Best in class Deep Learning Predictive Model

Informazioni relative al progetto

BEYOND

ID dell’accordo di sovvenzione: 817342

DOI

10.3030/817342

Progetto chiuso

Data della firma CE 2 Maggio 2018

Data di avvio 1 Giugno 2018

Data di completamento 30 Novembre 2018

Finanziato da

• INDUSTRIAL LEADERSHIP - Innovation In SMEs

Costo totale

€ 71 429,00

Contributo UE

€ 50 000,00

Coordinato da

HPA SRL
Italy

Periodic Reporting for period 1 - BEYOND (Best in class Deep Learning Predictive Model)

Periodo di rendicontazione: 2018-06-01 al 2018-11-30

Sintesi del contesto e degli obiettivi generali del progetto

More and more organizations are turning to predictive analytics to increase their bottom line and competitive advantage. Examples of industries benefiting from predictive analytics and machine learning include finance, energy, marketing, insurance, healthcare, and media communication. This list can be endless as virtually any industry might take advantage of predictive analytics to solve specific pains related to better and more timely approaching their corresponding market, by analysing technological trends as well as stocks or commodities prices, depending on the case.

But predictive analytics is difficult, and this is especially truth for time series forecasting, the application field of HPA's BEYOND. Time series analysis is the predictive technique used to forecast future values based on previously observed ones. Technically, this analysis is way harder than simple data classification and regression, as they add the complexity of order or temporal dependence between observations.
Traditionally, time series forecasting has been dominated by linear statistical methods (like ARIMA - AutoRegressive Integrated Moving Average) because they are well understood and effective on many problems. But these are traditional methods that suffer from several limitations.

Due to this limitations, existing techniques often depend on “hand-crafted” features that are expensive to create and require expert knowledge of the field. HPA’s solution tackles all these challenges using innovative Deep Learning proprietary algorithms in a low-cost solution.

Predictive analytics is widely used in many different industries so it’s no wonder that more and more organizations are turning to forecasting tools to increase their bottom line and competitive advantage. The market is expected to grow from EUR 3,1BN in 2016 to EUR 10BN by 2022, at a CAGR of 22,1%. As a recent Forbes’ study points out, an increasing percentage of managers (comprised between 50% and 70%) considers the adoption of predictive analytics solutions important or very important.

More accurate forecasting can result in immediate benefits both for companies and citizens. For example, one of BEYOND target industries is the Energy Trade, in which the increased accuracy in the prediction of how much electricity needs to be generated and customer demand, allows energy suppliers to reduce imbalance charges and ultimately reduce energy costs for citizens.

The overall objective of the BEYOND project is to develop a full-scale version of the solution and massively market BEYOND in Europe and other international markets in order for HPA to gain an important share of the Predictive Analytics market and obtain a leadership in the field.

The feasibility study that was completed allowed HPA to design the full scale architecture, implment an MVP of the solution, complete several testing sessions in real world environment and refine the business plan. In addition, the desicion to apply for a SME Instrument Phase-2 grant was taken and a preliminary workplan has been defined.

Lavoro eseguito dall’inizio del progetto fino alla fine del periodo coperto dalla relazione e principali risultati finora ottenuti

During the phase 1 implementation, HPA focussed on the realisation of a feasibility study for the BEYOND project. In performing such activity, Mr. Tilman Süss supported HPA as external coach (funded by the SME Instrument Phase 1), in particular regarding the definition of the business focus and the business model, and the possibility to enter the German market.

As first activity, a technical assessment of the BEYOND full-scale solution has been realised. Following several sessions of consultations with industry experts, interviews to prospects and the SME-I coaching, the team formalised the key features of the solution and designed the software modules that will form the final solution. Following the functional requirements, the full-scale architecture has been designed. The functional architecture has been designed to include the production, the stage and the development environments. Moreover, detailed requirements to ensure scalability, reliability and security have been defined.
Finally, the physical architecture has been determined by choosing the cloud environment (Amazon AWS) that will host the full-scale solution and all the services required to implement the architectural requirements. In doing so, a rigorous scalability analysis has also been completed, to assess costs and risks.

In parallel with the architecture design, an advanced BEYOND MVP has been developed by the HPA team in collaboration with external collaborators. This MVP already embeds the functional workflow of the functional requirements and has been deployed on a private HPA server. The MVP has also been extensively tested in real world environments by prospects operating in key target industries, in order to validate the accuracy of the predictive models as well as the key features that will set BEYOND apart from the competition. BEYOND performance have been excellent in all the testing and allowed HPA to land the first paid contracts with early adopters.

HPA then focussed on the definition of users’ needs and market size, and the definition of the business focus and model. The disruptive Forecast-as-a-Service model that BEYOND will adopt, will be based on a “freemium” approach, including free services, premium services, and on-line consultancy services. The business plan, including the a thorough competitors analysis, the commercialisation actions and the financial estimations has been then realised.
The SMEs will be the main target of BEYOND with a flexible pricing that will allow them to adopt advanced predictive models. In terms of verticals, the energy industry will be the first market to be penetrated. Additional industries will be then addressed one by one in order to allow for an organic growth of the company and an effective use of resources. HPA also outlined the long-term vision of BEYOND: in 10 years the company aims to transition from a simple B2B SaaS solution to an ambitious platform that could be used by other software house to develop their own predictive models via APIs.

The IPR has been also studied. A prior art analysis has been conducted, which outlined that there are no constraints in the freedom to operate from the side of the proposer. The possibility to apply for a patent will be deeply analysed with legal advice in the next phase. The possibility to protect the BEYOND software through copyrights has also been investigated. Finally, the BEYOND official logo has been designed and web domains purchased.
During this period, HPA has been very active in the search of investors and sources of funds. In this sense, a plan for the Phase 2 implementation has also been made. Such plan included the preliminary definition of the workplan, the duration of the project and the estimated funding.

Progressi oltre lo stato dell’arte e potenziale impatto previsto (incluso l’impatto socioeconomico e le implicazioni sociali più ampie del progetto fino ad ora)

BEYOND predictive models have been proved to be more accurate than many other competitors solutions thanks to testing activities conducted in real world environments.

Specifically, the following testing activities were completed during the Phase-1 project, in the following industry sectors:

ENERGY:
Client Need: Electricity can’t be stored efficiently or cheaply at scale, so electricity suppliers must balance the energy that they produce themselves or procure from third parties with the energy that their customers use. This means, ahead of time, forecasting how much electricity is going to be generated, forecasting customer demand, and taking any actions to balance them out: buying or selling additional electricity as required. Any imbalance between generation and demand can result in suppliers facing costly charges from National Grid, who are forced to act in real time to balance the system.
Testing Activities and Results: HPA offered a prospect the possibility to use BEYOND predictive models in parallel with the company current supplier predictive solution. Testing activities lasted 3 months during which BEYOND was trained on a dataset of 30-months of historical data. BEYOND extreme accuracy was successfully demonstrated thanks to a 3% average error against the 6% average error of the competitor solution. An estimate of the economical benefit provided by the higher accuracy was also calculated and amounted to €400.000.

PHYSICAL SECURITY MANAGEMENT:
Client Need: HPA's prospect wanted to enhance the itssolution by developing a new module for anomaly detection, to pursue the following objectives:
• Detect failures and malfunctioning.
• Detect real failures through statistical analysis.
• Detect behavioural anomalies that generate false malfunctioning.
Testing Activities and Results: Activities were structured in two phases:
• Phase 1 - during this first period the HPA team completed an analysis of the prospect historical data to develop the most appropriate predictive algorithms.
• Phase 2 – Beyond is being integrated into the prospect's solution, while a refinement is being performed on the predictive models in order to improve the accuracy and extend the analysis to the whole geographical area covered by the prospect.

TECHNOLOGY INTELLIGENCE
Client Need: HPA prospect wanted to develop a technology intelligence tool able to predict future technology trends in several innovation fields. The solution, to be integrated into existing tools and platforms, has been designed to be used by all the organisations operating in the innovation field (SMEs, Large companies, research centres, universities etc.).
Testing Activities and Results: HPA integrated the BEYOND MVP into the client platform and designed specific predictive models to support the forecasting of technology trends based on the analysis of datasets such as patents and scientific publications.

More in general BEYOND will allow many SMEs to have access to advanced predictive tools, today only used by large corporations, in order to become more competitive.

BEYOND by HPA - Logo