UC Davis College of Engineering researchers use automatic learning to discover new materials for high -efficiency solar cells. They perform complex experiences and apply various algorithms depending on automatic learning. Following studies, they found possible to predict the dynamic behavior of very high precision materials without the need for a large number of tests.
The study was published in the ACS energy letters in April.
The object of the research of scientists is the organic-inorganic hybrid perovskites (Hoip). Solar cells based on organic-inorganic hybrid perovskites are a rapid development area of alternative energy. These molecules have initiated the development of a new class of photovoltaic devices – the solar cells of Perovskite. Their first prototypes were created in 2009.
Perovskites are comparable to silicon efficiency for the manufacture of solar cells, but they are lighter and cheaper to produce, which means that they have the potential to be used in a wide variety of applications, including electrumintescent devices.
However, there is an unresolved problem with devices based on perovskite. The problem is that they tend to decompose more quickly than silicon when exposed to humidity, oxygen, light, heat and stress.
The challenge for scientists is to find such perovskites that would combine high efficiency with resistance to environmental conditions. Using only testing and error methods, it is very difficult to quantify the behavior of perovskites under the influence of each stress factor, because a multidimensional parameter space is involved.
The structure of perovskite is generally described by the ABX3 formula, where:
A is a cation in the form of an organic group (carbon -based) or inorganic.
B is a cation in the form of lead or tin.
X is an anion, a halogenure based on chlorine, iodine, fluorine or its combinations.
As you can see, the number of possible chemical combinations is enormous in itself. In addition, each of these combinations must be evaluated under multiple environmental conditions. These two requirements lead to a combinatorial explosion. We obtain a hyperparameter space which cannot be explored by conventional experimental methods.
In the first key step towards solving these problems, the researchers of the UC Davis College of Engineering, led by the students of Marina Leite and graduates Meghna Srivastava and Abigail Hering, decided to test whether the automatic learning algorithms could be effective in testing and predicting the effects of humidity on the degradation of materials.
They built a system to measure the efficiency of photoluminescence of five different perovskite films under 6 -hour repeated humidity cycles that simulate accelerated day and night meteorological patterns based on summer days typical of northern California. Using a high speed configuration, they collected 50 spectra of photoluminescence each hour and 7,200 spectra in a single experience, which is enough for a reliable analysis based on automatic learning.
The researchers then applied three automatic learning models to data ensembles and generated forecasts of photoluminescence responses dependent on the environment and compared their precision quantitatively. They used linear regression (LR), the echo state network (ESN) and the seasonal integrated automatic moving average with exogenous regressor algorithms (Sarimax) and have found values of the average quadratic error of standard root (NRMSE). The predictions of the model were compared to the physical results measured in the laboratory. The linear regression model had a 54%NRMSE value, the Echo state neural network had 47%NRMSE, and Samax did the best with only 8%in NRMSE.
The high and coherent precision of Sarimax, even when monitoring long-term changes compared to a 50-hour window, shows the capacity of this algorithm to model non-linear complex data from various organic-per-organic-inorganic hybrid compositions. Overall, the precise predictions of the chronological series illustrate the potential of approaches based on data for the stability studies of perovskite and reveal the promise of automation – data science and automatic learning as tools to further develop this new material.
Researchers note in their article that the generalization of their methods with several compositions can help reduce the time necessary to set up a composition, which is currently the main strangulation neck of the Perovskites design process for absorption and light emission devices.
In particular, the combination of Sarimax with long-term long-term memory models (LSTMS) can allow the prediction of the chemistry of perovskite beyond the training set, which will also lead to a precise evaluation of the stability of the compositions currently sub-studied.
In the future, scientists plan to extend their work by adding environmental stressors other than humidity (such as oxygen, temperature, light and tension). Combinations of many stressors can simulate operating conditions in various geographic locations, providing an overview of the stability of Hoip solar cells without the need for long experiences in each individual location.
