How can reduce improve my ANN accuracy and reduce overfitting?

Question

My ANN model produces classic overfitting characteristics, producing high R2 values (90-99%) but low accuracy scores (10-40%). I'm currently inputting 28655 data entries, using 8 input features to predict solar radiation.

My model is as follows:

df = pd.read_csv (r'D:\ABC Masters Diss\Penarth 1 year.csv')
print (df)
df[:5]
df.info()
#Features
X = df[["Cloud Coverage", "Probability of Precipitation", "Dry Bulb Temp", "Dew Point Temp", "Relative Humidity", "Apparent Temp", "Time of Day", "Month"]]
#Target
Y = df["Solar Radiation"]
X_train, X_test, Y_train, Y_test = train_test_split(X, 
                                                    Y, 
                                                    test_size = 0.2)
scaler = StandardScaler()
X_train_std = scaler.fit_transform(X_train)
X_test_std = scaler.transform(X_test)
#Initialise the ANN
model = Sequential()
#Input layer and hidden layer
model.add(Dense(50, activation= 'relu', input_dim = 8))
#second layer
model.add(Dense(units = 50, activation = 'relu'))
#output layer
model.add(Dense(units = 1))
model.compile(optimizer = 'adam', loss = 'mean_squared_error', metrics = ['accuracy'])
model.fit(X_train_std, Y_train, batch_size = 10, epochs = 10)
pred_a1 = model.predict(X_test_std)
pd.DataFrame({"Actual": Y_test, "Predicted": pred_a1.flatten()})[:5]

I've tried altering the architecture, no. of layers & neurons etc. I've also tried changing the activation, adding drop out and kernel constraints and altering the learning rate(see below). Any ideas on how to improve this?

#Input layer and hidden layer
model.add(Dense(50, activation= 'relu', input_dim = 8, kernel_constraint=MaxNorm(3)))
model.add(Dropout(0.2))
#second layer
model.add(Dense(units = 50, activation = 'relu', kernel_constraint=MaxNorm(3)))
model.add(Dropout(0.2))
#output layer
model.add(Dense(units = 1, activation = 'relu'))
sgd = SGD(learning_rate=0.01, momentum=0.9)
model.compile(optimizer = 'adam', loss = 'mean_squared_error', metrics = ['accuracy'])
model.fit(X_train_std, Y_train, batch_size = 10, epochs = 100)