The recently completed AGROS establishment trial is taking researchers another whole step closer to the realisation of a fully self-reliant nursery . In the trial , an Artificial Intelligence algorithm based on Reinforcement Learning controlled the climate in a semi - commercial nursery , resulting in a productive cucumber harvest . Another milestone was marked by the successful deployment of a Digital Twin , which controlled greenhouse climate , irrigation scheme , and crop management . Anja Dieleman , AGROS project leader and research worker at Wageningen University & Research ’s business unit , Greenhouse Horticulture , partake in a sum-up of the successful proof trial and its results .
" In the past two years , we work out on the building blocks for autonomously manipulate cultivation of cucumber . We determine the plant traits essential to determination - qualification for craw direction and clime control and choose the sensor to measure said trait . To control the greenhouse autonomously , we develop algorithmic program based on two approaches : a mechanistic model - establish Digital Twin and a automobile learning algorithm based on Reinforcement Learning , " explains Anja . Early this yr saw the kickoff of the validation trial : These two coming and a grower ’s reference were practice in a greenhouse trial with real cucumber vine crops to evaluate their execution at the WUR enquiry facilities in Bleiswijk ( the Netherlands ) .
The objective was to reach the high possible net profit , which was determined by the balance between variable cost ( electricity , natural gas , CO2 ) and benefit ( number of reap cucumbers , calculate on fruit weight unit ) . On May 11 , the last cucumbers were glean in the greenhouse compartment of the AGROS establishment trial .
product in growers ' compartment matchespredictionThe climate control and crop direction in the ' growers ' compartment was base on current grower knowledge and best drill , with a pre - defined culture and irrigation strategy that target to realize balanced crop increase and production . The ground of the scheme was the crop ’s demand for assimilates , provided by sun , extra LED inflammation , and the temperature and CO2 supply strategy . The irrigation strategy resulted in a harvest with high growth at the starting time of culture . When the first flowers appear , the fruit pruning strategy was determined based on the number of newly formed leave-taking and the ask light levels . The experts ' concept was to grow a vigorous crop , with the supposition that its product would cover for the costs of electrical energy , oestrus , and CO2 . " The number of yield harvested very tight matched the predicted yield , which was base on the cultivation plan made at the trial ’s beginning . It shows the lustiness of the program . This control strategy prove to be favourable : the last profit in the grower ' compartment was the highest , " says Anja .
Closest to autonomous refinement in commercial practiceThe Digital Twin that controlled the second compartment was generated by the combined crop and mood good example of the business building block Greenhouse Horticulture . In this close - to - real environment , the Digital Twin fix the ideal control strategy based on the responses of the simulated climate and virtual cucumber harvest . It used real - prison term data from climate sensors and manual crop measurement to ' ego - calibrate ' and improve its control scheme during the run . The Digital Twin realized an objective control scheme , in which it balance actual costs against require benefit : the cucumbers that would be harvested in the next two hebdomad . Although the number of harvested cucumbers was low , the Digital Twin made optimal use of sport in electricity prices , resulting in the lowest cost per kWh of electricity used . Anja : " take care at the near future tense , the Digital Twin is closest to the practical program of independent polish in commercial practice . It can take decisions objectively in complex nursery production systems , balancing variable resourcefulness monetary value and product prices . "
pioneer fully autonomous greenhouse control with Reinforcement LearningIn the third greenhouse compartment , the clime was see to it by a Reinforcement Learning ( RL ) algorithm that was cultivate on virtual data lot of cucumber vine crops and climate . " AI software in greenhouse horticulture is still in its early childhood , and this was one of the first sentence a greenhouse was keep in line fully autonomously by a Reinforcement Learning algorithm , " aver Anja . The example could control actuators like kindling , sieve use , carbonic acid gas compactness , and heat but was not trained to control irrigation and fruit pruning . In the first three hebdomad of cultivation , control was re-create from the ' grower ' compartment , and when the first fruits were set , the RL hire over . This ' black box ' framework resulted in a greenhouse climate that differed well from the other two compartments , but the crop proved to be able-bodied to cope well with the larger fluctuation in temperature . The result was good yield production , even with the limitation of a pre - set fruit pruning scheme .
Incorporating sensor into the ascendance loop"Ideally , all controls would have swear on clime and craw sensors providing uninterrupted , machine-driven , and objective information . However , in this validation trial run , we still based control on manual measurements to ensure that organisation failures would not halter the control systems . The next step in the development of self-directed glasshouse control is to comprise sensing element into the ascendency loop , which would require racy sensing element supported by soft detector - based solutions , " concludes Anja .
reference : wur.nl
