Machine Learning methods to estimate the performance of aquafarms

BlueBRIDGE receives funding from the European Union’s Horizon 2020
research and innovation programme under grant agreement No. 675680 www.bluebridge-vres.eu
Konstantinos Bovolis
kbovolis@i2s.gr
Machine Learning methods to
estimate the performance of aquafarms
Supporting Blue Growth with
innovative applications based on
EU e-infrastructures
14-15 February 2018, Brussels

Outline
Challenges and Needs
BlueBRIDGE Solution
BlueEconomy: Performance Evaluation, Benchmarking and
Decision Making
Case Study
Conclusion
15/2/2018 2
“Supporting Blue Growth with innovative applications based on
EU e-infrastructures”, 14-15 February 2018, Brussels

Challenges that have to be addressed:
• maintaining the economic viability of the sector by
reducing costs and increasing production
• guaranteeing high quality food and animal welfare
• addressing environmental concerns.
All aquaculture producers are concerned about improving the
performance of their companies in terms of cost, feed
conversion, growth rate and mortality and at the same time, be
sustainable and environmental friendly
Challenges and Needs
15/2/2018 3

It is not only equipment and
hardware!
Unfortunately, answering this question is not that simple
• Aquafarmers can invest in the latest technology for cages or on
the most advanced feeding systems but they cannot forget two
key aspects:
i. an aquaculture comes with its own array of environmental
challenges that have a huge impact on production system;
ii. an aquaculture business can be sustainable only if they are
able to continuously monitor and improve its performance
15/2/2018 4

Provide innovative data analytics and machine learning services
that will benefit all the stakeholders of the aquaculture sector
The aim is to support:
• Companies to maximize the growth rate, reduce costs and
minimize the impact on the environment
• Investors to make efficient identification of strategic locations of
interest and select the most profitable investments
• Governments and environmental agencies to evaluate the current
situation and define policies
• Researchers to generate new knowledge and evaluate the
practical indicators of aquafarming performance
BlueBRIDGE Solution
15/2/2018 5

Blue Economy
Performance Evaluation, Benchmarking &
Decision Making
Goal:
Estimate/create KPIs Tables (biol. FCR, SFR,
Mortality Rate) based on historical data using
Machine Learning Techniques (i.e. GAMs, MARS)
Define a Site:
•location
•temperature
profile
Setup Site
Performance Evaluation
Estimate KPIs:
• Collect data
• Upload data
• Generate
models
Setup Model
Benchmarking & Decision Making
Goals:
• Create accurate and feasible production plans
• Benchmark the performance against the
competition
Perform production
planning by:
• Create scenarios
• Assess the KPIs
• Benchmarking
What-If Analysis
Decision
15/2/2018 6

Aquafarms engagement
10 Aquaculture companies have already started to utilizing BlueBRIDGE
services and tools, via their own VREs:
 ARDAG Aquaculture
 iLKNAK Aquaculture
 GALAXIDI MARINE FARM S.A.
 NIREUS AQUACULTURE S.A.
 MARKELLOS AQUACULTURE LEROS S.A.
 STRATOS AQUACULTURES
 ALIEIA S.A.
 FORKYS
 ELLINIKA PSARIA
 KIMAGRO FISH FARMING LTD
15/2/2018 7

Case Study:
Performance Evaluation,
Benchmarking &Decision Making
How to evaluate the
performance of Sea
Bream production at
site A over different
stocking months?
Define the Site A
(Setup Site)
1
Create a production
model for Site A
(Setup Model)
2
Create hypothetical
scenarios for Site A
(What-If)
3
Evaluate results:
• Production KPIs
• Benchmarking
4
15/2/2018 8

Step 1: Setup Sites
• Aquafarm manager can define the average temperature fortnightly and the
geographical location of the site of interest (i.e. Site A)
Case Study:
15/2/2018 9

Step 2: Setup Model
• Aquafarm manager can develop reliable and powerful Machine Leaning
models, which are capable to estimate vital production indicators, such as
biological FCR, SFR and Mortality Rate, providing real historical production
data and details regarding the production of the specific fish species (i.e. Sea
Bream) of the site of interest (i.e. Site A)
• For the particular case study, aquafarmer needs to upload production data for
different stocking periods for the Sea Bream species at the Site A
Note:
• Very often data need to be cleaned and preprocessed before the analysis is
executed
• ‘Setup Model’ tool includes processes so as to remove automatically
inconsistent entries and outliers from the processed data
• However, aquafarmer is responsible to provide to the system good quality
data
Case Study:
15/2/2018 10

Step 2: Setup Model
Case Study:
15/2/2018 11

Step 2: Setup Model - Results
• The outcome of the modeling process is a simulation of the relationship
between growth, feeding and temperature
• Development of tables for
 Biological FCR,
 Feeding Rate and
 Mortality Rate
in terms of fish weight (Avg. Weight Categories) and temperatures (Avg. Sea
Temperature)
Case Study:
15/2/2018 12

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
1 3.47 3.09 2.72 2.38 2.07 1.81 1.59 1.42 1.29 1.21 1.16 1.14 1.14 1.15 1.17 1.19
3 3.46 3.08 2.72 2.37 2.07 1.80 1.58 1.41 1.28 1.20 1.15 1.14 1.13 1.15 1.16 1.18
8 3.44 3.06 2.70 2.35 2.05 1.78 1.56 1.39 1.26 1.18 1.13 1.12 1.11 1.13 1.14 1.16
20 3.39 3.01 2.65 2.31 2.00 1.74 1.52 1.34 1.22 1.13 1.09 1.07 1.07 1.08 1.10 1.11
50 3.32 2.94 2.58 2.24 1.93 1.66 1.45 1.27 1.15 1.06 1.02 1.00 1.00 1.01 1.02 1.04
100 3.53 3.16 2.79 2.45 2.14 1.88 1.66 1.49 1.36 1.28 1.23 1.21 1.21 1.22 1.24 1.26
150 4.10 3.72 3.36 3.02 2.71 2.44 2.22 2.05 1.93 1.84 1.80 1.78 1.78 1.79 1.80 1.82
200 4.48 4.11 3.74 3.40 3.09 2.83 2.61 2.44 2.31 2.23 2.18 2.16 2.16 2.17 2.19 2.21
250 4.50 4.12 3.76 3.41 3.11 2.84 2.62 2.45 2.32 2.24 2.19 2.17 2.17 2.18 2.20 2.22
300 4.38 4.01 3.64 3.30 2.99 2.73 2.51 2.34 2.21 2.13 2.08 2.06 2.06 2.07 2.09 2.11
350 4.37 3.99 3.63 3.28 2.98 2.71 2.49 2.32 2.19 2.11 2.06 2.04 2.04 2.05 2.07 2.09
400 4.48 4.11 3.74 3.40 3.09 2.83 2.61 2.44 2.31 2.23 2.18 2.16 2.16 2.17 2.19 2.21
Step 2: Setup Model – Results of Machine Learning process
Avg. Sea Temperature
Avg. Weight
Categories
Case Study:
15/2/2018 13

0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
4.50
5.00
1 3 8 20 50 100 150 200 250 300 350 400
BiologicalFCR
Average Weight Categories
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
Case Study:
15/2/2018 14

0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
4.50
5.00
12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
BiologicalFCR
Temperature
1
3
8
20
50
100
150
200
250
300
350
400
Case Study:
15/2/2018 15

Case Study:
Step 3: What-If Analysis
• Aquafarm manager can draw a hypothesis and evaluate it, using an already
existing production model
• The ‘What-If Analysis’ tool:
 calculates production indicators which are able to estimate the
performance of the fish growth
 presents the results in a meaningful tables and interactive graphs
 benchmark the production performance against competition over the
same hypothesis
15/2/2018 16

Case Study:
Step 3: What-If Analysis
15/2/2018 17

Case Study:
Step 3: What-If Analysis – Case Study
• Baseline scenario: evaluate the Sea Bream production performance whether a
population of 500.000 fish will be stocked at “Site A” in December (01/12)
with initial average weight 2 grs and they are cultivated for 18 months (harvest
date 31/05)
• Alternative scenario: stock the fish 3 months later, namely on March (01/03).
Thus, the harvest date will be at the end of August (31/08). The other
conditions are similar with baseline scenario
15/2/2018 18

Case Study:
Step 3: What-If Analysis – Case Study Results
Baseline Scenario Alternative Scenario
15/2/2018 19

“Supporting Blue Growth with innovative applications based
on EU e-infrastructures”, 14-15 February 2018, Brussels
Case Study:
15/2/2018 20

Case Study:
Monthly Feed Consumption
Dec-17 2360.23 Mar-18 1741.49
Jan-18 3520.59 Apr-18 2837.87
Feb-18 3269.62 May-18 4128.76
Mar-18 4231.18 Jun-18 4615.31
Apr-18 2181.23 Jul-18 12295.91
May-18 3752.54 Aug-18 25072.67
Jun-18 9849.37 Sep-18 32578.39
Jul-18 22691.04 Oct-18 39285.06
Aug-18 34990.53 Nov-18 38698.9
Sep-18 41002.50 Dec-18 32145.91
Oct-18 45299.59 Jan-19 14659.85
Nov-18 41002.40 Feb-19 13555.94
Dec-18 30098.66 Mar-19 15467.33
Jan-19 16653.27 Apr-19 19664.4
Feb-19 9741.25 May-19 26381.92
Mar-19 9689.18 Jun-19 42909.94
Apr-19 14294.53 Jul-19 56629.56
May-19 23285.18 Aug-19 61458.86
18 317912.89 18 444128.07
317.91 444.13
 283.130 tons saving around 11% comparing with
the total feed consumption at baseline scenario
(317.91 tons)
 However, after the cultivation of a duration of
18th months (31/05) the average weight in
baseline scenario is estimated to be 347.78 grs
against 305.32 grs of the alternative scenario
15/2/2018 21

Conclusions
IT
Providers
Gain knowledge from
the aquaculture
domain
New approaches to
face problems
Combine production
and techno-
economical models
15/2/2018 22
Aquaculture
New perspectives to
overcome production
problems
Enrich capabilities to
process historical
production data
Benchmarking – change
mentality towards to
open sector
Encourage to use
innovative cloud-based
apps, such as BlueBRIDGE

Any Questions?
15/2/2018 23
http://www.bluebridge-vres.eu/

Step 2: Setup Model – Sample Data
A. Periodic or “Sampling to Sampling” dataset contains data which are
gathered from sequential samplings by an aquaculture company:
• datefrom: the date when the sampling period is started
• dateto: the date when the sampling period is terminated
• openweight: the fish average weight at the beginning of the sampling
period
• closeweight: the fish average weight at the end of the sampling period
• avgtemperature: the average sea temperature of the sampling period
• openfishno: the number of fish at the begin of the sampling period
• closefishno: the number of fish at the end of the sampling period
Case Study:
15/2/2018 24

calculated attributes (KPIs production indicators):
• fcr: Biological Feed Conversion Rate, which is calculated from the
number of kilograms of feed used to produce one kilogram of fish,
measured at the end of the sampling period
• mortalityrate: ratio of dead fishes at the end of the sampling period
• sfr: Suggested Feed Ratio, which indicates the quantity of feed given
to the fishes over the period, measured at the end of the sampling
period
• sgr: Specific Growth Rate, which indicates the growth of the fish in a
particular period, measured at the end of the sampling period
Case Study:
Step 2: Setup Model – Sample Data
15/2/2018 25

Step 2: Setup Model – Weight limits
B. Weight Categories Dataset:
contains user-defined categories of average
fish weight which are corresponded to each
production KPI (FCR, SFR, SGR and
Mortality Rate)
FCR SFR SGR Mortality
1 0.50 1 1
3 1 3 3
8 2 8 8
20 3 20 20
50 5 50 50
100 8 100 100
150 10 150 150
200 15 200 200
250 20 250 250
300 30 300 300
350 50 350 350
400 100 400 400
1000 120 1000 1000
150
200
250
300
350
400
450
500
600
Case Study:
15/2/2018 26

Machine Learning methods to estimate the performance of aquafarms

More Related Content

Similar to Machine Learning methods to estimate the performance of aquafarms

More from Blue BRIDGE

Recently uploaded

Machine Learning methods to estimate the performance of aquafarms

Editor's Notes