Python 股票预测分析系统-LSTM

基于上一个版本的arima 算法进行改进，该次修改只修改了预测部分的算法从，arima 算法改为lstm 算法

现在的预测结果：

之前的预测结果：

从结果的图片上可以看出LSTM 更贴近真实的数据曲线，准确性也相比ARIMA 算法要好，但是LSTM 的运行时间较长：

而AIRMA 的运行时间很短。在代码的复杂度上LSTM 也是要高于ARIMA
 

def get_stock_hist_realtimedeal(code, level):
    """
    历史分时数据接口
    :param code: 股票代码
    :param level: 分时级别
    :return:
    """
    try:
        if level in ['5', '15', '30', '60', 'Day', 'Week', 'Month', 'Year']:
            return resp(data=StockApi.get_stock_hist_realtimedeal(code, level))
        elif level == 'Forecast':
            data = pandas.DataFrame(StockApi.get_stock_hist_realtimedeal(code, 'Day'))
            data.to_csv('./data/Day.csv', index=False)
            data = pd.read_csv('./data/Day.csv', index_col=0, parse_dates=[0])
            stock_week = data['o'].resample('W').mean().dropna()
            stock_train = stock_week['2020':'2023'].dropna()
            stock_data = stock_week['2023':'2024'].dropna()

            scaler = MinMaxScaler(feature_range=(0, 1))
            stock_train_scaled = scaler.fit_transform(stock_train.values.reshape(-1, 1))

            # Preparing dataset for training
            X_train = []
            y_train = []
            for i in range(60, len(stock_train_scaled)):
                X_train.append(stock_train_scaled[i - 60:i, 0])
                y_train.append(stock_train_scaled[i, 0])

            X_train, y_train = np.array(X_train), np.array(y_train)
            X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[1], 1))

            # Build the LSTM model
            model = Sequential()
            model.add(LSTM(units=50, return_sequences=True, input_shape=(X_train.shape[1], 1)))
            model.add(LSTM(units=50))
            model.add(Dense(1))

            model.compile(optimizer='adam', loss='mean_squared_error')
            model.fit(X_train, y_train, epochs=100, batch_size=32)

            # Forecasting
            inputs = stock_week[len(stock_week) - len(stock_data) - 60:].values
            inputs = inputs.reshape(-1, 1)
            inputs = scaler.transform(inputs)

            X_test = []
            for i in range(60, len(inputs)):
                X_test.append(inputs[i - 60:i, 0])
            X_test = np.array(X_test)
            X_test = np.reshape(X_test, (X_test.shape[0], X_test.shape[1], 1))

            predicted_stock_price = model.predict(X_test)
            predicted_stock_price = scaler.inverse_transform(predicted_stock_price)

            # Converting predictions to JSON
            pred_list = predicted_stock_price.flatten().tolist()
            predictions = [{'date': str(index.date()), 'val': format(val, '.2f')} for index, val in
                           zip(stock_data.index, pred_list)]
            if isinstance(stock_data.index, pd.DatetimeIndex):
                stock_data.index = stock_data.index.strftime('%Y-%m-%d')
            if isinstance(stock_data, pd.Series):
                origin_data = [{'date': index, 'val': format(value, '.2f')} for index, value in stock_data.items()]
            else:
                origin_data = [{'date': index, 'val': format(value, '.2f')} for index, value in
                               stock_data.itertuples(index=True, name=None)]
            predictions_data = [{'date': str(item['date']), 'val': item['val']} for item in predictions]

            obj = {'origin': json.dumps(origin_data), 'pred': json.dumps(predictions_data)}
            print(obj)
            # Generating the response object
            return resp(data=json.dumps(obj))
        else:
            return resp(ResponseEnum.STOCK_LEVEL_INVALID.value['code'], ResponseEnum.STOCK_LEVEL_INVALID.value['msg'])
    except Exception as e:
        print(e)
        return resp(ResponseEnum.OUTER_INTERFACE_EXCEPTION.value['code'],
                    ResponseEnum.OUTER_INTERFACE_EXCEPTION.value['msg'])

演示视频地址：
Python股票预测分析系统-LSTM_哔哩哔哩_bilibili