NeuroPrice Navigator MT5

NeuroPrice Navigator 2.0 — TCNN‑LSTM Neural Network EA with Attention Mechanism

Brief Description

NeuroPrice Navigator 2.0 is a fully automated trading advisor for MetaTrader 5 that uses a hybrid TCNN‑LSTM neural network with a Feature Attention mechanism to predict price direction. The advisor self‑trains on historical data, adapts to changing market conditions, and includes a built‑in multi‑level risk management system.

How It Works

The advisor is built on a proprietary neural network model combining three components:

Temporal Convolutional Networks (TCNN) analyze a sequence of 32 previous bars, extracting local patterns and impulsive market movements. Two convolutional layers (24 and 12 filters) recognize short‑ and medium‑term price structures.

Feature Attention Mechanism evaluates the importance of each feature at each moment, automatically focusing on the most significant signals and ignoring market noise.

Long Short‑Term Memory (LSTM) processes the temporal sequence after the convolutional layers and attention mechanism, capturing long‑term dependencies and market context. A recurrent layer of 12 LSTM blocks considers history when forming the forecast.

The model outputs the probability of price growth P(up) in the range 0 to 1. Based on this value, a trading signal is generated:

  • P(up) ≥ 0.65 — strong BUY signal

  • P(up) ≤ 0.35 — strong SELL signal

  • 0.35 < P(up) < 0.65 — uncertainty zone, no trade is opened

The advisor retrains every 48 bars on a rolling history window (300 bars by default), using Backpropagation Through Time with adaptive learning rate and L2 regularization. Training includes cross‑validation with chronological fold splitting, preventing overfitting and ensuring robustness to market changes.

Advantages

  • Self‑learning and adaptability — no manual optimization for a specific currency pair is required; the advisor trains directly on the chart where it is installed and rebuilds model weights when the market regime changes.

  • Hybrid architecture — convolutional layers, attention mechanism, and LSTM capture short‑term impulses, medium‑term trends, and long‑term context simultaneously.

  • Multi‑level deposit protection — adaptive lot calculation from free margin, drawdown protection, limit on simultaneous positions, spread control, and volatility filter.

  • Flexible position management — real and virtual SL/TP, trailing stop with activation and step settings, and dynamic order distance based on ATR.

  • Information panel — displays current signal, model confidence, account status, drawdown, and training progress, plus manual BUY / SELL / CLOSE buttons.

  • State preservation — model parameters, optimizer state, and feature cache are saved to a single file, so training continues after terminal restart.

Input Parameters

Main Settings

Parameter Description Default
MagicNumber Unique identifier of the advisor's orders 2456
SessionStart Trading session start (UTC hours) 8
SessionEnd Trading session end (UTC hours) 22
UTCOffset Server time offset from UTC (hours) 0
EnableDebugLogging Enable detailed logging false
FixSeedForTesting Fix random number generator for testing true

Risk Management

Parameter Description Default
LotCalculationMode Lot mode: fixed or from free margin LOT_MODE_FIXED
FixedLotSize Fixed lot size 0.01
RiskPercent Risk per trade (% of free margin) 1.0
UseEquityProtection Reduce lot on high drawdown true
MaxDrawdownPercent Drawdown (%) after which lot reduction begins 20.0
MaxConcurrentTrades Maximum simultaneous positions 3

Execution & Broker

Parameter Description Default
InpOrderComment Order comment Trade
MaxSpreadPoints Maximum allowed spread (points) 40
MaxSlippagePoints Maximum slippage (points) 5
ExecutionRetries Order send attempts 3
ExecutionRetryDelayMs Delay between attempts (ms) 250
EnableExecutionDiagnostics Detailed execution diagnostics false

Real SL / TP

Parameter Description Default
StopLoss Stop loss (points), 0 — disabled 400
TakeProfit Take profit (points), 0 — disabled 300

Virtual SL / TP

Parameter Description Default
UseVirtualSL Enable virtual stop loss false
VirtualStopLoss Virtual stop loss (points) 0
UseVirtualTP Enable virtual take profit false
VirtualTakeProfit Virtual take profit (points) 0

Trailing Stop

Parameter Description Default
UseTralling_Stop Enable trailing stop true
UseAveragePrice Use average entry price for all positions true
TrallingStart Profit to activate trailing (points) 40.0
TrailingDistance Distance from price to SL (points) 20.0
TrailingStep Minimum SL improvement for modification (points) 5.0

Order Distance

Parameter Description Default
FixedDistancePoints Fixed distance (points) 30
UseDynamicDistance Use ATR-based distance true
DynamicDistanceAtrPeriod ATR period for dynamic distance 12
DynamicDistanceMultiplier Multiplier for dynamic distance 1.5

TCNN‑LSTM Model

Parameter Description Default
ML_TrainingPeriod Training period (bars) 300
ML_RetrainInterval Retraining interval (bars) 48
Lookback_Window History depth (bars) 32
ForecastBars Bars for forecast 3
ForecastThreshold Forecast threshold (ATR multiplier) 0.5
CNN_Filters Number of CNN filters 24
Kernel_Size Convolution kernel size 5
LSTM_Units Number of LSTM units 12
ML_LearningRate_Input Base learning rate 0.005
ML_DropoutRate Dropout during training 0.20
ML_BuyThreshold BUY threshold: P(up) ≥ value 0.65
ML_SellThreshold SELL threshold: P(up) ≤ value 0.35
L2_Lambda L2 regularization coefficient 0.001
Target_MSE Target MSE for early stopping 0.1
Max_Epochs Maximum training epochs 9
ML_BatchSize Batch size 16
ML_WalkForwardFolds Cross-validation folds 1
ML_LRPatience Epochs without improvement before LR reduction 6
ML_LRDecayFactor LR reduction factor 0.5
ML_MinLearningRate Minimum learning rate 0.0001
ML_BPTT_GradClipNorm Global gradient clipping norm 5.0
ML_BPTT_SampleClipNorm Per-sample gradient clipping norm 10.0
ML_BPTT_StateClip LSTM state limit 10.0
ML_BPTT_DeltaClip Limit for dh/dc per BPTT step 5.0

Feature Attention

Parameter Description Default
Attention_Heads Number of attention channels 2
Enable_Contextual_Attention Enable contextual attention true
Attention_Temperature Temperature for scaling 1.0
Attention_Dropout Dropout for attention 0.1

Indicators

Parameter Description Default
ADX_Period ADX period 10
CCI_Period CCI period 14
WPR_Period Williams %R period 12
STD_Dev_Period Standard deviation period 12
MFI_Period_New Money Flow Index period 10
ATR_Period ATR period 12

Volatility Filter

Parameter Description Default
Volatility_Filter Enable volatility filter false
Min_Volatility Minimum volatility (ATR value) 0.0003
Max_Volatility Maximum volatility (ATR value) 0.012

Usage Recommendations

The advisor is designed for currency pairs with moderate volatility (e.g., EURUSD, GBPUSD, USDJPY). Before running on a live account, test it in the MetaTrader 5 Strategy Tester with a fixed initial deposit (e.g., $10,000) and 1:100 leverage. Default parameters are optimized for the M15 timeframe and can be used without additional configuration.


Plus de l'auteur
NeuroPrice Navigator
Aleksandr Zavolskov
NeuroPrice Navigator 2.0 — intelligent trading advisor based on the TCNN-LSTM neural network model with an attention mechanism Brief description NeuroPrice Navigator 2.0 is a fully automated trading advisor for MetaTrader 4 that uses a hybrid TCNN-LSTM neural network architecture with a Feature Attention mechanism to predict the direction of price movement and make trading decisions. The advisor trains itself on historical data, adapts to changing market conditions, and is equipped with a built-
KNN Pattern Hunter
Aleksandr Zavolskov
KNN Pattern Hunter 2.0 An automated trading advisor that learns from market history and adapts to volatility changes Short Description KNN Pattern Hunter 2.0 is a fully automated trading robot for MetaTrader 4 built on the K-Nearest Neighbors (KNN) machine learning algorithm. The advisor analyzes historical market patterns and finds the most similar past situations to predict future price movement. The key innovation in version 2.0 is adaptive rolling Z-normalization : the robot continuously rec
Filtrer:
Aucun avis
Répondre à l'avis