PlannedServicesMarktconsultatie

RWS Challenge Machine-Learning model for Wave Parameters

Rijkswaterstaat Water, Verkeer en Leefomgeving · Marktconsultatie · 1 lots · 426401
Type
Services
72200000 · Software programming a
Estimated value
Not published
To deadline
Ongoing
Knock-outs
n/a
exclusion grounds
Award basis
Best price-quality ratio
4 criteria
Assess manuallyconfidence high

Rijkswaterstaat is organizing a challenge for the development of a fast machine-learning model for wave parameters in the North Sea. Successful participants may be invited for a multiple tender (MVO) for a proof-of-concept and further rollout to a pre-production model.

Services · Marktconsultatie · National procedure

Complete timeline
Contract type
Services
Marktconsultatie
Estimated value
Estimate not published
Submission deadline
Ongoing
Scope
National
National procedure
Lots
1
1 lots
Main CPV code
Software programming and consultancy
Location
NL
CharacteristicsCPV 72ServicesMachine LearningAIWater models

01What is being requested

Development of a Machine-Learning model for calculating wave parameters (significant wave height and swell height) at North Sea locations. The assignment starts with a Challenge (market consultation) and may lead to a Multiple Tender (MVO) for a Proof-of-concept and the subsequent development into a pre-production model.

The purpose of the market consultation is to gain insight into innovative solution possibilities for future, fast ML water models.

72200000Services
1RWS Challenge Machine-Learning model for Wave Parameters

02Strategic insight

Strategic insight · AI analysis
During the Challenge, focus on a robust training approach that directly aligns with the operational needs (MLOps) and the user requirements of the HMC. Ensure strict compliance with the technical requirements, particularly the hardware constraints (VRAM/CPU) and the determinism requirement. Do not optimize the model solely on general accuracy, but specifically on the Peak Error for extreme values, as this determines 40% of the performance score.
Read automatically from the tender documents using AI. Always verify against the original documents.

03Points of attention

Important · 8
The Plan of Approach may cover a maximum of 2 single-sided A4 pages (excluding cover page).Important
The implementation must use Python version 3.12 or higher.Important
For dependency management, the use of pyproject.toml is mandatory.Important
A fixed seed (random-seed=42) must be used for stochastic operations.Important
Models that exhibit more than 5% deviation during repeated training will be excluded.Important
Inference must run on a single GPU with a maximum of 16GB VRAM.Important
CPU memory usage must not structurally exceed 12GB during inference.Important
Participants in the Challenge receive no compensation for participation.Important

04Award criteria

Performance: Total score of the model (RMSE, Peak Error and Skill Score)weight n/a

Assessment based on a validation test with three core metrics: General Accuracy (RMSE) for 40%, Performance at Extremes (Peak Error) for 40%, and Predictive Power (Skill Score) for 20%.

Qualitative: Alignment with user requirements and application20%

The extent to which the proposed training approach aligns with the wishes and needs of the end users and the final application (such as ensemble predictions).

Qualitative: Operational deployment and deployment20%

The extent to which the bidder takes into account the desire to deploy the model operationally and manage it as a production-ready solution that meets the technical requirements.

Price: Tender price for pre-productionweight n/a

Tender price with a price ceiling of € 95,000 excl. VAT.

05Process & timeline

Publication of Challenge via Tenderned / Digishape
26 mei 2026
Registration for information session
29 mei 2026
Online information session – teams
1 juni 2026 14:00
Closing date for submitting questions
5 juni 2026 11:00
Publication of answers to questions
8 juni 2026 17:00
Submission of Plan of Approach
15 juni 2026 11:00
DigiShape Pitches
19 juni 2026
Announcement of follow-up and decision to issue MVO
23 juni 2026
Publication of MVO via TenderNed
25 juni 2026
Closing date for MVO tender
31 augustus 2026
Final award based on Proof-of-concept
18 september 2026
Contract start date
7 oktober 2026
Einddatum overeenkomst
15 februari 2027

06Value in context

Estimate not published

The contracting authority did not publish an estimated value — common for a large share of contracts. The EU threshold for diensten is € 221.000, for reference.

07Likely competitors

#Likely bidderFitWins
1COMPAREX Nederland B.V.SME9130×
2Conclusion B.V.SME8518×
3IlionX Group B.V.SME7713×
4CGI GovernmentUnknown7612×
5KPMG Advisory N.V.SME7615×
6Sogeti Nederland B.V. SME7516×

08Tender documents

TN591656 - EFE1 Vrijwillige aankondiging van voorafgaande marktconsultatie 20260526135912pdfMay 26, 2026 · 171 KB
31219005 Challenge Machine Learning-model voor GolfparameterspdfMay 26, 2026 · 610 KB

09Legal themes that may be relevant here

10Frequently asked questions

What is the purpose of the RWS Challenge?
The purpose is to gain insight into innovative solution possibilities for future, fast Machine Learning water models within the AI Impulse Program for water movement models (AIWaM).
What is the possibility of a follow-up assignment after the market consultation?
Upon sufficient results and potential from the Challenge, the assignment may be continued via a multiple tender (MVO).
What expertise is required to participate in this tender?
Highly specialized knowledge of both Artificial Intelligence (AI) and hydraulic engineering parameters is required to develop models that meet the functional and technical requirements.
On which criteria can a potential multiple tender (MVO) be assessed?
The assessment during an MVO focuses on the total score of the model, the alignment with user requirements and application, and the operational deployment.

Automatically compiled from the official tender data and documents.

11Estimated value versus the market

p25
€ 327K
median
€ 800K
p75
€ 2,5 mln
deze opdracht

Gegunde waarden in CPV 72 · diensten n=1436