A Neural Network-Based Low-Cost Solar Irradiance Sensor

Abstract

Measuring solar irradiance allows for direct maximization\ud of the efficiency in photovoltaic power plants. However,\ud devices for solar irradiance sensing, such as pyranometers and\ud pyrheliometers, are expensive and difficult to calibrate and thus\ud seldom utilized in photovoltaic power plants. Indirect methods\ud are instead implemented in order to maximize efficiency.\ud This paper proposes a novel approach for solar irradiance\ud measurement based on neural networks, which may, in turn,\ud be used to maximize efficiency directly. An initial estimate\ud suggests the cost of the sensor proposed herein may be price\ud competitive with other inexpensive solutions available in the\ud market, making the device a good candidate for large deployment\ud in photovoltaic power plants. The proposed sensor is implemented\ud through a photovoltaic cell, a temperature sensor, and a low–\ud cost microcontroller. The use of a microcontroller allows for\ud easy calibration, updates, and enhancement by simply adding\ud code libraries. Furthermore, it can be interfaced via standard\ud communication means with other control devices; integrated into\ud control schemes; and remote–controlled through its embedded\ud web server. The proposed approach is validated through experimental\ud prototyping and compared against a commercial device

Two FPGA-Oriented High-Speed Irradiance Virtual Sensors for Photovoltaic Plants

“…The solution proposed in [16] (sketched in Fig. 4) consists of a low-cost PV element (composed by a series of few solar cells), a temperature sensor and a known testing resistor R test , suitably chosen according to the PV cell I-V characteristic.…”

“…The first one (which relies on a NN) has been proposed and successfully applied in [16], but a FPGA implementation was not available. The second one is a novel approach based on PWAS functions.…”

“…This would guarantee a higher accuracy since the calibration procedure can take into account also the effects of parasitic components, inaccurate datasheets information, the dependence on temperature T of the whole measuring circuit, variability in the components (e.g., R test ), etc. Here, the first strategy (successfully applied in [16] with very accurate results) is used, with the resistance R test equal to 921 Ω. The five parameters used for modeling the adopted solar cell (IXYS KXOB22-01X8, rated at 3.4 V/3.8 mA at MPP under standard testing conditions) and obtained by using datasheet information, are the following: A test set must also be created to validate the performances of the sensor on different data sets with respect to those used for the training process.…”

“…By following the theoretical approach described in [28], starting from initial guessed values for n h and n train , the iterative process yielded n h = 8 and n train = 50 (N = 10 and M = 5). The reader is referred to [16] for a further deepening on the set up of the NN for this problem and [28] for the detailed explanation of the theory.…”

“…The neural network approach is widely adopted for virtual sensors for PV applications [16]- [20]. In our case the chosen architecture implements a recently proposed lowcost and extremely accurate virtual solar sensor based on a neural network (NN) algorithm [16], appositely tailored for the estimation of the irradiance from the knowledge of voltage, current and temperature of a PV cell. The second architecture implements a novel virtual irradiance sensor, based on piecewise-affine functions defined over uniform simplicial partitions (PWAS functions).…”