Neuro Pulse AI

Multi-Strategy Expert Advisor Based on a Deep Neural Network

Trading Instrument: EURUSD. It is not recommended to run on other instruments — the network was trained strictly for this pair.

Trading Timeframe: Attach the EA to M15 (H4 will also be used for the neural network).

Brokers are recommended: with GMT time offset: +2 winter, +3 summer, as the neural network was trained on this offset.

Deposit: Depends on the Risk parameter and account type. (For example, it can work with $5 if it is on a cent account and Risk=500). It is recommended that your deposit exceeds the Risk parameter (e.g., for Risk=5000, deposits of 5000 and above are suitable).

Recommendations for the Risk parameter: Risk > 5000 — normal mode, Risk > 10000 — safe mode, Risk = 500–2000 — high-risk, Risk < 500 — extreme.

Commission: If your account's commission is less than 5 or absent, adjust the Manual Commission parameter accordingly. The lower the broker's commission and spread, the better for stability. If your broker offers a spread (or commission) refund program — Rebate — activate this service; this way, part of your trading costs will be returned.

1. Technological Foundation

1.1. Neural Network Architecture

The EA is based on a deep neural network (proprietary development, no third-party libraries), trained using the author's indicator Neuro Future.

• Network Structure: Four-layer model.

  • Outer Layers: Input and output layers.

  • Inner Layers: Two hidden data processing layers.

• Input Data: The network receives an array of 96 H4 bar prices, specially prepared.

• Output Data: Price direction forecast.

1.2. Training Process

• Training Period: 3 years of historical data.

• Instrument: EURUSD, H4 timeframe.

• Training Dataset: 5,500 specially prepared examples (input and output signals).

1.3. Adaptability to the Period

Tests of this neural network demonstrated high efficiency on the M15 period. The EA tends toward profitability even without deep parameter optimization, using mechanisms of adaptation to current volatility.

2. Innovative Capital Management System (Multi-Strategy)

A unique optimization and order management system is applied to eliminate local drawdowns.

2.1. Virtual Strategies System

• Strategy Generation: An internal algorithm creates a multitude of virtual trading systems with random parameters.

• Statistics Collection: Each strategy undergoes a series of tests, following which complete trade statistics are gathered.

• Selection of Leaders: A TOP list of the best configurations is formed. The final version of the EA simultaneously runs 20 best virtual strategies using the signals of one neural network.

2.2. Virtual Capital Management

• Each of the 20 strategies operates with a limited virtual deposit.

• Strategy Stop-Loss: If a specific strategy's virtual deposit exceeds the loss threshold, it forcibly closes all related positions (partial loss fixation).

3. Position Opening and Management Logic

3.1. Entry Conditions

A triple neural network confirmation is used to filter false signals:

1. Forecast on the first bar (start of the forecasting window).

2. Forecast on the middle bar.

3. Forecast on the last bar.

A signal is considered valid if all three forecasts match in direction and their values exceed the set minimum signal strength threshold.

3.2. Position Management Logic (Averaging and Reversal)

The EA's behavior when the price moves against an open position depends on the current neural network readings:

• Scenario A (Trend Persists): If the neural network confirms the initial direction, position averaging occurs with a small increasing lot multiplier.

• Scenario B (Trend Change): If the neural network shows the opposite direction, a full position reversal occurs (closing current positions and opening in the opposite direction).

• Scenario C (Uncertainty): If the neural network produces multidirectional signals within the window, averaging and reversal are not performed (standby mode).

4. Protective Mechanisms and Service Functions

4.1. Auto Restart

The system monitors the overall total profit. When a set percentage threshold is reached, all real and virtual positions are forcibly closed. Trading starts anew "from a clean slate," allowing accumulated profits to be locked in and the load on the deposit to be reset.

4.2. Drawdown Management

Flexible settings for controlling maximum drawdown are implemented:

• Option 1: When a critical drawdown level is reached, all positions are closed, and trading resumes from scratch.

• Option 2: Close all positions, save state in memory, and restart trading only after a user-defined timeout ("cool-down" period) has elapsed.

4.3. Weekend Manager

Before the market closes, the EA can forcibly close all positions and save the current state to avoid gaps on Monday. This function can be disabled.

4.4. Trading Filters

• Time Filter: Ability to set allowed/forbidden trading hours.

• News Filter: Built-in integration with the MT5 terminal's economic calendar to block trading during the release of important news.

5. Key Product Advantages

• Real Neural Network: Deep machine learning on 5,500 examples.

• Multi-Strategy Stability: Parallel operation of 20 best systems on a single signal eliminates the impact of a single losing streak on the balance curve.

• Adaptive Management: The logic itself determines when to average and when to reverse a position, eliminating "luck-based" trading.

6. Fault Tolerance and System Requirements

6.1. Interruption Protection

The EA is equipped with a built-in safety system for unforeseen stops (connection failures, terminal restart, power outage):

• Data of virtual strategies and current operating parameters are saved to files.

• After restoring the connection or restarting the terminal, the EA automatically loads the state from the files and continues operation from the same point without losing strategy logic.

6.2. Recommended Environment

For stable round-the-clock operation, it is recommended to use:

• A dedicated personal computer (PC).

• A remote server with a desktop (VPS/VDS).