layer and leads to faster convergence.
Further, the neural network model 220 can be trained to account for wiring losses that may vary based on a topology exhibited by a PV array. The below discussion provides details on generation of the synthetic irradiance data 242 with modeled wiring losses for training the neural network model 220. By training the neural network model 220 on the synthetic irradiance data 242 generated using the simulation module 240 discussed herein, the neural network model 220 can inherently incorporate into its decision-making process tradeoffs associated with wiring losses that result from different topology configurations.
A. Data Generation
Sample panel-level irradiances (e.g., synthetic irradiance data 242) for a 5×5 PV array can be generated using a binary mapping scheme, where “0” is assigned to panels that are unshaded and “1” is assigned to panels that are unshaded.
0→irr˜[α,1000] (1)
1→irr˜[50,α], (2)
where α=586 W/m2 is the threshold chosen for considering whether a panel is shaded or not. The same irradiance realization is used for the unshaded (0) and shaded (1) panels respectively in the array. As shown in
In one implementation, a comprehensive dataset of synthetic irradiance data was generated including 8,000 irradiance instances covering a wide range of partial shading scenarios by sampling these uniform distributions for randomly chosen binary assignments.
PV topology reconfiguration as considered by the system 200 can be posed as a supervised classification problem. A dataset {(xi, yi)}i=1M can be constructed where xiϵn denotes the input (e.g., operating data 212, including irradiance instances measured at each respective panel 102 of
B. Simulation Setup:
Each panel 102 of the PV array 100 can be represented as a by-pass diode in parallel. As such, the PV array simulation model 250 illustrated in
For a first simulated topology configuration, each simulated linkage of the second subset 256 is “deactivated” to configure the PV array simulation model 250 according to the SP topology configuration. For a second simulated topology configuration, each simulated linkage of the second subset 256 is selectively activated or deactivated to configure the PV array simulation model 250 according to the BL topology configuration. For a third simulated topology configuration, each simulated linkage of the second subset 256 is activated or deactivated to configure the PV array simulation model 250 according to the HC topology configuration. For a fourth simulated topology configuration, each simulated linkage of the second subset 256 is activated to configure the PV array simulation model 250 according to the TCT topology configuration.
Wiring losses representing re-configurable links and conditions of the PV array 100 can be modeled at the PV array simulation model 250 using resistors placed between each of the linkages originating or terminating at the panels. Importantly, the wire loss is dependent upon a simulated topology configuration exhibited by the PV array simulation model 250. Specifically, one embodiment of the PV array simulation model 250 used a resistance of R1=0.01Ω to model the wiring losses between the panels connected by linkages across adjacent strings (e.g., for linkages belonging to the first subset 254) and a resistance R2=0.005Ω to model the wiring losses between panels connected by linkages between the panels of a string (e.g., for linkages belonging to the second subset 256). While the embodiments discussed herein are shown with a 5×5 PV array as an example, the system and methods outlined above can be applied to larger or smaller PV arrays of other dimensions.
Generating synthetic irradiance data representing a PV array of any dimension, modeling resultant power generation incorporating wire losses for re-configurable links, and training the neural network model 220 to identify the best topology for a shading condition enables adaptability of the system 200.
C. Design of the Regularized Neural Network
Neural networks (NNs) have produced state-of-the-art performance in a variety of applications including PV array fault detection. As shown in
The example of
In one implementation, the number of neurons for the hidden layers 224A-224F were respectively chosen to be 64, 64, 128, 256, 64 and 64. Each hidden layer 224A-224F of the neural network model 220 applies an affine transformation followed by a non-linear activation function (Rectified Linear Unit —ReLU), with application of a dropout policy and batchnorm on the features from the previous layer. As discussed above, the dropout policy randomly sets the weights and gradients of p % (p=0.2 in one example) of the neurons in a layer (e.g., hidden layers 224A-224F) to zero during every training iteration thereby regularizing the neural network model 220 and preventing overfitting. Application of batchnorm at the output of each respective hidden layer tackles internal covariate shifts between every layer and leads to faster convergence.
The categorical label ŷi is predicted as follows from the output layer 226 including 4 neurons with the softmax activation function applied at the output layer 226:
where zi,j represents the jth logit learned by the output layer 226 for the ith sample and C=4 represents the number of topologies. 90% of the simulated dataset was used for training the neural network model 220 (scaled to have zero mean and unit variance) the remaining 10% was used for testing the neural network model 220. During training, the neural network model 220 can be optimized using categorical cross entropy loss with Adam optimizer for 200 epochs with a learning rate of 10−3.
A. System Performance
B. Merit of PV Topology Reconfiguration
In this subsection, the viability of topology reconfiguration for a wide range of partial shading scenarios is assessed. The assessment determines the number of cases where reconfiguring from one (current) topology to another (switched) topology for that irradiance instance produces a power improvement (Pimp) greater than a specified threshold. For example, by assuming SP to be the ‘current’ topology, the number of cases where the ‘switched’ topology produces maximum power across the four topologies is found. Power improvement is defined as Pimp=Pswitched−Pcurrent where Pswitched and Pcurrent are the MPPs of the switched and current topologies. For quantitative evaluation, considering SP as the current topology, the percentage of cases where Pimp>γ while switching to other topology classes (BL, HC and TCT) is determined. Here, an empirical threshold γ=SOW is considered. Using simulated data, one assessment found that the percentage of cases of switching from SP to BL, HC and TCT were observed at 72.32%, 68.44% and 84.5% respectively. In addition, the average percentage improvement in power when switching from SP to the other topologies was found to be approximately 11%. These illustrate that even under modeled losses optimizing the topology of the solar array can still lead to significant power improvements. As such, the system 200 is scalable and can be effectively utilized to perform the reconfiguration.
Device 500 comprises one or more network interfaces 510 (e.g., wired, wireless, PLC, etc.), at least one processor 520, and a memory 540 interconnected by a system bus 550, as well as a power supply 560 (e.g., battery, plug-in, etc.). The neural network model 220 (
Network interface(s) 510 include the mechanical, electrical, and signaling circuitry for communicating data over the communication links coupled to a communication network. Network interfaces 510 are configured to transmit and/or receive data using a variety of different communication protocols. As illustrated, the box representing network interfaces 510 is shown for simplicity, and it is appreciated that such interfaces may represent different types of network connections such as wireless and wired (physical) connections. Network interfaces 510 are shown separately from power supply 560, however it is appreciated that the interfaces that support PLC protocols may communicate through power supply 560 and/or may be an integral component coupled to power supply 560.
Memory 540 includes a plurality of storage locations that are addressable by processor 520 and network interfaces 510 for storing software programs and data structures associated with the embodiments described herein. In some embodiments, device 500 may have limited memory or no memory (e.g., no memory for storage other than for programs/processes operating on the device and associated caches).
Processor 520 comprises hardware elements or logic adapted to execute the software programs (e.g., instructions) and manipulate data structures 545. An operating system 542, portions of which are typically resident in memory 540 and executed by the processor, functionally organizes device 500 by, inter alia, invoking operations in support of software processes and/or services executing on the device. These software processes and/or services may include photovoltaic topology reconfiguration processes/services 590. Note that while photovoltaic topology reconfiguration processes/services 590 is illustrated in centralized memory 540, alternative embodiments provide for the process to be operated within the network interfaces 510, such as a component of a MAC layer, and/or as part of a distributed computing network environment.
It will be apparent to those skilled in the art that other processor and memory types, including various computer-readable media, may be used to store and execute program instructions pertaining to the techniques described herein. Also, while the description illustrates various processes, it is expressly contemplated that various processes may be embodied as modules or engines configured to operate in accordance with the techniques herein (e.g., according to the functionality of a similar process). In this context, the term module and engine may be interchangeable. In general, the term module or engine refers to model or an organization of interrelated software components/functions. Further, while the photovoltaic topology reconfiguration processes/services 590 is shown as a standalone process, those skilled in the art will appreciate that this process may be executed as a routine or module within other processes.
With reference to
As shown, step 604 can include various sub-steps for implementation of the functionality of step 604. For example, step 604 can include step 606, which includes receiving, at a first input layer of the neural network model, the operating data including the set of irradiance values for the plurality of panels of the photovoltaic array. Step 608 of step 604 can include receiving, at a hidden layer of a plurality of hidden layers of the neural network model, an output of a previous layer of the neural network model. Step 610 of step 604 includes applying, at the hidden layer of the plurality of hidden layers of the neural network model, an affine transformation followed by a non-linear activation function to the output of the previous layer.
While the method 600 shown in
Step 702 of method 700 includes generating or otherwise accessing a set of synthetic irradiance data of the set of labeled training data representing a shading condition. Step 704 of method 700 follows step 702 and includes applying the set of synthetic irradiance data as input to a PV array simulation model for one or more simulated topology configurations to obtain a set of simulated power output values, the PV array simulation model having simulated panels connected by simulated linkages that incorporate wire loss, where the wire loss is dependent upon a simulated topology configuration exhibited by the PV array simulation model 250.
Step 704 can include various sub-steps, including step 706 of method 700 that includes deactivating simulated linkages of a second subset according to an SP topology configuration (e.g., during simulation, to generate a simulated power output value for the PV array simulation model assuming the SP topology configuration that would result from shading conditions represented by the set of synthetic irradiance data). Step 704 can also include step 708 of method 700 that includes deactivating or activating simulated linkages of the second subset according to a BL topology configuration (e.g., during simulation, to generate a simulated power output value for the PV array simulation model assuming the BL topology configuration that would result from shading conditions represented by the set of synthetic irradiance data). Step 704 can further include step 710 of method 700 that includes deactivating or activating simulated linkages of the second subset according to an HC topology configuration (e.g., during simulation, to generate a simulated power output value for the PV array simulation model assuming the HC topology configuration that would result from shading conditions represented by the set of synthetic irradiance data). Step 704 can also include step 712 of method 700 that includes activating simulated linkages of the second subset according to a TCT topology configuration (e.g., during simulation, to generate a simulated power output value for the PV array simulation model assuming the HC topology configuration that would result from shading conditions represented by the set of synthetic irradiance data). While steps 706-712 are directed to generating simulated power outputs for respective SP, BL, HC and TCT topology configurations, additional steps can be included with respect to simulating power outputs for additional topology configurations to adapt the set of labeled training data accordingly. Further, while steps 706-712 discuss activating and deactivating simulated linkages belonging to the second subset (e.g., of
Step 714 of method 700 follows step 704 (and associated sub-steps), and includes identifying, based on the set of simulated power output values, a simulated topology configuration of the one or more simulated topology configurations that produces a maximum simulated power output value of the set of simulated power output values. Step 716 of method 700 follows step 714 and includes assigning the simulated topology configuration that produces the maximum simulated power output value as a label associated with the set of synthetic irradiance data for inclusion within the set of labeled training data.
Step 802 of method 800 includes applying one or more sets of synthetic irradiance data of a set of labeled training data (e.g., after generating the set of labeled training data according to method 700 shown in
It should be understood from the foregoing that, while particular embodiments have been illustrated and described, various modifications can be made thereto without departing from the spirit and scope of the invention as will be apparent to those skilled in the art. Such changes and modifications are within the scope and teachings of this invention as defined in the claims appended hereto.
This is a U.S. Non-Provisional patent application that claims benefit to U.S. Provisional patent application Ser. No. 63/319,514 filed 14 Mar. 2022, which is herein incorporated by reference in its entirety.
This invention was made with government support under 1646542 and 2019068 awarded by the National Science Foundation. The government has certain rights in the invention.
Number | Date | Country | |
---|---|---|---|
63319514 | Mar 2022 | US |