QString QDjangoCompiler::databaseColumn(const QString &name)
{
QDjangoMetaModel model = baseModel;
QString modelPath;
QString modelRef = referenceModel(QString(), &model);
QStringList bits = name.split(QLatin1String("__"));
while (bits.size() > 1) {
const QByteArray fk = bits.first().toLatin1();
QDjangoMetaModel foreignModel;
if (!model.foreignFields().contains(fk)) {
// this might be a reverse relation, so look for the model
// and if it exists continue
foreignModel = QDjango::metaModel(fk);
if (!foreignModel.isValid())
break;
reverseModelRefs[bits.first()] = foreignModel.primaryKey();
} else {
foreignModel = QDjango::metaModel(model.foreignFields()[fk]);
}
// store reference
if (!modelPath.isEmpty())
modelPath += QLatin1String("__");
modelPath += bits.first();
modelRef = referenceModel(modelPath, &foreignModel);
model = foreignModel;
bits.takeFirst();
}
const QDjangoMetaField field = model.localField(bits.join(QLatin1String("__")).toLatin1());
return modelRef + QLatin1Char('.') + driver->escapeIdentifier(field.column(), QSqlDriver::FieldName);
}
void pid_update(neural_state_t *n_s, float rin, int encoder_counter, int command_distance_count){
n_s->rf_out = referenceModel(rin, n_s->rf_out_1);
n_s->e = n_s->rf_out - (float)encoder_counter;
n_s->rf_out_2 = n_s->rf_out_1;
n_s->rf_out_1 = n_s->rf_out;
n_s->desire = n_s->desire +(int)n_s->rf_out;
// Incremental PID
n_s->xc[0] = n_s->e - n_s->e_1;
n_s->xc[1] = n_s->e;
n_s->xc[2] = n_s->e - (2 * n_s->e_1) + n_s->e_2;
// Calculate the output of PID controller
n_s->du = n_s->kp * n_s->xc[0] + n_s->ki * n_s->xc[1] + n_s->kd * n_s->xc[2];
n_s->u = n_s->u_1 + n_s->du;
if (n_s->u < 0)
{
n_s->u = 0;
}else if(n_s->u > 110) {
n_s->u = 110;
}
// update yout(k-1) u(k-1)
n_s->yout_1 = n_s->x[1];
n_s->u_2 = n_s->u_1;
n_s->u_1 = n_s->u;
// update e(k-1) e(k-2)
n_s->e_2 = n_s->e_1;
n_s->e_1 = n_s->e;
}
QStringList QDjangoCompiler::fieldNames(bool recurse, QDjangoMetaModel *metaModel, const QString &modelPath)
{
QStringList columns;
if (!metaModel)
metaModel = &baseModel;
// store reference
const QString tableName = referenceModel(modelPath, metaModel);
foreach (const QDjangoMetaField &field, metaModel->localFields())
columns << tableName + QLatin1Char('.') + driver->escapeIdentifier(field.column(), QSqlDriver::FieldName);
if (!recurse)
return columns;
// recurse for foreign keys
const QString pathPrefix = modelPath.isEmpty() ? QString() : (modelPath + QLatin1String("__"));
foreach (const QByteArray &fkName, metaModel->foreignFields().keys()) {
QDjangoMetaModel metaForeign = QDjango::metaModel(metaModel->foreignFields()[fkName]);
columns += fieldNames(recurse, &metaForeign, pathPrefix + QString::fromLatin1(fkName));
}
return columns;
}
QString QDjangoCompiler::databaseColumn(const QString &name)
{
QDjangoMetaModel model = baseModel;
QString modelName;
QString modelPath;
QString modelRef = referenceModel(QString(), &model, false);
QStringList bits = name.split(QLatin1String("__"));
while (bits.size() > 1) {
const QByteArray fk = bits.first().toLatin1();
QDjangoMetaModel foreignModel;
bool foreignNullable = false;
if (!modelPath.isEmpty())
modelPath += QLatin1String("__");
modelPath += bits.first();
if (!model.foreignFields().contains(fk)) {
// this might be a reverse relation, so look for the model
// and if it exists continue
foreignModel = QDjango::metaModel(fk);
QDjangoReverseReference rev;
const QMap<QByteArray, QByteArray> foreignFields = foreignModel.foreignFields();
foreach (const QByteArray &foreignKey, foreignFields.keys()) {
if (foreignFields[foreignKey] == baseModel.className()) {
rev.leftHandKey = foreignModel.localField(foreignKey + "_id").column();
break;
}
}
if (rev.leftHandKey.isEmpty()) {
qWarning() << "Invalid field lookup" << name;
return QString();
}
rev.rightHandKey = foreignModel.primaryKey();
reverseModelRefs[modelPath] = rev;
} else {
void neural_update(neural_state_t *n_s, float rin, int encoder_counter, int command_distance_count){
if (car_state != CAR_STATE_IDLE)
{
int i = 0, j = 0;
// 1. RBFNN Output
n_s->ynout = 0;
for ( i = 0; i < neuralNumber; ++i)
{
n_s->norm_c_2[i] = pow2((n_s->x[0] - n_s->c[0][i]), 2) + pow2((n_s->x[1] - n_s->c[1][i]), 2) + pow2((n_s->x[2] - n_s->c[2][i]), 2);
float tmp = (((-0.5) * n_s->norm_c_2[i]));
tmp = tmp/ pow2(n_s->b[i], 2);
n_s->h[i] = exponential(tmp);
n_s->ynout = n_s->ynout + (n_s->h[i])*(n_s->w[i]);
}
n_s->ynout_sum += n_s->ynout;
// 2 Get the motor speed of last clock cycle, and calculate the error of rbf
n_s->erbf = encoder_counter - n_s->ynout;
n_s->erbf_record[4] = n_s->erbf_record[3];
n_s->erbf_record[3] = n_s->erbf_record[2];
n_s->erbf_record[2] = n_s->erbf_record[1];
n_s->erbf_record[1] = n_s->erbf_record[0];
n_s->erbf_record[0] = abs2(n_s->erbf);
n_s->erbf_avg = (n_s->erbf_record[0] + n_s->erbf_record[1] + n_s->erbf_record[2] + n_s->erbf_record[3] + n_s->erbf_record[4])/5.0;
// 3. Update w of RBFNN
for ( i = 0; i < neuralNumber; ++i)
{
float tmp = n_s->erbf * n_s->h[i];
n_s->dw[i] = n_s->eta * tmp;
n_s->w_1[i] = n_s->w[i];
n_s->w[i] = n_s->w_1[i] + n_s->dw[i];
}
// 4. Update bj
for ( i = 0; i < neuralNumber; ++i)
{
float tmp = n_s->eta * n_s->erbf;
tmp = tmp * n_s->w[i];
tmp = tmp * n_s->h[i];
tmp = tmp * n_s->norm_c_2[i];
tmp = tmp / pow2(n_s->b[i], 3);
n_s->db[i] = tmp;
n_s->b_1[i] = n_s->b[i];
n_s->b[i] = n_s->b_1[i] + n_s->db[i];
}
// 5. Update Cj
for ( i = 0; i < neuralNumber; ++i)
{
for ( j = 0; j < centerNumber; ++j)
{
float tmp = n_s->eta * n_s->erbf;
tmp = tmp * n_s->w[i];
tmp = tmp * n_s->h[i];
tmp = tmp * (n_s->x[j] - n_s->c[j][i]);
tmp = tmp / pow2(n_s->b[i], 2);
n_s->dc[j][i] = tmp;
n_s->c_1[j][i] = n_s->c[j][i];
n_s->c[j][i] = n_s->c_1[j][i] + n_s->dc[j][i];
}
}
// 6. Calculate Jacobian
n_s->yu = 0;
for ( i = 0; i < neuralNumber; ++i)
{
float tmp = n_s->w[i] * n_s->h[i];
float tmp2 = (-1) * n_s->x[0];
tmp = tmp * (tmp2 + n_s->c[0][i]);
tmp = tmp / pow2(n_s->b[i], 2);
n_s->yu = n_s->yu + tmp;
}
n_s->dyu = n_s->yu;
// 6.2 Calculate the error with reference model
n_s->rf_out = referenceModel(rin, n_s->rf_out_1);
n_s->e = n_s->rf_out - (float)encoder_counter;
n_s->total_err += abs2(n_s->e);
n_s->rf_out_2 = n_s->rf_out_1;
n_s->rf_out_1 = n_s->rf_out;
n_s->desire = n_s->desire +(int)n_s->rf_out;
// 8. Incremental PID
n_s->xc[0] = n_s->e - n_s->e_1;
n_s->xc[1] = n_s->e;
n_s->xc[2] = n_s->e - (2 * n_s->e_1) + n_s->e_2;
//int tmp_erbf = (int)(abs2(erbf)*100);
int tmp_erbf = (command_distance_count - n_s -> ynout_sum)/command_distance_count;
tmp_erbf = tmp_erbf * 100;
if (tmp_erbf < 10)
{
n_s -> erbf_correct_times ++;
}
if ((tmp_erbf < 10) && (encoder_counter > 1) && (n_s->erbf_correct_times >30 && n_s->erbf_avg < 1)){
n_s->erbf_correct_times ++;
float kp_add = n_s->eta * n_s->e;
kp_add = kp_add * n_s->dyu;
kp_add = kp_add * n_s->xc[0];
float ki_add = n_s->eta * n_s->e;
ki_add = ki_add * n_s->dyu;
ki_add = ki_add * n_s->xc[1];
float kd_add = n_s->eta * n_s->e;
kd_add = kd_add * n_s->dyu;
kd_add = kd_add * n_s->xc[2];
// 10. update kp(k-1) ki(k-1) kd(k-1)
n_s->kp_1 = n_s->kp;
n_s->ki_1 = n_s->ki;
n_s->kd_1 = n_s->kd;
// 9. Update the parameter of PID controller
if (n_s->stop_tune == 0 && car_state == CAR_STATE_MOVE_FORWARD){
n_s->kp = n_s->kp_1 + kp_add;
n_s->ki = n_s->ki_1 + ki_add;
// n_s->kd = n_s->kd_1 + kd_add;
if (n_s->kp <= 0)
{
n_s->kp = 0;
}
if (n_s->ki <= 0)
{
n_s->ki = 0;
}
if (n_s->kd <= 0)
{
//n_s->kd = 0;
}
}
}
// 11. Calculate the output of PID controller
n_s->du = n_s->kp * n_s->xc[0] + n_s->ki * n_s->xc[1] + n_s->kd * n_s->xc[2];
n_s->u = n_s->u_1 + n_s->du;
if (n_s->u < 0)
{
n_s->u = 0;
}else if(n_s->u > 110) {
n_s->u = 110;
}
// 12. update yout(k-1) u(k-1)
n_s->yout_1 = n_s->x[1];
n_s->u_2 = n_s->u_1;
n_s->u_1 = n_s->u;
// 13. update e(k-1) e(k-2)
n_s->e_2 = n_s->e_1;
n_s->e_1 = n_s->e;
// 14. update input of RBFNN
n_s->x[0] = n_s->du;
n_s->x[1] = (float)encoder_counter;
n_s->x[2] = n_s->yout_1;
}
}
