Use TL494 that has a reference and feedback loop to stabilize the voltage.
http://www.ti.com/lit/an/slva001e/slva001e.pdf
If you can use microcontroller, use its PWM channel and ADC to achieve this.
Any MCU can be used. For example, AtTiny series from Atmel.
You need to program the MCU with what you want.
MCU can also help you achieve temperature compensation of the charging voltage and many other things e.g. over-voltage and over-current protection.
Contrary to what has been said above, 555 is not unreliable and not inefficient.
It can be used as a regulated booster than a simple booster (as it is in your case) but you need to modify your design.
In your current design add a fixed reference voltage source and an op-amp. Let op-amp compare the voltage at the output to the reference voltage. If reference voltage is 1V, you need to divide the output voltage by 14.4 (using voltage divider network) and then let op-amp compare the voltages. Output of the op-amp should manipulate (increase or decrease) the duty factor so that the output voltage is fixed at 14.4V regardless of load or input voltage variations.
It takes almost 2 mA just to charge and discharge the gate of your MOSFET. You're also wasting about 5 mA in R1, since it is grounded through pin 7 about half the time. Your voltage feedback divider is drawing about 1 mA from the high-voltage rail, which translates to more than 20 mA at the input.
There's a problem with using a 555 to drive a large MOSFET: The limited output current of the 555 means that the MOSFET can't switch quickly from full-off to full-on and back again. It spends a lot of time (relatively speaking) in a transition region, in which it dissipates a significant amount of your input power instead of delivering that power to the output. The MOSFET has a total gate charge of 63 nC, and the 555 has a maximum output current of about 200 mA, which means it takes a minimum of 63 nC / 200 mA = 315 ns to charge or discharge the gate. If you're using a CMOS 555, the output current is much less and the switching time is correspondingly longer.
If you add a gate driver chip between the 555 and the MOSFET (one that's capable of peak currents of 1-2A), you'll see a marked increase in overall efficiency. A real boost controller chip will often have such drivers built in.
If you're serious about developing switchmode power converters, you definitely need to get an oscilloscope so that you can see these effects for yourself.
That regulator design is also rather crappy for another reason. The power through a boost mode converter is regulated by varying the duty cycle of the switching element. In this circuit, the feedback is created by using a transistor to pull down on the control voltage node of the 555, which reduces the upper switching threshold. However, because of the way the 555 is constructed, this also reduces the lower switching threshold by a proportional amount. This means that the change in duty cycle as the ouptut voltage rises is much less than you might otherwise think. It has a bigger effect on the frequency of the output pulses, but this isn't relevant. Again, switching to a proper boost controller chip would solve this problem.
By the way, the "regulator" part of the circuit is NOT using the input voltage as its reference, it's using the forward voltage of Q1's B-E junction as its reference.
As Spehro points out, a 100 µH inductor at a switching frequency of 30 kHz — nominal on time = 16 µs — with a 9V source is going to reach a peak current of 1.44 A. This is really abusing the hell out of a 9V battery, not to mention the I2R losses in both the inductor and the MOSFET. This is also uncomfortably close to the saturation current of the inductor, which only exacerbates the losses.
Best Answer
The simplest solution would be to use a 1:10 transformer to transform the 5V swing of the 555 output to about 50V at the transducer. Put a capacitor (about 0.1 µF) between the 555 and the transformer to eliminate the DC bias.
However, you should note that 1.66 MHz is rather high for a 555 in the first place, and it might not be able to drive much power at that frequency.
Also, you'll probably get better results if you can get the duty cycle of the waveform closer to 50%.