An inverter power supply control method suitable for a water surface unmanned ship

Abstract

The application relates to a control method of an inverter power supply suitable for a water surface unmanned ship, and the control method relates to an inverter power supply control device which comprises a storage battery, a DC-DC voltage boosting conversion unit, an LC filter unit, a DC-AC conversion unit, a rectification and filter unit, a voltage and current signal acquisition unit, an STM32F767 chip and a TMS320C6678 chip; the storage battery, the DC-DC voltage boosting conversion unit, the LC filter unit, the DC-AC conversion unit and the rectification and filter unit are electrically connected in sequence; the control method is as follows: DC 12V or 24V power supply is inverted into AC 220V, and the inversion is mainly realized through two processes, namely, DC-DC voltage boosting conversion and DC-AC conversion; in the DC-AC process, a BP neural network indirect model reference self-adaptive control algorithm is used to control output SPWM waves. The application realizes conversion of DC power into high-quality AC power.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power supply control of unmanned surface vehicle, and particularly relates to an inverter power supply control method suitable for unmanned surface vehicle. BACKGROUND

[0002] Some large surveying ships, manned island surveying boats or local fishing boats equipped with single-beam depth sounders, multi-beam depth sounding systems, magnetometers, gravimeters go to the sea area for surveying and mapping, and there are great safety hazards, and it is difficult to obtain seabed topographic information near the islands. The unmanned surface vehicle carrying single-beam depth sounders, multi-beam depth sounding systems, magnetometers, gravimeters for marine gravity, magnetic force, topographic and geomorphic surveying can solve the problem. Various marine surveying and mapping instruments and equipment are precision instruments and equipment, and have high requirements for power supply voltage and electromagnetic compatibility (EMC). A power supply system with high reliability and stable output is the primary guarantee for the unmanned surface vehicle to perform marine surveying and mapping tasks.

[0003] The marine surveying and mapping instruments and equipment such as multi-beam sonar and single-beam depth sounder provide AC 220V power supply for the power supply at the end of the unmanned surface vehicle, and have certain requirements for the quality of the AC 220V power supply. From the perspective of equipment and personnel safety, the power supply of the unmanned surface vehicle generally uses 12V storage battery, 12V UPS battery or generator (outputting DC 12V or 24V) as the main power supply. Therefore, the inverter power supply as an energy conversion device plays an important role in the power supply system at the end of the unmanned surface vehicle. As the number of electric equipment and load operation equipment of the unmanned surface vehicle increases, the number of AC 220V power supply equipment also gradually increases. A series of problems such as harmonic increase and waveform distortion caused by various nonlinear loads (surveying and mapping equipment) will seriously affect the normal work of other equipment, and will also cause serious electromagnetic interference to the equipment of the unmanned surface vehicle, and even will seriously affect the entire power supply system of the unmanned surface vehicle. SUMMARY

[0004] The application aims to overcome the deficiencies of the prior art, and provides an inverter power supply control method suitable for unmanned surface vehicle.

[0005] The above-mentioned purpose of the application is realized by the following technical scheme:

[0006] The application discloses a control method of an inverter power supply suitable for a water surface unmanned ship.

[0007] The control method is to inversely convert a direct current 12V or 24V power supply into an alternating current 220V, and is mainly realized through two processes, i.e., DC-DC voltage conversion and DC-AC conversion. The first stage adopts DC-DC conversion, 2 groups of symmetrical push-pull circuits are used as main topologies, a high-frequency transformer adopts a double-transformer form, the primary winding is connected in parallel with one group of push-pull circuits, the same time sequence control mode is adopted, the PWM waveform is outputted by STM32F767 for program control and driving of the two groups of switching tubes, the two groups of switching tubes are alternately turned on under the driving of the pulse width modulation technology (PWM) outputted by the STM32F767, the high-frequency transformer outputs a high-frequency square wave and adopts full-bridge rectification. The direct current 310V outputted by the DC-DC conversion is filtered through an LC filter to obtain a boosted direct current; the second stage adopts DC-AC conversion, a full-bridge inverter circuit is adopted, the TMS320C6678 chip is used to complete the control of the full-bridge inverter circuit, a sine wave pulse width modulation technology (SPWM) control task is completed, a sine alternating current output is obtained, the sine alternating current output is rectified and filtered to supply the use equipment. The TMS320C6678 chip uses an important peripheral enhanced pulse width modulation ePWM module to realize the comparison output of a triangular wave and a sine wave to generate an SPWM wave; the voltage and current signal acquisition unit acquires the battery voltage, the voltage after the DC-DC voltage conversion and the voltage and power after rectification and filtering in real time, and protects the overcurrent, overvoltage, undervoltage and overload of the whole inverter circuit.

[0008] Further, the control method adopts an inverter power supply control method based on a neural network indirect model reference adaptive control, wherein the TMS320C6678 chip controls the output of an SPWM wave through a BP neural network indirect model reference adaptive control algorithm.

[0009] Further: the indirect model reference self-adaptive control algorithm based on the BP neural network comprises a reference model, a controlled object, a neural network identifier and a neural network controller; the neural network identifier is used for establishing a neural network model of DC-AC conversion by using the characteristics of the neural network, introducing the model into a DC-AC conversion control process, identifying the controlled system model (DC-AC) by using the neural network, and then continuously training the neural network controller by using test data, so that the output of the controlled system tracks the output of the reference model; the reference model is a reference control model of the DC-AC process; the controlled object is DC-AC (12V to 220V AC or 24V to 220V AC); the neural network controller is used for continuously identifying the controlled object (DC-AC) by using the neural network algorithm, continuously training by using the actual DC-AC data, so that the output of the controlled system tracks the output of the reference model, and high-quality AC power is output for the end measurement and surveying equipment of the unmanned ship.

[0010] The application has the advantages and positive effects that:

[0011] 1、 The application designs an inverter power supply device and a control method meeting the use requirements of the unmanned ship according to the specific inverter power supply function requirements of the water surface unmanned ship, and specifically relates to an inverter power supply control device, which comprises a storage battery, a DC-DC step-up conversion unit, an LC filter unit, a DC-AC conversion unit, a rectification and filter unit, a voltage and current signal acquisition unit, an STM32F767 chip and a TMS320C6678 chip; the storage battery, the DC-DC step-up conversion unit, the LC filter unit, the DC-AC conversion unit and the rectification and filter unit are electrically connected in sequence to inversely convert the DC 12V or 24V power supply into AC 220V, so that the DC power is converted into AC power.

[0012] 2、 The control method adopts an inverter power supply control method based on the neural network indirect model reference self-adaptive control, has the characteristics of good nonlinear control performance, strong anti-interference ability, self-adaptation and autonomous learning, and is suitable for the inverter power supply control and use environment of the water surface unmanned ship. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 It is a principle block diagram of the inverter power supply device of the unmanned ship of the application;

[0014] Figure 2 It is a principle diagram of the voltage signal and current signal acquisition circuit of the application

[0015] Figure 3 It is a model reference self-adaptive principle diagram of the application;

[0016] Figure 4 Structure diagram of neural network system identifier of the present application;

[0017] Figure 5 Structure diagram of BP neural network of the present application;

[0018] Figure 6 Flow chart of identification of the present application;

[0019] Figure 7 Flow chart of reference model algorithm of the present application;

[0020] Figure 8 Reference model output and controlled object output curve of the present application;

[0021] Figure 9 Input curve diagram of controlled object of the present application;

[0022] Figure 10 Error curve diagram of system output of the present application. DETAILED DESCRIPTION

[0023] The structure of the present application will be further described below in combination with the accompanying drawings and by examples. It should be noted that the present examples are narrative rather than limiting.

[0024] A control device for an inverter power supply suitable for an unmanned surface vehicle, please see Figures 1-10 The application point is: including a battery (12V or 24V), a DC-DC boost conversion unit, an LC filter unit, a DC-AC conversion unit, a rectification and filter unit, a voltage and current signal acquisition unit, an STM32F767 chip and a TMS320C6678 chip; the battery, the DC-DC boost conversion unit, the LC filter unit, the DC-AC conversion unit and the rectification and filter unit are electrically connected in sequence.

[0025] The control method of the inverter power supply control device is to invert the DC 12V or 24V power supply into AC 220V, which is mainly realized through two processes: DC-DC boost conversion and DC-AC conversion. The first stage adopts DC-DC conversion, and the DC-DC conversion adopts 2 groups of symmetrical push-pull circuits as the main topology, and the high-frequency transformer adopts a double transformer form, and the primary side windings are connected in parallel and connected to a group of push-pull circuits respectively, and the same timing control mode is adopted, and the PWM waveform is output by STM32F767 for program control and driving two groups of switching tubes. The two groups of switching tubes are alternately turned on under the PWM drive of STM32F767 output, and the high-frequency transformer outputs a high-frequency square wave and adopts full-bridge rectification. The DC signal output by the DC-DC conversion is filtered by LC after the subsequent DC-DC conversion to obtain the boosted DC 310V. The inverter power supply device automatically adjusts the output PWM by collecting the voltage value after DC-DC boost, so as to adjust the size of the boosted voltage. The second stage adopts DC-AC conversion, and the DC-AC conversion adopts a full-bridge inverter circuit, and a TMS320C6678 chip is used to complete the SPWM control task of the full-bridge inverter circuit to obtain a sinusoidal AC output. The inverter power supply device collects voltage and current signals in real time through the TMS320C6678 chip, and then outputs the corresponding control signals, and generates SPWM drive signals to control the turn-on and turn-off of the inverter circuit switching tube.

[0026] The inverter power supply device collects the voltage value after DC-AC conversion in real time through the TMS320C6678 chip, and controls the output SPWM through the BP neural network indirect model reference adaptive control algorithm, so as to control and adjust the quality of the AC power supply, so that the output AC power meets the index. After rectification and filtering, the AC power is provided for the boat end equipment of the unmanned ship. At the same time, the inverter power supply device collects the battery voltage and the current in the line in real time, and realizes the overcurrent, overvoltage, undervoltage and overload protection of the entire inverter circuit.

[0027] As Figure 2The voltage signal acquisition, current signal acquisition circuit schematic diagram is shown, the required battery, DC-DC boost circuit, DC-AC AC part circuit is connected through ADC1_VCC, ADC1_IN, ADC1_GND and ADC2_VCC, ADC2_IN, ADC2_GND multiple voltage signal acquisition and current signal acquisition sensor (current sensor), specifically, the signal is given to the 11 pin (Sin+) of T6560P module, and the 11 pin 3 pin (Sout) of T6560P module is output to AD8552 chip. After conversion, the signal is transmitted to the PF3 pin of STM32F767 chip or the signal acquisition pin of TMS320C6678 chip. The inverter power supply device collects the battery voltage and the current in the circuit in real time, and realizes the overcurrent, overvoltage, undervoltage and overload protection of the entire inverter circuit.

[0028] The control method is an indirect model reference adaptive control method based on BP neural network. The indirect model reference adaptive control algorithm based on BP neural network is composed of reference model, controlled object, neural network identifier (NNI) and neural network controller (NNC), as shown below. Figure 3 r is the reference input of the control system, under the condition of meeting the performance index reference model y m , the difference e c between the output is used to adjust the weight of the neural network controller, the identification parameters of the identifier are set through offline training, and then online identification is carried out, the identification error signal is provided for the neural network controller (NNC), and the parameters of the controller are trained, u is the output of the controller NNC (Neural Networks controller); y is the output of the controlled object; e i is the difference between the identification model and the output of the controlled object, which is used to train the parameters of the identifier NNI; e c is the difference between the output of the reference model and the controlled object, which is used to train the control parameters of the neural network. Obviously, the output error e c =y m cannot be directly used to train the neural network controller, and the difference Δu between the control value and the actual control value must be combined with the control difference value generated by the neural network identifier, and the correction value is obtained by backward propagation in the network identifier, and then used to train the neural network controller. Therefore, in order to obtain better control effect, the mathematical model of the controlled object (DC-AC direct current to alternating current) is accurately identified by using the neural network identifier, and then e m(t) Online learning and correction are performed to achieve the purpose of control. The neural network indirect model reference adaptive control uses two neural networks: a controller network and an identifier network. First, a neural network is used to identify the model of the controlled system (DC-AC direct current to alternating current), and then a neural network controller is trained to make the output of the controlled system track the output of the reference model.

[0029] The reference model is selected as a reference object of the controlled object, and the following first-order model is selected:

[0030] y = 1 / aTs + 1 (1)

[0031] is the reference model. By selecting parameters a and T, the reference model can obtain better output. The selection of parameters generally meets the requirement of small steady-state error, and the adjustment time is long or short, depending on the specific situation of the control system. In engineering tests, the reference model in the range of 0.46≤α≤0.51 cannot effectively control the system, so the control effect is not satisfactory. When 0<α≤0.46, the system can obtain a relatively satisfactory control effect, and when α>0.51, the system has no output, so parameters in this range cannot be selected to construct the reference model.

[0032] The following second-order model is selected

[0033]

[0034] as the reference model.

[0035] When ξ>1, the second-order system is in an over-damped state, and the system response is too slow, so this parameter setting is not used.

[0036] When ξ=1, the system is in a critical damping state, which has a shorter rise time and faster response speed compared to the over-damped state.

[0037] When ξ<1, the system is in an under-damped state. In the response curve under this state, the smaller the damping ratio, the larger the overshoot, and the shorter the rise time.

[0038] When ξ=0, the system is in an undamped state, and when ξ<0, the second-order system is unstable, so these two cases are not considered.

[0039] The structure of the neural network identifier in the inverter power control method is shown in Figure 4 The identification is to determine the identification model of the system in the transformed observable input and output data, so that the error criterion function is minimized, that is:

[0040] J = ||y(t)-y n (t)|| = ||e|| < ε (3)

[0041] y(t) and y n (t) are the output responses of the system and model respectively, where ε > 0 is the preset identification accuracy, which is a given minimum value. For single-input, single-output model, the NARX is used to describe the nonlinear autoregressive model as the identification model of neural network, and its mathematical description is as follows:

[0042]

[0043] Wherein, it is assumed that:

[0044] The model structure is known, that is, m, n are known;

[0045] u(k), y(k) are measurable;

[0046] For all possible input u, the system output y is uniformly bounded, that is, the system is stable;

[0047] The BP neural network has one or more sigmoid hidden layers and linear output layer, as shown in Figure 5 In the BP neural network, the input signal is forward propagated and the error is backward propagated. The activation function of hidden layer neurons generally uses sigmoid function or hyperbolic tangent function, or its deformation function, and the output layer neuron activation function can be selected according to the actual situation. If it is a linear function, the output of the entire network can take any value. If the S-shaped function f(x) = 1 / 1+exp(-αx), 0 < α < 1 is selected, the output of the entire network is limited within a certain range. In the identification process, the BP neural network uses three-layer neural network, the activation function of hidden layer is hyperbolic tangent function, the neurons of output layer are linear neurons, the input signal is sine signal, the sampling time is 0.001 s, and the learning parameters of network are α = 0.05 and η = 0.3 respectively.

[0048] The design steps of BP neural network are as follows:

[0049] ① Initialization, set the connection weights ω ij ,υ jk

[0050] Threshold θ j ,θ k The initial value is a random number between [-1, 1].

[0051] ② Provide training samples, select training samples (X p , D p ), wherein X p = [x1, x2, … x m ] T is the system input D p = [d1, d2, … dn ] T The system expects the input.

[0052] The forward propagation calculation of the input signal, the input signal of the hidden layer and the output layer According to

[0053]

[0054] The calculation is performed.

[0055] The backward propagation calculation of the error signal, from the output layer to the input layer according to

[0056] δ k = e k f'(u k ) (6)

[0057]

[0058] The value of the generalization error of each layer of neurons is calculated.

[0059] The correction calculation of the weight and threshold of each layer of the network is performed according to the following formula respectively.

[0060]

[0061]

[0062]

[0063] Step 2, return to step 2, and recalculate each learning sample according to the corrected network connection weight and threshold, until the set network target function E p < ε, which means that the network is convergent, i.e. the control effect is achieved, ε is a pre-defined minimum value, or the learning is ended when the pre-defined maximum learning times is reached. The system identification flow chart is shown in Figure 6

[0064] The model reference controller part (referring to the above-mentioned BP neural network based model reference adaptive controller), the reference model of the controlled object, the output of the reference model is y m (k), the output y(k) of the controlled object required by the control system can track the output y m (k) of the reference model.

[0065] The tracking error is:

[0066] e(k) = y m (k) - y(k) (11) ​​

[0067] The performance index is:

[0068]

[0069] The output of the controller is BP neural network

[0070]

[0071] m is the number of BP network hidden layer neurons, w j is the connection value between the jth network hidden layer neuron and the output layer, h j is the output of the jth neuron. The weight vector of the network is W = [w1, w2, w3…wn]T. m T The learning algorithm of the network weight value can be obtained by gradient descent method as follows:

[0072]

[0073] w j (k) = w j (k-1) + Δw j (k) + αw j (k) (15)

[0074] wherein η is the learning rate, and α is the momentum factor.

[0075] To verify the effect of the algorithm, the online simulation and debugging of the algorithm program are carried out on the matlab software, and the model of the controlled object is a nonlinear system:

[0076]

[0077] The reference model is a first-order system:

[0078] y m (k) = 0.6y m (k-1) + r(k) (17)

[0079] The input signal is selected as:

[0080] r(k) = 0.5sin(0.005πk) (18)

[0081] The BP neural network is adopted for control, the sine signal is taken as the input excitation applied to the controlled object and the output y(k) of the controlled object to form sample data for offline training of the neural network identifier NNI to obtain the identification model of the controlled object; and the same input signal is taken as the input of the reference model, and the output and the output y(k) of the controlled object form sample data for offline training of the neural network controller NNC. The BP neural network structure adopts Figure 4 ​The activation function of the network hidden layer is a sigmoid function, the learning rate η is 0.46, the momentum factor α is 0.01, the network initial value is a random number between 0 and 1, and the performance index ε is 0.005. Figure 6 As shown in the figure.

[0082] The simulation result is shown in the figure. Figure 8 The contrastive graph of the system output y(k) following the reference model output y m (k) can be seen, the system output y(k) can track the reference model output y Figure 10 (k) well, and even can keep following state completely after a period of time. The system output error transformation graph is shown in the figure, the system error is unstable at the beginning, and finally fluctuates around zero value with very small amplitude to keep stable state, because the initial value of the neural network weight is given randomly, the weight adjustment needs a certain time, but after the online learning of the neural network, the connection weight gradually tends to the optimal value, the system error gradually tends to zero, and the control effect of the system is gradually improved.

[0083] In conclusion, the reference model with good control performance and the BP neural network controller capable of approximating any nonlinear system are designed, the weight of the neural network controller is adjusted by the output error of the reference model and the controlled object. The control performance of the nonlinear controlled object is simulated and verified, the simulation result shows that the method has good nonlinear approximation performance, reduces the harmonic increase, waveform distortion and the like caused by various nonlinear loads (measuring and mapping equipment), has good control effect, and reduces the electromagnetic interference of the inverter power supply device to the end equipment of the unmanned ship.

[0084] Although the embodiments and the drawings of the present application are disclosed for the purpose of illustration, those skilled in the art can understand that various substitutions, transformations and modifications are possible without departing from the spirit of the present application and the appended claims, therefore, the scope of the present application is not limited to the disclosed contents of the embodiments and the drawings.

Claims

1. A control method of an inverter power supply suitable for a water surface unmanned vehicle, characterized in that: The method relates to an inverter power supply control device which comprises a storage battery, a DC-DC step-up conversion unit, an LC filter unit, a DC-AC conversion unit, a rectification and filter unit, a voltage and current signal acquisition unit, an STM32F767 chip and a TMS320C6678 chip; the storage battery, the DC-DC step-up conversion unit, the LC filter unit, the DC-AC conversion unit and the rectification and filter unit are electrically connected in sequence; The control method is: converting a direct current 12V or 24V power supply into an alternating current 220V through two processes: DC-DC step-up conversion and DC-AC conversion; the first stage adopts DC-DC conversion, 2 groups of symmetrical push-pull circuits are adopted as main topologies in the DC-DC conversion, a double-transformer form is adopted for a high-frequency transformer, primary side windings are connected in parallel and connected with a group of push-pull circuits respectively, the same time sequence control mode is adopted, and the two groups of switching tubes are driven by PWM waveforms output by an STM32F767 through program control; the two groups of switching tubes are alternately turned on under the PWM driving of the STM32F767, the high-frequency transformer outputs a high-frequency square wave and adopts full-bridge rectification; the signal output by the DC-DC conversion is subjected to LC filtering to obtain a step-up direct current 310V; the second stage adopts DC-AC conversion, a full-bridge inverter circuit is adopted in the DC-AC conversion, a TMS320C6678 chip is adopted to complete the SPWM control task of the full-bridge inverter circuit, a sinusoidal alternating current output is obtained, and the sinusoidal alternating current output is supplied to a use device after rectification and filtering; meanwhile, a voltage and current signal acquisition unit acquires the storage battery voltage, the voltage after DC-DC step-up conversion and the voltage and power after rectification and filtering in real time, and protects the entire inverter circuit from overcurrent, overvoltage, undervoltage and overload; The method is an inverter power supply control method based on a neural network indirect model reference adaptive control; a TMS320C6678 chip controls and outputs an SPWM wave through a BP neural network indirect model reference adaptive control algorithm; The BP neural network indirect model reference adaptive control algorithm comprises a reference model, a controlled object, a neural network identifier and a neural network controller; The neural network identifier is used for establishing a neural network model of direct current to alternating current by using the characteristics of the neural network, introducing the model into a direct current to alternating current control process, identifying the controlled system model by using the neural network, continuously training the neural network controller through test data, making the controlled system output track the reference model output, comparing the alternating current output information with a given reference signal through online monitoring, adjusting the SPWM output, and achieving the output control effect of the direct current to alternating current; the reference model is a reference control model of the direct current to alternating current process; the controlled object is the direct current to alternating current; The neural network controller is used for continuously identifying the controlled object by using the neural network algorithm, continuously training the actual direct current to alternating current data, making the controlled system output track the reference model output, and achieving the output of high-quality alternating current for use of a boat end measurement surveying and mapping equipment of an unmanned ship.

Images