Statsmodel ARIMA prediction mismatch

Question

I have written the below code to forecast data using ARIMA of stats models,but my results are not matching with the actual data and the predicted values become almost constant after first few predictions giving a straight horizontal line on graph.

And if the prediction is for 2nd differencing order because i have used d=2,how can i get the prediction for original data for the same model.

arima_mod = sm.tsa.ARIMA(df, (1,2,0)).fit()
print(arima_mod.params)
print(arima_mod.summary())
predict_workshop = arima_mod.predict('2011-04-01', '2011-05-30',dynamic=True)
print(predict_workshop)

Actual data

2011-04-01      356.839  
2011-04-02      363.524  
2011-04-03      332.864  
2011-04-04      336.228  
2011-04-05      264.749  
2011-04-06      321.212  
2011-04-07      384.382  
2011-04-08      273.250  
2011-04-09      307.062  
2011-04-10      326.247  
2011-04-11      222.521  
2011-04-12      135.326  
2011-04-13      374.953  
2011-04-14      329.583  
2011-04-15      358.853  
2011-04-16      343.169  
2011-04-17      312.086  
2011-04-18      339.302  
2011-04-19      300.534  
2011-04-20      367.166  
2011-04-21      178.670  
2011-04-22      320.823  
2011-04-23      349.995  
2011-04-24      323.120  
2011-04-25      331.665  
2011-04-26      352.993  
2011-04-27      359.253  
2011-04-28      308.281  
2011-04-29      329.357  
2011-04-30      301.873  

predicted value

2011-04-01   -50.693560  
2011-04-02    30.715553  
2011-04-03   -19.081318  
2011-04-04    11.378766  
2011-04-05    -7.253263  
2011-04-06     4.143701  
2011-04-07    -2.827670  
2011-04-08     1.436625  
2011-04-09    -1.171787  
2011-04-10     0.423744  
2011-04-11    -0.552221  
2011-04-12     0.044764  
2011-04-13    -0.320404  
2011-04-14    -0.097036  
2011-04-15    -0.233667  
2011-04-16    -0.150092  
2011-04-17    -0.201214  
2011-04-18    -0.169943  
2011-04-19    -0.189071  
2011-04-20    -0.177371  
2011-04-21    -0.184528  
2011-04-22    -0.180150  
2011-04-23    -0.182828  
2011-04-24    -0.181190  
2011-04-25    -0.182192  
2011-04-26    -0.181579  
2011-04-27    -0.181954  
2011-04-28    -0.181724  
2011-04-29    -0.181865  
2011-04-30    -0.181779  

Answer 1

A little hint. You should include a dummy intervention variable at period 11,12, 21. There is no need to double difference this model. Just an intercept and 3 intervention variables would work. Y(T) = 332.20
+[X1(T)][(- 196.87 )] :PULSE 12 +[X2(T)][(- 153.53 )] :PULSE 21 +[X3(T)][(- 109.68 )] :PULSE 11 + + [A(T)]