Intelligent control method and system of MOPA laser

Abstract

The application relates to the technical field of laser control, in particular to an intelligent control method and system of an MOPA laser. The method comprises the following steps: constructing a target training data set and a one-dimensional convolutional neural network model; training the one-dimensional convolutional neural network model based on the target training data set, combining a target loss function and a back propagation algorithm, and obtaining a target model; acquiring current operation parameters of the MOPA laser; inputting the current operation parameters into the target model, and outputting target control parameters from the target model; the target control parameters comprise working currents of pump modules in the MOPA laser and working temperatures of doped optical fibers; adjusting the working currents of the pump modules in the MOPA laser and the working temperatures of the doped optical fibers based on the target control parameters; the working temperatures of the doped optical fibers comprise a plurality of different target temperatures, and each target temperature corresponds to a different position on the doped optical fiber. In this way, the problems caused by nonlinear effects under high-power pumping can be reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of laser control, in particular to an intelligent control method and system of a MOPA laser. BACKGROUND

[0002] At present, in the tunable high-power single-frequency fiber laser designed by using a MOPA (Master Oscillator Power-Amplifier) architecture, a power amplifier serves as a core component and is responsible for amplifying a low-power single-frequency signal generated by a seed source to a high-power level, so as to realize efficient energy output.

[0003] However, under high-power operating conditions, nonlinear effects become the main bottleneck restricting performance improvement. This is because the high optical power density inside the amplifier can induce various nonlinear processes, such as stimulated Brillouin scattering (SBS) and stimulated Raman scattering (SRS). These effects can cause a part of the forward transmission pump light to be scattered into backward transmission nonlinear light, thereby causing energy loss. When the pump light power increases to a critical value (SBS threshold), the nonlinear effect will be sharply enhanced, so that the output power of the amplifier no longer increases linearly with the increase of the pump power, but most of the energy is reflected back to the seed source, causing power saturation phenomenon. As can be seen, nonlinear effects not only seriously restrict the further improvement of output power, but also cause spectral characteristics to deteriorate, such as spectral width broadening and side mode suppression ratio (SMSR) decrease, and cause system instability, including power fluctuation and phase noise increase, ultimately affecting the overall performance and reliability of the laser.

[0004] Therefore, it is urgent to propose a regulation scheme of a MOPA laser to solve the above problems. SUMMARY

[0005] In order to solve the above problems, the embodiments of the present application provide an intelligent control method and system of a MOPA laser, which reduces the problems caused by nonlinear effects under high-power pumping.

[0006] To achieve the above objectives, in a first aspect, an intelligent control method for a MOPA laser is provided, comprising: constructing a target training dataset and a one-dimensional convolutional neural network model, wherein the target training dataset is constructed based on the historical operating parameters of the MOPA laser; training the one-dimensional convolutional neural network model based on the target training dataset, combining a target loss function and a backpropagation algorithm, to obtain a target model; obtaining the current operating parameters of the MOPA laser; wherein the current operating parameters include the current target output power, the current actual output power, the current pump light power of the pump module in the MOPA laser, the current backscattered light power in the MOPA laser, the current ambient temperature, and multiple current temperatures of the doped fiber in the MOPA laser; each current temperature is the temperature at a different location on the doped fiber; inputting the current operating parameters into the target model, and the target model outputting target control parameters; the target control parameters include the operating current of the pump module in the MOPA laser and the operating temperature of the doped fiber; adjusting the operating current of the pump module in the MOPA laser and the operating temperature of the doped fiber based on the target control parameters, wherein the operating temperature of the doped fiber includes multiple different target temperatures, each target temperature corresponding to a different location on the doped fiber.

[0007] In one possible implementation, constructing the target training dataset includes: obtaining multiple sets of historical operating parameters of the MOPA laser; the historical operating parameters include the historical target output power of the MOPA laser. Historical actual output power of MOPA lasers Historical pump power of the pump module in the MOPA laser Historical backscattered light power in MOPA lasers Historical environmental temperature And the first historical temperature of doped fiber in MOPA lasers. Second highest temperature in history and the third highest temperature in history Based on multiple sets of historical operating parameters, multiple initial input feature vector samples are constructed; these initial input feature vector samples are input into the initial model, and the initial model outputs the control parameters at time i; the backscattered light power at time i is obtained. With actual output power The operation of the MOPA laser was controlled using control parameters; the backscattered power at time i+1 was obtained. and actual output power Based on backscattered light power Actual output power Backscattered light power and actual output power , determine whether the initial input feature vector sample is a target input feature vector, the target training data set including the target input feature vector; and normalize each target input feature vector in the target training data set.

[0008] In a possible implementation, based on a plurality of sets of historical operation parameters, a plurality of initial input feature vector samples are constructed, including: performing data cleaning on the plurality of sets of historical operation parameters to obtain a plurality of sets of historical operation parameters whose deviations satisfy preset requirements; sorting the plurality of sets of historical operation parameters whose deviations satisfy the preset requirements in an order of acquisition time to obtain an initial data matrix; and selecting a plurality of initial input feature vector samples by using a preset window and sliding on the initial data matrix according to a preset sliding step.

[0009] In a possible implementation, whether the initial input feature vector sample is a target input feature vector is determined according to the backscattering light power , the actual output power , the backscattering light power , and the actual output power , and includes: calculating a first ratio at i moment according to and according to a preset formula; calculating a second ratio at i+1 moment according to and according to a preset formula; and determining that the initial input feature vector sample is a target input feature vector in a case where the second ratio satisfies preset requirements; the target training data set including the target input feature vector; wherein the preset requirements include that the second ratio is less than the first ratio, and / or the second ratio is less than a proportion threshold.

[0010] In a possible implementation, the preset formula includes:

[0011] ;

[0012] wherein, includes and , is the first ratio, is the second ratio; includes and , includes and .

[0013] In a possible implementation, the target loss function includes:

[0014] ; and

[0015] in, This is the number of samples in the current training batch. These are the weighting coefficients for power tracking loss. These are the weighting coefficients for backward nonlinear optical power suppression loss. It is the first The actual output power of each target input feature vector It is the first The target output power of a target input feature vector. For the first The first ratio is calculated from the target input feature vectors.

[0016] In one possible implementation, the control parameters include a first target temperature, a second target temperature, a third target temperature, and a pump target current; the first target temperature is the first temperature of the doped fiber in the MOPA laser. The target temperature and the second target temperature are the second temperatures of the doped fiber in the MOPA laser. The target temperature, the third target temperature is the third temperature of the doped fiber in the MOPA laser. The target temperature is defined as the target temperature, and the target pump current is defined as the target current of the pump module. The one-dimensional convolutional neural network model includes a first constraint, a second constraint, a third constraint, and a fourth constraint. The first constraint includes: the first target temperature, the second target temperature, and the third target temperature corresponding to the target control parameters are all within a preset range. The second constraint includes: the minimum temperature difference between the first temperature and the second temperature is defined as follows: The minimum temperature difference between the second and third temperatures is , , The third constraint includes: , The maximum allowed gradient; the fourth constraint includes: ,in, This is the minimum current of the pump module. This is the maximum current of the pump module.

[0017] In one possible implementation, before obtaining the current operating parameters of the MOPA laser, the process includes: lightweighting the target model; deploying the target model on a microcontroller; and initializing the target model.

[0018] In one possible implementation, lightweighting the target model includes: converting a first parameter format of the target model to a second parameter format, wherein the second parameter format occupies less space than the first parameter format; and / or pruning the target model.

[0019] In a second aspect, an intelligent control system of a MOPA laser is provided, comprising: a seed source module, a first coupler, an amplifier module, a regulation module, and a control module; the amplifier module comprises a pump module, a first combiner, and a doped optical fiber; the first combiner comprises a first input end, a second input end, and a first output end; the seed source module is in optical communication with the first input end, the pump module is in optical communication with the second input end, and the doped optical fiber is in optical communication with the first output end; the seed source module is configured to output seed laser; the pump module is configured to output pump light; the first combiner is configured to combine the seed laser and the pump light according to a first combining ratio to form first target laser, and output the first target laser to the doped optical fiber; the doped optical fiber is configured to oscillate the first target laser to form second target laser; the first coupler is arranged on an output light path of the seed source module, and comprises a second output end configured to output backscattered light generated by the amplifier module; the control module is in communication connection with the regulation module, and is configured to: construct a target training data set and a one-dimensional convolutional neural network model, the target training data set being constructed based on historical running parameters of the MOPA laser; train the one-dimensional convolutional neural network model based on the target training data set, a target loss function, and a back propagation algorithm, to obtain a target model; obtain current running parameters of the MOPA laser; wherein the current running parameters comprise a current target output power of the MOPA laser, a current actual output power, a current pump light power of the pump module in the MOPA laser, a current backscattered light power in the MOPA laser, a current environmental temperature, and a plurality of current temperatures of the doped optical fiber in the MOPA laser; each current temperature is a temperature at a different position of the doped optical fiber; input the current running parameters into the target model, and output target regulation parameters from the target model; the target regulation parameters comprise a working current of the pump module in the MOPA laser and a working temperature of the doped optical fiber; the regulation module comprises a first regulation submodule and a second regulation submodule; the first regulation submodule is configured to adjust the working current of the pump module in the MOPA laser based on the target regulation parameters; the second regulation submodule is configured to adjust the working temperature of the doped optical fiber based on the target regulation parameters; the number of the second regulation submodules is a plurality, and the plurality of second regulation submodules are arranged along the doped optical fiber in sequence, and each second regulation submodule is configured to regulate the working temperature at a different position of the doped optical fiber.

[0020] The intelligent control method and system of the MOPA laser provided by the embodiments of the present application can dynamically adjust the temperature distribution of each section of the doped optical fiber and the working current of the pump module, thereby maximizing the suppression of the influence of nonlinear effects while ensuring that the output power of the laser reaches the target value accurately, to meet the requirements of high-precision laser processing, fiber communication, and other scenarios for the performance of the laser. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0022] Figure 1 is a schematic diagram of an intelligent control system of a MOPA laser provided by the embodiments of the present application;

[0023] Figure 2 is a schematic diagram of a second regulation and control sub-module provided by the embodiments of the present application;

[0024] Figure 3 is another schematic diagram of a second regulation and control sub-module provided by the embodiments of the present application;

[0025] Figure 4 is a flowchart of an intelligent control method of a MOPA laser provided by the embodiments of the present application;

[0026] Figure 5 is a flowchart of an intelligent control method of a MOPA laser provided by the embodiments of the present application;

[0027] Figure 6 is an architecture schematic diagram of a target model provided by the embodiments of the present application;

[0028] Figure 7 is another flowchart of an intelligent control method of a MOPA laser provided by the embodiments of the present application.

[0029] Reference signs:

[0030] 100, an intelligent control system of a MOPA laser; 10, a seed source module; 11, a seed source laser; 12, a seed source controller; 20, a first coupler; 201, a third input end; 202, a second output end; 203, a third output end; 30, an amplifier module; 31, a pump module; 32, a first beam combiner; 321, a first input end; 322, a second input end; 323, a first output end; 33, a doped optical fiber; 40, a regulation and control module; 41, a first regulation and control sub-module; 42, a second regulation and control sub-module; 421, a heat sink; 4211, an optical fiber slot; 422, a thermoelectric cooler; 423, a temperature sensor; 424, a liquid cooling radiator; 50, a control module; 501, a first control sub-module; 502, a second control sub-module; 60, a first isolator; 70, a first silicon photodiode; 80, a second beam combiner; 90, a second silicon photodiode. DETAILED DESCRIPTION

[0031] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly described. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments of the present application, other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0032] To solve the above technical problems, the embodiments of the present application provide an intelligent control method and system of a MOPA laser, which reduces the problems caused by nonlinear effects under high-power pumping.

[0033] Figure 1 FIG. 1 is a schematic diagram of an intelligent control system of a MOPA laser provided by the embodiments of the present application.

[0034] In some embodiments, as shown in FIG. 1, the intelligent control system 100 of the MOPA laser includes a seed source module 10, a first coupler 20, an amplifier module 30, a regulation module 40 and a control module 50. Figure 1

[0035] The seed source module 10 is configured to output signal light. The signal light enters the amplifier module 30 through the first coupler 20. The amplifier module 30 is configured to amplify the signal light to obtain target output laser and output the target output laser. The control module 50 is configured to receive working parameters of the MOPA laser and output intelligent control instructions of the MOPA laser, so as to make the regulation module 40 regulate the working state of the MOPA laser according to the control instructions.

[0036] It can be understood that the MOPA laser includes the seed source module 10, the first coupler 20 and the amplifier module 30. The embodiments of the present application add detection devices and controller devices to the MOPA laser on the basis of the MOPA laser, so as to regulate the working state of the MOPA laser.

[0037] For example, the amplifier module 30 includes a pump module 31, a first beam combiner 32 and a doped optical fiber 33. The first beam combiner 32 includes a first input end 321, a second input end 322 and a first output end 323. The seed source module 10 is in optical communication with the first input end 321, the pump module 31 is in optical communication with the second input end 322, and the doped optical fiber 33 is in optical communication with the first output end 323. The pump module 31 is configured to output pump light. The first beam combiner 32 is configured to combine the signal light and the pump light according to a first beam combination ratio to form a first target laser, and output the first target laser to the doped optical fiber 33. The doped optical fiber 33 is configured to oscillate the first target laser to form a second target laser.

[0038] ​For example, the first coupler 20 is disposed on the output light path of the seed source module 10, and the first coupler 20 includes a third input end 201, a second output end 202, and a third output end 203. The first coupler 20 is configured to input signal light into the inside of the first coupler 20 through the third input end 201 and output the signal light through the third output end 203. The first coupler 20 is also configured to output the generated backscattered light of the amplifier module 30 through the second output end 202.

[0039] For example, the first coupler 20 is a 1% diagnostic tap coupler, which is a passive and precise optical signal splitter.

[0040] For example, the control module 50 is in communication connection with the regulation module 40, and the control module 50 is configured to execute an intelligent control method of the MOPA laser to cooperatively regulate the pump module 31 and the doped optical fiber 33 of the MOPA laser. By collecting the backscattered light signal in real time and feeding back to the control module 50, the target regulation parameters are dynamically adjusted, and the working current of the pump module 31 and the working stability of the doped optical fiber 33 are regulated respectively, so as to realize efficient suppression of nonlinear effects and thermal effects, thereby accurately controlling the power, waveform and stability of the output laser. Moreover, in the high-power output scene, the device can effectively improve the signal-to-noise ratio and beam quality of the laser, and ensure long-term stable operation of the system.

[0041] For example, the target regulation parameters include a first control parameter and a second control parameter. The regulation module 40 includes a first regulation sub-module 41 and a second regulation sub-module 42. The first regulation sub-module 41 is configured to adjust the working current of the pump module 31 in the MOPA laser based on the first control parameter. The second regulation sub-module 42 is configured to adjust the working temperature of the doped optical fiber 33 in the MOPA laser based on the second control parameter.

[0042] For example, the number of the second regulation sub-modules 42 is multiple, and the multiple second regulation sub-modules 42 are sequentially disposed along the doped optical fiber 33. Each second regulation sub-module 42 is configured to regulate the working temperature at different positions of the doped optical fiber 33.

[0043] In some embodiments, the seed source module 10 includes a seed source laser 11 and a seed source controller 12. The seed source controller 12 includes a seed source current driving circuit, a seed source temperature control circuit, and a seed source power detection circuit.

[0044] For example, the seed source current driving circuit proportionally integrates and differentiates the driving output current according to the set driving current and the current feedback signal. The seed source temperature control circuit proportionally integrates and differentiates the control current of the seed source thermoelectric cooler according to the set working temperature of the seed source laser 11 and the collected working temperature inside the seed source laser 11.

[0045] For example, the control precision of the current and the temperature is 1 ppb and 0.002 ℃ respectively. The long-term stable narrow pulse seed signal light is generated by the seed source laser 11 through the precise regulation of the seed source current driving circuit and the seed source temperature control circuit.

[0046] For example, the seed source laser 11 is an external cavity frequency stabilized semiconductor laser.

[0047] It can be understood that, in the proportional integration and differentiation adjustment of the current driving circuit and the temperature driving circuit, the working current and the working temperature of the current signal light are dynamically adjusted to the working current and the working temperature of the target signal light through the proportional integration and differentiation adjustment after the preset working current and the working temperature of the seed source module 10 are obtained.

[0048] In some embodiments, the intelligent control system 100 of the MOPA laser further includes a first isolator 60 arranged at the output end of the seed source laser. The first isolator 60 is configured to isolate the backscattered light generated by the amplifier module 30 in the subsequent optical path of the seed source module 10, so as to reduce the problem of the backscattered light returning along the optical path to damage the seed source module 10.

[0049] In some embodiments, the intelligent control system 100 of the MOPA laser further includes a first silicon photodiode 70. The 1% output end of the first coupler 20 is connected to the first silicon photodiode 70, so as to convert 1% of the backscattered light into an electrical signal and transmit the electrical signal to the control module 50 for monitoring and calculating the backscattered light power.

[0050] Specifically, for the power tunable laser, the signal light power generated by the seed source module 10 remains unchanged. When the signal light passes through the amplifier module 30, the photons are amplified in a “cloning” manner through the stimulated emission process, and the optical power is significantly increased. The process of amplifying the signal light power is to adjust the pump light power by controlling the pump current in the pump module 31, so as to achieve the purpose of adjusting the final output power.

[0051] In some embodiments, the pump module 31 includes a multimode pump. The first regulation sub-module 41 includes a multimode pump current driving circuit, a multimode pump temperature control circuit and a multimode pump power detection circuit.

[0052] The multi-mode pumping current driving circuit is configured to drive the multi-mode pumping current according to a current driving instruction sent by the control module 50 and a pump source power feedback signal The output current driving the multi-mode pumping is proportionally integrated and differentiated adjusted. The temperature control circuit is configured to control the temperature of the multi-mode pumping according to a set temperature control instruction and the collected actual temperature of the multi-mode pumping. The pump controller controls the current driving circuit and the temperature control circuit to control the pump light power and ensure the stability of the pump light power. In addition, the pump controller detects the amplifier pump power in real time through the power detection circuit .

[0053] In some embodiments, the intelligent control system 100 of the MOPA laser further comprises a second beam combiner 80, which is configured to combine the second target laser and the pump light.

[0054] For example, the second beam combiner 80 is a (1+1)×1 beam combiner, and the signal light is combined with the pump light through the (1+1)×1 pump beam combiner. The second beam combiner 80 has one signal light input port, one pump light input port and one laser output port. The signal light input port is connected to the signal light generated by the seed source module 10, and the pump light input port is connected to the pump light generated by the pump module 31. The output port is connected to a doped optical fiber 33, such as a double-clad erbium-ytterbium co-doped polarization maintaining optical fiber (EYDF PM 12 / 130 Nufern) as a gain medium. The final output laser passes through a second coupler, and the 0.1% output Tap end of the second coupler is connected to a second silicon photodiode 90.

[0055] For example, the control module 50 comprises an optical power detection module. The optical power detection module is connected to the first silicon photodiode 70 and the second silicon photodiode 90 respectively. The first silicon photodiode 70 is configured to convert the backscattering light signal into a first electrical signal and transmit it to the optical power detection module, and the second silicon photodiode 90 is configured to convert the final output laser (0.1%) signal of the laser into a second electrical signal and transmit it to the optical power detection module.

[0056] The optical power detection module is configured to identify the backscattering light power based on the first electrical signal The optical power detection module is further configured to identify the MOPA laser output power based on the second electrical signal .

[0057] In some embodiments, the control module 50 comprises a first control submodule 501 and a second control submodule 502, the first control submodule 501 is a control device for the MOPA laser itself, and the second control submodule 502 is an electronic device such as a computer, which can obtain the working parameters of the MOPA laser through the first control submodule 501, then transmit the working parameters to the second control submodule 502, the second control submodule 502 trains the initial model based on the obtained working parameters to obtain the target model, and after deploying the target model in the first control submodule 501, the first control submodule 501 can be directly used to regulate and control the operation of the laser. In this way, the second control submodule 502 can better adapt to the computing power required for model training, and after training the target model, it is deployed on the first control submodule 501 for running, so as to deploy intelligent algorithms on the MOPA laser, improve the intelligence of the MOPA laser control, and facilitate the integration and miniaturization of the device.

[0058] For example, the first control submodule 501 comprises an optical power detection module to detect the backscattering optical power and the actual output power of the laser.

[0059] It can be understood that the laser power index is mainly affected by the nonlinear effect (such as SBS), and the threshold simplified formula is as follows:

[0060]

[0061]

[0062] wherein, is the optical effective mode area, is the SBS gain coefficient, is the effective fiber length, is the absorption coefficient (unit: dB / m), and L is the physical length (unit: m) of the doped optical fiber 33.

[0063] In actual application, the value of a is determined by the material properties and doping concentration of the doped optical fiber 33, for example, the a of the double-clad erbium-ytterbium co-doped polarization maintaining optical fiber is usually in the range of 0.5~1.2 dB / m. By accurately measuring the physical length L and the absorption coefficient a of the doped optical fiber 33, the effective fiber length can be calculated, and then combined with the SBS gain coefficient and the optical effective mode area , the SBS threshold of the current system can be evaluated.

[0064] As can be seen, under the condition that the type of optical fiber is determined, in order to improve the SBS threshold, the fiber length of each device should be shortened as much as possible to reduce the value of .

[0065] Further, while reducing the length of the doped fiber 33, since the nonlinear effect in the doped fiber 33 is closely related to temperature, the embodiment of the present application applies a gradient temperature to the doped fiber 33 to be able to regulate the temperature of each position of the doped fiber 33, thereby setting a non-uniform and stable temperature distribution on the doped fiber 33, shifting the nonlinear gain peaks of different positions of the doped fiber 33, breaking the phase matching, broadening the gain bandwidth of the whole fiber, and reducing the peak gain, thereby effectively suppressing the nonlinear effect in the amplification process.

[0066] Specifically, the nonlinear frequency shift broadening caused by the gradient temperature That is, for every temperature unit, the nonlinear frequency shift decreases , wherein is the nonlinear frequency shift temperature coefficient. The required temperature difference depends on the nonlinear gain bandwidth to be broadened.

[0067] Figure 2 is a schematic diagram of a second regulating sub-module provided by the embodiment of the present application, Figure 3 is another schematic diagram of the second regulating sub-module provided by the embodiment of the present application. Wherein, Figure 2 and Figure 3 respectively show the schematic diagrams of the second regulating sub-module 42 at different angles.

[0068] In some embodiments, as shown in Figure 2 and Figure 3 , the second regulating sub-module 42 includes a plurality of heat sinks 421 arranged in sequence, each heat sink 421 is provided with a fiber groove 4211, and each heat sink 421 is provided with a thermoelectric cooler 422. The fiber groove 4211 is configured to accommodate the doped fiber 33. The thermoelectric cooler 422 is configured to regulate the temperature of the doped fiber 33.

[0069] For example, the fiber groove 4211 includes a U-shaped groove or a V-shaped groove, so that the doped fiber 33 has coaxiality when the doped fiber 33 is arranged in the fiber groove 4211.

[0070] As shown in Figure 2 , the doped fiber 33 is arranged in sequence along the heat sinks 421, and the doped fiber 33 is arranged in the fiber grooves 4211 of the heat sinks 421 in sequence. Each heat sink 421 provides different temperature control for the doped fiber 33 to provide gradient temperature control on the doped fiber 33.

[0071] Exemplarily, after the doped optical fiber 33 is arranged in the optical fiber groove 4211, a heat-conductive silicone grease can be injected into the optical fiber groove 4211 to ensure good heat conduction of the heat sink 421 to the doped optical fiber 33.

[0072] Exemplarily, each heat sink 421 is provided with a temperature sensor 423 to detect the temperature of the corresponding position of the doped optical fiber 33 corresponding to each heat sink 421.

[0073] For example, a counterbore is opened in the middle of each heat sink 421 near the optical fiber (about 3-8 mm away from the core of the optical fiber), a temperature sensor 423 (such as NTC) is placed in the counterbore, and glue is injected.

[0074] Exemplarily, the second regulating sub-module 42 further comprises a liquid cooling radiator 424. The thermoelectric cooler 422 is divided into a heating surface and a heat dissipation surface, the heat dissipation surface is tightly attached to the liquid cooling radiator 424, and the liquid cooling radiator 424 dissipates heat for the three thermoelectric coolers 422.

[0075] Exemplarily, the control of the temperature distribution of the doped optical fiber 33 is realized by the second regulating sub-module 42, the plurality of thermoelectric coolers 422 in the second regulating sub-module 42 are respectively controlled in temperature by a plurality of thermoelectric cooler module controllers on the control module 50, and the thermoelectric cooler module controllers are one-to-one corresponding to the thermoelectric coolers 422 to realize different temperature control.

[0076] The control module 50 sets a different target temperature for different positions of the doped optical fiber 33. The control module 50 further comprises a plurality of temperature reading circuits for reading state signals of the plurality of temperature sensors 423, the temperature reading circuits are one-to-one corresponding to the temperature sensors 423 to monitor the gradient temperature of each position on the doped optical fiber 33 in real time. The control module 50 further comprises a plurality of thermoelectric cooler module controllers, the thermoelectric cooler module controllers are configured to output temperature regulation instructions to regulate the temperature of each thermoelectric cooler 422, and the thermoelectric cooler module controllers are one-to-one corresponding to the thermoelectric coolers 422.

[0077] For example, when the doped optical fiber 33 has a three-section gradient temperature, the temperature of the doped optical fiber 33 close to the pump module 31 is a first temperature , a second temperature , and a third temperature in turn. The control module 50 outputs temperature control instructions to the three thermoelectric coolers 422 of the second regulating sub-module 42 through the three thermoelectric cooler module controllers to regulate the thermoelectric coolers 422, so as to realize the non-uniform and stable temperature distribution of the second regulating sub-module 42.

[0078] It can be understood that the stability of the gradient temperature directly affects the stability of the nonlinear effect suppression effect. Influenced by environmental temperature, fiber type, length and other conditions, the gradient temperature is not a simple fixed control quantity or linear relationship, but also affected by the dynamic changes inside the system. In addition, suppressing nonlinear effects is not just a matter of temperature control, but also coupled with multiple parameters such as pump power, signal light power, polarization state, etc.

[0079] It is worth noting that, for ease of understanding, only a three-section gradient temperature is used as an example in the following description, and it is not limited to only using a three-section gradient control.

[0080] To solve the above problems, the embodiment of the application also provides an intelligent control method of MOPA laser. The machine learning algorithm is used to interact with the laser system in real time, and the laser model is continuously trained and optimized through continuous learning. The best control instruction is automatically found to adjust the pump power, gradient temperature control and other actuators, so as to find a globally optimal working point, output the target power while effectively suppressing the nonlinear effect. And combine the machine learning algorithm with the traditional PID control strategy to ensure the accurate control of pump power and temperature gradient and improve the stability of the laser.

[0081] For example, the system realizes dynamic and accurate control of the laser power and the temperature gradient of the doped optical fiber 33 by running the machine learning model in the embedded hardware. Specifically, the system calculates the optimal non-uniform temperature distribution in real time according to different working conditions and target powers by the machine learning algorithm and applies it to the doped optical fiber 33, so that the nonlinear gain peaks at different positions of the optical fiber are shifted, thereby effectively suppressing the nonlinear effect in the amplifier and improving the tunable power accuracy and spectral quality of the laser.

[0082] Figure 4 is a flowchart of an intelligent control method of a MOPA laser provided by the embodiment of the application, Figure 5 is a flowchart of an intelligent control method of a MOPA laser provided by the embodiment of the application.

[0083] In combination with Figure 1 the intelligent control system of the MOPA laser shown in the figure, the embodiment of the application also provides an intelligent control method of a MOPA laser. The execution subject of the method can be a control module. As shown in Figure 4 and Figure 5 , the method includes the following steps S1-S5.

[0084] S1, construct a target training data set and a one-dimensional convolutional neural network model.

[0085] The target training data set is constructed based on the historical running parameters of the MOPA laser.

[0086] Figure 6 is the architecture schematic diagram of the target model provided by the embodiment of the application.

[0087] Exemplarily, as shown in the figure, Figure 6 the one-dimensional convolutional neural network model has one input layer, two hidden layers and one output layer. In order to meet the requirements of better expression ability and miniaturization of the network, the number of neurons of each layer is designed to be . The feature dimension of the input sample is regarded as a channel, and a series connection architecture of two time feature extractors is adopted to extract multi-scale time features, the convolution kernel sizes are kl and k2 respectively, and the filter sizes are f1 and f2 respectively. Each time feature extractor adopts ReLU function activation max pooling, and through the layer-by-layer increase of the two time feature extractors, rich feature combinations are captured.

[0088] Exemplarily, the target training data set is constructed, including the following steps S11-S17.

[0089] S11, a plurality of sets of historical operating parameters of the MOPA laser are acquired.

[0090] The historical operating parameters include the historical target output power of the MOPA laser, the historical actual output power of the MOPA laser, the historical pump light power of the pump module in the MOPA laser, the historical backscattered light power in the MOPA laser, the historical environment temperature , and the historical first temperature , the historical second temperature and the historical third temperature of the doped optical fiber in the MOPA laser.

[0091] Exemplarily, the plurality of sets of historical operating parameters can correspond to at least two different working conditions, that is, at least correspond to two groups of different environment temperatures and target output powers of the laser which are set in advance.

[0092] Exemplarily, the doped optical fiber of the laser is provided with three gradient temperature zones, that is, a plurality of historical temperatures corresponding to different positions respectively include the historical first temperature , the historical second temperature and the historical third temperature of the doped optical fiber, and the control module collects the laser sensing parameters including the environment temperature , the amplifier pump output power and the laser output power transmitted by the laser main control unit at a certain frequency under different working conditions. Three gradient temperatures of a doped fiber thermoelectric refrigeration device 、 、 Backward nonlinear scattered light power .

[0093] For example, one working condition can be set as an ambient temperature of 25℃ and a target output power of 10W of the laser, and another working condition can be set as an ambient temperature of 35℃ and a target output power of 15W of the laser. Under the two different working conditions, the historical actual output power of the laser The historical pump light power of the pump module The historical backward scattered light power The historical first temperature of the doped fiber The historical second temperature The historical third temperature will show different numerical distribution characteristics. After sorting, cleaning and labeling these multi-dimensional parameter data covering multiple working conditions, the target training data set capable of comprehensively reflecting the running state of the laser can be formed, and rich and representative sample support can be provided for the training of the one-dimensional convolutional neural network model.

[0094] Further, when obtaining the historical working parameters, the control parameters of the first control sub-module and the second control sub-module corresponding to each set of historical running parameters are also needed to train the initial model.

[0095] S12, based on multiple sets of historical running parameters, constructing multiple initial input feature vector samples.

[0096] Exemplarily, step S12 includes steps S121-S123.

[0097] S121, data cleaning is performed on the multiple sets of historical working parameters to obtain multiple sets of historical working parameters with deviations meeting preset requirements.

[0098] In this step, the data is cleaned to remove the data with large deviations in the obtained data. Specifically, the difference between the calculated or experienced actual value and the preset value can be used to determine the difference. When the difference is greater than the difference threshold, it is determined that the data has a large deviation, and the data is deleted to ensure the rationality of the data in the data set and improve the accuracy of the model in learning the actual laser working.

[0099] S122, the multiple sets of historical working parameters with deviations meeting the preset requirements are sorted in the order of acquisition time to obtain an initial data matrix.

[0100] Exemplarily, the control module carries out data cleaning on the collected data and arranges the collected data in the order of the "time stamp", respectively, a historical environmental temperature data set , a historical pump light power data set , a historical actual output power data set , a historical first temperature data set , a historical second temperature data set , a historical third temperature data set , a historical backscattered light power data set , wherein n is the number of collection.

[0101] Thus, according to the working principle of the laser, the input features of the pre-created model are determined, and the input features are 8, respectively, the environmental temperature, the amplifier pump power, the laser output power, the first temperature, the second temperature, the third temperature of the doped optical fiber, the back nonlinear scattering light power, and the laser target power, that is, the input feature vector .

[0102] S123, a preset window is adopted, and a plurality of initial input feature vector samples are selected by sliding on the initial data matrix according to a preset sliding step.

[0103] Exemplarily, a "window" of a fixed size L is moved along the time direction on the collected input data set to create N continuous input data matrices , that is, N data set samples are divided, and the sliding step of the window is K.

[0104] Suppose that the current time point index i is obtained, the input feature matrix is the historical data from i-L to i-1, which is expressed as

[0105] ;

[0106] , wherein L is the time step, and m is the feature dimension.

[0107] Exemplarily, the setting of the sliding step needs to consider the control real-time performance and the calculation efficiency: if high-precision dynamic control needs to be realized, the sliding step can be set to 1, that is, the window is slid by one time step each time to obtain the most dense feature sampling. If the calculation complexity needs to be reduced, the sliding step can be set to 1 / 2 or other proportions of the window length, which reduces the total amount of feature vectors under the premise of ensuring the representativeness of the features. Each initial input feature vector sample is spliced by the historical running parameters of a plurality of continuous time steps in the window in the time sequence, and the dimension thereof is equal to the window length multiplied by the dimension of a single set of historical running parameters (for example, when a single set of parameters contains 8 indexes, and the window length is 10, the feature vector dimension is 80), so that the running state details of each time in the window are completely retained, thereby providing comprehensive and continuous input basis for subsequent laser state modeling and control strategy optimization.

[0108] In this embodiment, a plurality of initial input feature vector samples are obtained by data cleaning, sorting and window division, which can effectively filter out abnormal data interference and ensure the reliability of the input features. Moreover, through time sequence sorting and sliding window division, the time sequence correlation information between the historical running parameters is fully retained, providing input samples with continuous dynamic characteristics for subsequent model training, which helps to improve the perception and prediction ability of the one-dimensional convolutional neural network model for the changes in the running state of the laser.

[0109] S13, inputting the initial input feature vector sample into an initial model, and outputting a control parameter at the i-th moment by the initial model.

[0110] For example, the output instruction sequence is the control parameter at the i-th moment .

[0111] Illustratively, the initial model is trained according to the historical running parameters and the actual control parameters after preprocessing. The actual control parameters refer to the control parameters of the first and second control sub-modules of the laser at the moment of running corresponding to each group of historical running parameters, which are used to train the initial model.

[0112] S14, obtaining the backscattering light power at the i-th moment and the actual output power , and regulating the laser to work by using the control parameter.

[0113] S15, obtaining the backscattering light power at the i+1-th moment and the actual output power .

[0114] S16, determining whether the initial input feature vector sample is a target input feature vector according to the backscattering light power , the actual output power , the backscattering light power and the actual output power , and the target training data set includes the target input feature vector.

[0115] Illustratively, step S16 includes steps S161-S163.

[0116] S161, calculating a first ratio at the i-th moment according to and according to a preset formula.

[0117] S162, calculating a second ratio at the i+1-th moment according to and according to a preset formula.

[0118] Exemplarily, the preset formula includes:

[0119]

[0120] wherein, includes and , is the first ratio, is the second ratio; includes and , includes and .

[0121] S163, in the case where the second ratio meets a preset requirement, determining the initial input feature vector sample as a target input feature vector; the target training data set includes the target input feature vector.

[0122] wherein, the preset requirement includes that the second ratio is less than the first ratio, and / or, the second ratio is less than a preset ratio threshold.

[0123] Exemplarily, the step S163 can include: comparing the second ratio with a preset threshold . If , it indicates that the ratio of the backscattering light power to the output power of the laser at the i+1 moment is in a range where the nonlinear effect is small, and it can be included in the target training data set; if , it indicates that the initial input feature vector sample corresponding to the working state of the laser produces a large nonlinear effect, which does not meet the screening condition of the target input feature vector, and the input parameter needs to be excluded or further adjusted to optimize the working efficiency of the laser.

[0124] In this embodiment, by comparing the ratio changes at the i moment and the i+1 moment, the regulation effect of the current control parameter on the running state of the laser can be intuitively reflected. If the second ratio is less than the first ratio, it indicates that the running performance of the laser has been optimized (such as the energy conversion efficiency is improved, the running stability is enhanced, etc.) after the control parameter corresponding to the initial input feature vector sample is used for regulation. Or, when the second ratio is further less than the preset ratio threshold, it indicates that the regulation effect has reached the expected performance standard. In this way, based on the screening method of the performance index ratio of adjacent moments, the input feature corresponding to the effective regulation can be accurately captured, so as to ensure that the target training data set contains only samples with actual optimization effect, providing high-quality data support for subsequent model learning of efficient control strategy, and helping to improve the accurate prediction and dynamic regulation ability of the model on the running state of the laser.

[0125] ​For example, the proportion threshold can be a pre-set fixed value, which is pre-set according to the design rated parameters of the MOPA laser, the performance requirements of the actual application scene, for example, 0.1, 0.08, etc.

[0126] For example, the proportion threshold can also be a statistical analysis result based on historical operation data of the laser, such as a dynamic threshold obtained by averaging or taking the median of the ratio data in multiple stable operation states.

[0127] For example, the proportion threshold can also be a variable threshold that is adaptively adjusted in combination with real-time environmental factors (such as working temperature, input voltage fluctuation) to ensure that the control effect can be accurately judged to meet the expected standard under different working conditions.

[0128] For example, the proportion threshold can also be periodically updated according to the aging degree of the laser, for example, after running for 1000 hours, the threshold range is recalibrated based on the latest performance detection data, so as to ensure the rationality and applicability of the threshold, and further improve the accuracy of the control effect judgment.

[0129] In some embodiments, after step S163, the target input feature vector can be further screened again through the following step S164, that is, by calculating the target output power at time i and the actual output power at time i+1, to determine whether the control parameter at time i is effective, that is, to make the actual output power of the MOPA laser closer to the target output power.

[0130] For example, step S164 includes calculating , that is, calculating the absolute difference between the second ratio and the first ratio , and determining whether the control parameter at time i is effective according to .

[0131] Further, the target input feature vector that satisfies is less than a pre-set threshold P can be used as data for subsequent training model.

[0132] S17, normalizing each target input feature vector in the target training data set.

[0133] For example, step S17 includes standardizing data preprocessing according to time step L, wherein μ is the mean and σ is the standard deviation; X includes historical target output power of the MOPA laser , historical actual output power of the MOPA laser , historical pump light power of the pump module in the MOPA laser , and historical backscattered light power in the MOPA laser​ , historical ambient temperature , and historical first temperature of the doped fiber in the MOPA laser , historical second temperature , and historical third temperature ; corresponding to X.

[0134] In this way, through steps S11-S17, the multi-dimensional key parameters in the operation process of the MOPA laser can be systematically integrated, the effective feature vectors highly related to the suppression of nonlinear effects and the accuracy of power output are screened out, and the dimensional differences between different parameters are eliminated through normalization processing, so as to obtain a target training data set which is reasonable in structure, complete in information and has strong representation.

[0135] Specifically, the data set not only covers the output performance indicators of the MOPA laser (such as actual output power, backscattered light power), but also includes the core control variables affecting the performance (such as pump light power, temperature of each section of the doped fiber) and environmental interference factors (such as ambient temperature), providing a solid data foundation for the training of the subsequent one-dimensional convolutional neural network model, ensuring that the model can learn the complex coupling relationship between the parameters of the MOPA laser, and then providing reliable decision basis for real-time regulation.

[0136] Specifically, the preset formula is the ratio of back nonlinear light power to total light power. In combination with the preset formula, it can be understood that the embodiments of the application reduce the nonlinear effects generated in the amplification process of the MOPA laser by adjusting the gradient temperature of the doped fiber thermoelectric refrigeration device. With the adjustment of the gradient temperature control module of the doped fiber, the back nonlinear scattered light power is reduced. In other words, the working current of the doped fiber temperature and the pump module is regulated by the instruction output by the trained target model, so as to realize the regulation of the backscattered light.

[0137] For example, the working current of the pump module in the process of regulating the target power of the laser output is adjusted by adjusting the pump power instruction , and the three thermoelectric coolers are real-time and synchronous regulated by the instruction , so that is less than a threshold , or as small as possible. The output vector of the target model is .

[0138] For example, M (≥100) target training samples are obtained, of which 70% × M data sets are training sets, 10% × M data sets are validation sets, and 20% × M data sets are test sets.

[0139] S2, based on the target training dataset, combines the target loss function and backpropagation algorithm to train a one-dimensional convolutional neural network model to obtain the target model.

[0140] For example, the target loss function include:

[0141] ;

[0142] in, It is the number of samples in the current training batch during target training. These are the weighting coefficients for power tracking loss. These are the weighting coefficients for backward nonlinear optical power suppression loss. It is the first The actual output power of each target input feature vector It is the first The target output power of a target input feature vector. For the first The first ratio is calculated from the target input feature vectors.

[0143] For example, the loss function is a core component of machine learning and represents its optimization objective. In this embodiment, the error between the output power and the target power at the next time step is summed. As the basis for calculating the loss function, the goal of model function learning is defined as improving the accuracy of laser output power and reducing... That is, to reduce the proportion of backward nonlinear optical power.

[0144] Specifically, this loss function uses the mean squared error (MSE) to measure the output power tracking accuracy, and L2 regularization to constrain the eigenvalues. .

[0145] For example, the objective loss function can be superimposed with control smoothing line loss, safety constraint loss and other functions according to actual needs to achieve multi-objective optimization control.

[0146] For example, the training process uses the backpropagation algorithm. The backpropagation algorithm is a gradient calculation algorithm based on the chain rule. The backpropagation algorithm mainly consists of three stages: the forward propagation stage, the backpropagation stage, and the iteration stage.

[0147] In the forward propagation phase, batch input sequence data flows from the input layer to the output layer, and the output control parameters are calculated layer by layer. The loss is then calculated based on the composite loss function. In the backpropagation phase, gradients are calculated from the output layer back to the input layer, and gradient descent is used to update the network parameters. The iterative phase involves repeated iterations of multi-objective tradeoffs until the loss converges or the gradient vanishes.

[0148] It is worth noting that in the process of training the model, only the data in the target data set can be used, or the initial obtained historical data can be directly input into the target model. The data in the historical data that is part of the initial input feature vector filtered out by S16 (i.e. not the target input feature vector) can represent part of the abnormal output of the laser, which can also enable the model to learn some output characteristics of the laser.

[0149] S3, obtaining a current operating parameter of the MOPA laser.

[0150] The current operating parameter includes a current target output power of the MOPA laser, a current actual output power, a current pump light power of a pump module in the MOPA laser, a current backscattered light power in the MOPA laser, a current environment temperature, and a plurality of current temperatures of the doped optical fiber in the MOPA laser. Each current temperature is a temperature of a different position of the doped optical fiber.

[0151] For example, the plurality of current temperatures include a current first temperature, a current second temperature, and a current third temperature. The current first temperature, the current second temperature, and the current third temperature are temperatures of different positions of the doped optical fiber.

[0152] S4, inputting the current operating parameter into a target model, and outputting a target control parameter by the target model.

[0153] The target control parameter includes a working current of the pump module in the MOPA laser and a working temperature of the doped optical fiber.

[0154] S5, adjusting the working current of the pump module and the working temperature of the doped optical fiber in the MOPA laser based on the target control parameter.

[0155] The working temperature of the doped optical fiber includes a plurality of different target temperatures, and each target temperature corresponds to a different position on the doped optical fiber.

[0156] In this embodiment, the temperature distribution of each section of the doped optical fiber and the working current of the pump module can be dynamically adjusted to maximize the suppression of the influence of nonlinear effects while ensuring that the output power of the MOPA laser accurately reaches the target value, thereby meeting the stringent requirements of high-precision laser processing, fiber communication, and other scenarios for the performance of the MOPA laser.

[0157] In addition, when the environment temperature fluctuates slightly (such as ±5℃ fluctuation), the machine learning model can quickly complete parameter adjustment to ensure that the output characteristics are not disturbed by the environment.

[0158] Exemplarily, the first control submodule is configured to acquire the historical working parameters and send the historical working parameters to the second control submodule. The second control submodule is configured to train a model based on the historical working parameters to obtain a target model. The second control submodule is further configured to deploy the target model in the first control submodule. The first control submodule is further configured to acquire actual working parameters of the MOPA laser, obtain a target control instruction based on the target model and the actual working parameters, and control the first control submodule and the second control submodule to work based on the target control instruction.

[0159] Exemplarily, the control parameters include a first target temperature, a second target temperature, a third target temperature, and a pump target current; the first target temperature is a target temperature of a first temperature of a doped optical fiber in the MOPA laser , the second target temperature is a target temperature of a second temperature of the doped optical fiber in the MOPA laser , the third target temperature is a target temperature of a third temperature of the doped optical fiber in the MOPA laser , and the pump target current is a target amount of current of the pump module.

[0160] Specifically, step S5 includes controlling a thermoelectric cooler of the doped optical fiber corresponding to the first target temperature to work based on the first target temperature; controlling a thermoelectric cooler of the doped optical fiber corresponding to the second target temperature to work based on the second target temperature; controlling a thermoelectric cooler of the doped optical fiber corresponding to the third target temperature to work based on the third target temperature; and controlling working of a multimode pump current circuit in the pump module based on the target current amount.

[0161] Exemplarily, a constraint layer is further constructed in the one-dimensional convolutional neural network model to ensure that the control output of the thermoelectric cooling module conforms to the physical reality. The output of the second hidden layer is connected to the physical constraint layer, and the output of the constraint layer is connected to the final output layer of the model.

[0162] For example, the one-dimensional convolutional neural network model includes a first constraint, a second constraint, a third constraint, and a fourth constraint.

[0163] The first constraint includes that the first target temperature, the second target temperature, and the third target temperature corresponding to the target control parameter are all within a preset range. In this way, the temperature range can be limited through the first constraint to prevent dewing caused by excessively low temperature and damage to the laser caused by excessively high temperature.

[0164] The second constraint includes that a minimum temperature difference between the first temperature and the second temperature is , and a minimum temperature difference between the second temperature and the third temperature is , , . In this way, the temperature gradient can be constrained to ensure that the temperature gradient direction is correct and effectively suppresses the nonlinear effect.

[0165] The third constraint comprises: , is the maximum allowed gradient. In this way, the maximum gradient can be limited (to prevent thermal shock), and the temperature change is too fast to cause thermal stress.

[0166] The fourth constraint comprises: wherein, is the minimum value of the pump module current, is the maximum value of the pump module current. In this way, the amplifier pump drive current range can be limited to prevent damage to the pump or to achieve the amplification power.

[0167] In some embodiments, the intelligent control method of the MOPA laser provided by the embodiments of the present application is deployed on a single-chip microcomputer for execution. Before obtaining the current operating parameters of the MOPA laser, that is, before actually regulating and controlling the MOPA laser, the target model also needs to be deployed on the single-chip microcomputer.

[0168] The exemplary step S3 further comprises the following steps S3a-S3c.

[0169] S3a, lightening the target model.

[0170] In this step, model lightening and conversion can be performed. For example, converting model parameters from 32-bit floating-point numbers to 8-bit integers, and converting all weights and activation values to integers, which can greatly reduce the model size (about 75%) and speed up the calculation, and has little effect on the accuracy.

[0171] Exemplarily, the step S3a comprises the following steps S3a1 and / or S3a2.

[0172] S3a1, converting a first parameter format of the target model into a second parameter format, the second parameter format occupying a smaller space than the first parameter format.

[0173] In this way, the model size can be effectively compressed by parameter format conversion, for example, converting 32-bit floating-point parameters in the model to 16-bit fixed-point or 8-bit quantization, which significantly reduces the parameter storage space occupation under the premise of ensuring controllable control accuracy loss.

[0174] S3a2, pruning the target model.

[0175] In this way, the model pruning can reduce the computational complexity and inference delay of the model by removing redundant weight connections or neuron nodes in the model, and further reduce the model size.

[0176] In this embodiment, performing one or both of the two alone or in combination can make the lightweight target model better adapt to the limited memory resources and computing power level of the single-chip microcomputer, ensuring that the intelligent control algorithm of the subsequent MOPA laser can be stably and efficiently executed on the single-chip microcomputer, meeting the needs of real-time regulation and control.

[0177] S3b, deploying the target model on the single-chip microcomputer.

[0178] In this step, the trained and optimized model is converted into a format suitable for embedded single-chip microcomputers (for example,.tflite) and imported into the single-chip microcomputer, and a suitable inference engine (TensorFlow Lite for Microcontrollers is selected in this embodiment) is selected to run the imported model.

[0179] S3c, initializing the target model.

[0180] In this step, inference code is written in the single-chip microcomputer to initialize the model and the interpreter. An input tensor (tensor) and an output tensor are opened in the program. The sensor data and target power value (after the same preprocessing and normalization) collected in real time are filled into the input tensor, and the interpreter is called for inference. The result prediction value is obtained from the output tensor, and the corresponding operation is performed.

[0181] In this embodiment, through the cooperative operation of steps S3a to S3c, the target model not only realizes efficient adaptation to the hardware environment of the single-chip microcomputer, but also completes the initialization configuration after deployment, so that it can quickly respond to subsequent control instruction input, laying a solid foundation for the laser parameter regulation based on model output in step S3. Moreover, the initialization process can ensure that the model loads the preset weight parameters and running environment variables, eliminating possible parameter deviation or environmental conflicts after deployment, further ensuring the stability and reliability of the model running on the single-chip microcomputer, so that the intelligent control process of the MOPA laser can proceed smoothly as designed.

[0182] For example, the target model can directly perform inference on the device terminal, which has great advantages in real-time performance, security, and cost optimization. At the same time, due to the limitations of the resources of the embedded single-chip microcomputer, such as the computing power of the single-chip microcomputer being determined by the CPU frequency and the running memory (generally, CPU frequency is several tens to several hundred MHz, and running memory is several tens to several hundred KB), the lightweight model (one-dimensional convolutional neural network) is used for model training in this design, which can greatly reduce the required running memory.

[0183] In some embodiments, the embedded microcontroller can also transmit new data generated in the real environment back to the processor, which can be used to retrain and optimize the model, forming a closed-loop system to continuously evolve the model. The final optimal model realizes real-time calculation and decision-making in the local embedded microcontroller.

[0184] For example, the embedded microcontroller is the core controller of the control module. Moreover, the peripheral circuit in the control module is the basis for achieving the control target. The temperature reading circuit can convert the temperature sensor signal into an analog voltage signal and transmit it to the embedded microcontroller, which receives the ambient temperature and the temperature state of the three thermoelectric coolers in real time. The final output end of the laser fiber is split into 1% of the output light connected to the second silicon photodiode, and the laser output power detection circuit converts the second silicon photodiode signal into a voltage signal through a transimpedance amplifier and transmits it to the single-chip microcomputer, which monitors the output power and its changes in real time. Similarly, the backward nonlinear optical power detection circuit transmits the nonlinear optical power output received by the first silicon photodiode to the single-chip microcomputer, which monitors the backward nonlinear optical power and its changes in real time.

[0185] The communication circuit on the seed source controller is connected to the seed source communication circuit on the main control unit, realizing the function of communication between the seed source controller and the main control board through RS232. The embedded microcontroller exchanges data with the seed source controller through the seed source module communication circuit, which can detect or control the related parameters of the seed source module, including collecting the seed source pump temperature and seed light power.

[0186] The laser in the embodiment of the application adopts a seed source module with stable output power and wavelength, and the stable seed light generated is amplified by the amplification unit module and the doped optical fiber to output a target laser with adjustable power. The final output laser power and the generated nonlinear effect are mainly determined by the power of the multimode pump in the amplification unit module, and the gradient temperature control of the doped optical fiber can effectively reduce the nonlinear effect.

[0187] The communication circuit on the pump controller is connected to the pump module communication circuit on the main control unit, realizing the function of communication between the pump controller and the main control board through RS232. The embedded microcomputer exchanges data with the pump controller through the pump module communication circuit, which can detect and control the related parameters of the amplification unit module, including collecting the pump temperature and pump light power in the amplifier module, setting the multimode pump driving current, temperature control instruction, etc.

[0188] The environmental temperature, amplifier pump power, laser output power, backward nonlinear optical power, first temperature, second temperature, third temperature of the doped optical fiber thermoelectric refrigeration device and the set target power collected by the embedded single-chip microcomputer are converted into numerical values to form the input tensor of the model, the interpreter is called for reasoning to obtain the result prediction output parameter value from the output tensor, including the multimode pump driving current set value in the amplifier module and the three thermoelectric cooler temperature adjustment values. Among them, the multimode pump driving current set value output by the single-chip microcomputer is transmitted to the pump controller through the communication circuit, and the pump controller drives the multimode pump to generate corresponding "excitation light" according to the received current set value, thereby realizing the adjustment of the laser output power. The three thermoelectric cooler temperature adjustment values are respectively output to the first thermoelectric cooler temperature adjustment circuit, the second thermoelectric cooler temperature adjustment circuit and the third thermoelectric cooler temperature adjustment circuit through the single-chip microcomputer, and finally the three thermoelectric coolers in the gradient temperature module are adjusted through the corresponding thermoelectric cooler driving circuit, thereby realizing the control of the gradient temperature module and achieving the suppression of the generation and influence of the nonlinear effect.

[0189] Figure 7 is another flowchart of the intelligent control method of the MOPA laser provided by the embodiment of the present application.

[0190] In some embodiments, the control module further comprises a processor configured to execute the intelligent control method of the MOPA laser, as shown in Figure 7 The intelligent control method of the MOPA laser comprises the following steps S100-S700.

[0191] S100, the laser main control unit collects the laser operating parameters and transmits them to the processor through the communication line.

[0192] For example, the laser main control unit can be an optical power detection module in the control module.

[0193] S200, the processor cleans the data and forms an ordered data set according to the collected "time stamp" to construct the input feature vector of the model .

[0194] S300, according to the ratio of the backward nonlinear optical power to the total optical power , the high-quality samples are screened, a one-dimensional convolutional neural network model is constructed, and the model is trained using high-quality samples.

[0195] S400, a target loss function is constructed , and a back propagation algorithm is used until the loss converges or the gradient disappears, and the target model is obtained.

[0196] S500, the target model is lightened and pruned and deployed in the embedded single-chip microcomputer.

[0197] S600, the embedded single-chip microcomputer runs the model, and outputs control instructions , a module for adjusting the output power and temperature gradient of the laser.

[0198] S700, the embedded single-chip microcomputer processor runs, and transmits new data generated in the real environment back to the processor for retraining and optimizing the model.

[0199] It is noted that other embodiments of the application will be readily apparent to those skilled in the art from the disclosure herein. The application is intended to embrace all such alterations, permutations, and variances that fall within the scope of this application, including adapting the application to various methods of use or behaviors to promote the general purpose of the application. The above specification, examples and data provide essential control and description of the application, and are

[0200] It is to be understood that the application is not limited to the precise details of construction and the arrangement of components described herein and illustrated in the drawings. The scope of the application should be determined by the terms of the appended claims and their equivalents.

Claims

1. A method for intelligent control of a MOPA laser, characterized in that, include: A target training dataset and a one-dimensional convolutional neural network model are constructed, wherein the target training dataset is constructed based on the historical operating parameters of the MOPA laser; Based on the target training dataset, the one-dimensional convolutional neural network model is trained by combining the target loss function and the backpropagation algorithm to obtain the target model; Obtain the current operating parameters of the MOPA laser; wherein, the current operating parameters include the current target output power, the current actual output power, the current pump light power of the pump module in the MOPA laser, the current backscattered light power in the MOPA laser, the current ambient temperature, and multiple current temperatures of the doped fiber in the MOPA laser; each current temperature is the temperature at a different location of the doped fiber; The current operating parameters are input into the target model, and the target model outputs target control parameters; the target control parameters include the operating current of the pump module and the operating temperature of the doped fiber in the MOPA laser. Based on the target control parameters, the operating current of the pump module and the operating temperature of the doped fiber in the MOPA laser are adjusted. The operating temperature of the doped fiber includes multiple different target temperatures, each of which corresponds to a different position on the doped fiber.

2. The intelligent control method of a MOPA laser according to claim 1, wherein, Construct the target training dataset, including: acquire a plurality of sets of historical operating parameters of the MOPA laser; the historical operating parameters include historical target output power of the MOPA laser , historical actual output power of the MOPA laser , historical pump light power of a pump module in the MOPA laser , historical backscattered light power in the MOPA laser , historical ambient temperature , and historical first temperature of a doped optical fiber in the MOPA laser , historical second temperature , and historical third temperature ​ Based on the aforementioned historical operating parameters, multiple initial input feature vector samples are constructed; The initial input feature vector sample is input into the initial model, and the initial model outputs the control parameters at time i. Obtaining backscattered light power at time i with actual output power and using the control parameters to regulate the operation of the MOPA laser acquiring the backscattered light power at the i+1 time and the actual output power ; determining whether the initial input feature vector sample is a target input feature vector based on backscattered light power , actual output power , backscattered light power , and actual output power , the target training data set including the target input feature vector; Each of the target input feature vectors in the target training dataset is normalized.

3. The intelligent control method of a MOPA laser according to claim 2, wherein, Based on multiple sets of historical operating parameters, multiple initial input feature vector samples are constructed, including: Data cleaning is performed on multiple sets of historical operating parameters to obtain multiple sets of historical operating parameters whose deviations meet preset requirements; The historical operating parameters that meet the preset requirements are sorted according to the order of acquisition time to obtain the initial data matrix; Using a preset window, multiple initial input feature vector samples are selected by sliding the window across the initial data matrix at a preset sliding step size.

4. The intelligent control method for the MOPA laser according to claim 2, characterized in that, The initial input feature vector sample is determined to be a target input feature vector according to the backscattering light power , the actual output power , the backscattering light power , and the actual output power , comprising: According to a preset formula, a first ratio at i moment is calculated according to and ​ According to the preset formula, the second ratio at the i+1 moment is calculated according to and . If the second ratio meets the preset requirements, the initial input feature vector sample is determined to be the target input feature vector; the target training dataset includes the target input feature vector. The preset requirement includes the second ratio being less than the first ratio, and / or the second ratio being less than a ratio threshold.

5. The intelligent control method of a MOPA laser according to claim 4, wherein, The preset formula includes: ； wherein comprising and , said is said first ratio, said is said second ratio; comprising and , comprising and .

6. The intelligent control method for the MOPA laser according to any one of claims 1-5, characterized in that, The target loss function comprises: ； wherein, is the number of samples in the current training batch, is a weight coefficient of the power tracking loss, is a weight coefficient of the backward nonlinear optical power suppression loss, is the actual output power of the first target input feature vector, is the actual output power of the i-th target input feature vector, is the target output power of the first target input feature vector, is the target output power of the i-th target input feature vector, is the first ratio calculated for the first target input feature vector, and is the first ratio calculated for the i-th target input feature vector.

7. The intelligent control method for the MOPA laser according to any one of claims 1-5, characterized in that, The control parameters include a first target temperature. Second target temperature Third target temperature and pump target current The first target temperature is the first temperature of the doped fiber in the MOPA laser. The target temperature, and the second target temperature are the second temperatures of the doped fiber in the MOPA laser. The target temperature, wherein the third target temperature is the third temperature of the doped fiber in the MOPA laser. The target temperature, and the target pump current is the target current of the pump module; The one-dimensional convolutional neural network model includes a first constraint, a second constraint, a third constraint, and a fourth constraint; The first constraint includes: the first target temperature, the second target temperature, and the third target temperature corresponding to the target control parameter are all within a preset range; The second constraint includes: the minimum temperature difference between the first target temperature and the second target temperature is The minimum temperature difference between the second target temperature and the third target temperature is , , ， ; The third constraint includes: ， , The maximum allowed gradient; The fourth constraint includes: ,in, This is the minimum current of the pump module. This is the maximum current of the pump module.

8. The intelligent control method for the MOPA laser according to any one of claims 1-5, characterized in that, Before obtaining the current operating parameters of the MOPA laser, the process also includes: Lighten the target model; The target model is deployed on a microcontroller; Initialize the target model.

9. The intelligent control method for the MOPA laser according to claim 8, characterized in that, The lightweighting of the target model includes: Convert the first parameter format of the target model to a second parameter format, wherein the second parameter format occupies less space than the first parameter format; and / or, The target model is pruned.

10. An intelligent control system for a MOPA laser, characterized in that, include: Seed source module, first coupler, amplifier module, regulation module, and control module; The amplifier module includes a pump module, a first combiner, and a doped fiber; The first beam combiner includes a first input terminal, a second input terminal, and a first output terminal; The seed source module is optically connected to the first input terminal, the pump module is optically connected to the second input terminal, and the doped fiber is optically connected to the first output terminal. The seed source module is configured to output a seed laser; The pump module is configured to output pump light; The first beam combiner is configured to combine the seed laser and the pump light according to a first beam combining ratio to form a first target laser, and output the first target laser to the doped fiber; The doped fiber is configured to oscillate the first target laser to form a second target laser; The first coupler is disposed on the output optical path of the seed source module. The first coupler includes a second output terminal, which is configured to output the backscattered light generated by the amplifier module. The control module is communicatively connected to the regulation module. The control module is configured to: construct a target training dataset and a one-dimensional convolutional neural network model, wherein the target training dataset is constructed based on the historical operating parameters of the MOPA laser; based on the target training dataset, combined with a target loss function and a backpropagation algorithm, train the one-dimensional convolutional neural network model to obtain a target model; obtain the current operating parameters of the MOPA laser; wherein the current operating parameters include the current target output power, the current actual output power, the current pump light power of the pump module in the MOPA laser, the current backscattered light power in the MOPA laser, the current ambient temperature, and multiple current temperatures of the doped fiber in the MOPA laser; each current temperature is the temperature at a different location in the doped fiber; input the current operating parameters into the target model, and the target model outputs target regulation parameters; the target regulation parameters include the operating current of the pump module in the MOPA laser and the operating temperature of the doped fiber; The control module includes a first control submodule and a second control submodule; The first control submodule is configured to adjust the operating current of the pump module in the MOPA laser based on the target control parameters; The second control submodule is configured to adjust the operating temperature of the doped fiber in the MOPA laser based on the target control parameters; The number of the second control sub-modules is multiple, and the multiple second control sub-modules are arranged sequentially along the doped optical fiber. Each second control sub-module is configured to control the operating temperature at different positions of the doped optical fiber.

Images