Gwajin Amfani da Makamashi da Sawun Carbon don Sabis na IoT Masu Gudanar da AI

Teburin Abubuwan Ciki

1. Gabatarwa

Haɗaɗɗiyar Hankalin Wucin Gadi (AI) tare da sabis na Intanet na Abubuwa (IoT) yana canza kwamfuta na gefe zuwa Harsashi, yana haifar da sabbin kalubale don gwajin amfani da makamashi da sawun carbon. Kayan aikin gwajin IoT na yanzu ba su da cikakkun iyawar ma'auni na makamashi da hayaƙin carbon, suna barin masu haɓakawa ba tare da bayanan tasirin muhalli mai mahimmanci ba.

2. Bayanan Bincike

2.1 Juyin Halittar Harsashi

Kayan aikin IoT sun samo asali daga maƙasudai masu sauƙi zuwa na'urori masu sarƙaƙƙiya tare da na'urori masu hanzari da ke iya tallafawa ayyukan AI. Girman da rarraba sabis na IoT masu gudanar da AI suna ci gaba da ƙaruwa, tare da Gartner yana hasashen cewa 75% na bayanan kamfani za a ƙirƙira kuma a sarrafa su a gefe.

2.2 Kalubalen Amfani da Makamashi

Bukatun lissafin AI suna girma sosai, suna ninka kowane watanni 4 idan aka kwatanta da lokacin Dokar Moore na watanni 24. Cibiyoyin bayanai a halin yanzu suna cinye kusan TWh 200 a shekara, tare da Google ta ba da rahoton cewa 15% na amfani da makamashi ana danganta shi da ayyukan AI/ML.

3. Tsarin Fasaha

3.1 Hanyar Ƙirar Makamashi

Samfurin amfani da makamashi don sabis na IoT masu gudanar da AI yana la'akari da duka abubuwan lissafi da na sadarwa. Jimlar amfani da makamashi $E_{total}$ ana iya bayyana shi kamar haka:

$E_{total} = E_{compute} + E_{communication} + E_{idle}$

Inda $E_{compute}$ ke wakiltar makamashin da aka cinye yayin ƙaddamarwa da horar da samfurin AI, $E_{communication}$ yana lissafin makamashin watsa bayanai, kuma $E_{idle}$ ya ƙunshi ainihin amfani da makamashi.

3.2 Lissafin Hayaƙin Carbon

Ana lissafin hayaƙin carbon bisa amfani da makamashi da abubuwan ƙarfin carbon na yanki:

$CO_2 = \sum_{i=1}^{n} E_i \times CI_i$

Inda $E_i$ shine makamashin da aka cinye a wuri $i$, kuma $CI_i$ shine ƙarfin carbon na hanyar wutar lantarki a wannan wuri.

4. Sakamakon Gwaji

Ƙimar gwaji ta nuna bambance-bambance masu mahimmanci a cikin amfani da makamashi a cikin gine-ginen samfurin AI daban-daban da yanayin turawa. Tsarin gwaji ya bayyana cewa:

• Samfuran tushen CNN sun cinye makamashi 23% ƙasa da daidai gine-ginen Transformer
• Tura gefe ya rage jinkiri da 47% amma ya ƙara amfani da makamashi da 18% idan aka kwatanta da tura gajimare kawai
• Dabarun ƙididdige samfurin sun sami ceton makamashi 35% tare da ƙarancin asarar daidaito

5. Aiwar Code

A ƙasa akwai sauƙaƙan aiwar Python don kimanta amfani da makamashi:

class EnergyMonitor:
    def __init__(self, carbon_intensity=0.5):
        self.carbon_intensity = carbon_intensity  # kgCO2/kWh
        
    def estimate_energy(self, model_size, inference_time, device_power):
        """Kimanta amfani da makamashi don ƙaddamarwar AI"""
        energy_kwh = (device_power * inference_time) / 3600000
        carbon_emissions = energy_kwh * self.carbon_intensity
        return {
            'energy_kwh': energy_kwh,
            'carbon_kg': carbon_emissions,
            'model_size': model_size
        }
    
    def optimize_deployment(self, models, locations):
        """Ingantaccen tura samfurin mai sanin carbon"""
        best_config = None
        min_carbon = float('inf')
        
        for model in models:
            for location in locations:
                carbon = self.calculate_carbon_footprint(model, location)
                if carbon < min_carbon:
                    min_carbon = carbon
                    best_config = (model, location)
        
        return best_config, min_carbon

6. Aikace-aikacen Gaba

Binciken ya nuna wasu madaidaitan makoma masu ban sha'awa:

• Tsara Aiki Mai Sanin Carbon: Rarraba aikin aiki bisa bayanan ƙarfin carbon na ainihin lokaci
• Ingantaccen Koyon Tarayya: Horar da AI mai rarrabuwa mai ingancin makamashi a cikin na'urori na gefe
• Haɗin Ginin Kayan Aiki-Software: Na'urori masu hanzari na musamman don AI na gefe mai ingancin makamashi
• Ma'auni na Daidaitacce: Ma'aunin makamashi da carbon na masana'antu gaba ɗaya don sabis na IoT masu gudanar da AI

7. Bayanan Dogaro

1. Trihinas, D., da sauransu. "Zuwa Gwajin Amfani da Makamashi da Sawun Carbon don Sabis na IoT Masu Gudanar da AI." IEEE IC2E 2022.
2. Strubell, E., da sauransu. "Abubuwan Makamashi da Manufofi don Koyo Mai Zurfi a cikin NLP." ACL 2019.
3. Schwartz, R., da sauransu. "Green AI." Communications of the ACM 2020.
4. Zhu, J., da sauransu. "CycleGAN: Fassarar Hoto zuwa Hoto mara Biyu ta Amfani da Cibiyoyin Adawa masu Daidaitaccen Zagaye." ICCV 2017.
5. Hukumar Tarayyar Turai. "Yarjejeniyar Green ta EU." 2020.