Multi-rotor unmanned aerial vehicle feedforward control method based on dynamic cascade pulse neural network

Abstract

The invention belongs to the technical field of unmanned aerial vehicle control, and particularly relates to a multi-rotor unmanned aerial vehicle feedforward control method based on a dynamic cascade pulse neural network. The method comprises the following steps: S1, constructing a network model of a dynamic cascade pulse neural network based on a pulse signal; s2, constructing a cost function of the dynamic cascade pulse neural network based on the network model and the pulse error; s3, solving the weight of the dynamic cascade pulse neural network; s4, setting a preset similarity threshold value, and determining a dynamic cascade structure learning rule based on the preset similarity threshold value; and S5, obtaining an output signal of the dynamic cascade pulse neural network according to the weight of the dynamic cascade pulse neural network and the dynamic cascade structure learning rule, and realizing feedforward control of the multi-rotor unmanned aerial vehicle based on the output signal. According to the invention, the adaptability and robustness of the multi-rotor unmanned aerial vehicle in a complex flight environment are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of unmanned aerial vehicle control, and in particular relates to a multi-rotor unmanned aerial vehicle feedforward control method based on a dynamic cascade pulse neural network. Background Art

[0002] In recent years, multi-rotor UAVs have shown broad application prospects in the military and civilian fields due to their flexible flight mode, strong maneuverability, and vertical take-off and landing. As the application of multi-rotor UAVs becomes more and more extensive, their flight environment is becoming more and more complex, and they even need to meet the requirements of flying in complex terrains such as high-altitude mountains, jungles, narrow buildings, and lakes. The flight environment often has characteristics such as variable wind disturbances and complex uncertainties, which have an adverse effect on the flight control of multi-rotor UAVs.

[0003] Composite control is based on the feedback controller, and introduces feedforward control to further improve the dynamic response performance of the system, so that it can suppress complex environmental disturbances in a timely manner without waiting for the system to have feedback errors. At present, due to the adaptive, self-organizing and self-learning characteristics of artificial neural networks, many scholars use it as a feedforward controller to realize composite control of UAVs.

[0004] As a third-generation neural network model that is close to the bionic mechanism, the pulse neural network has an operation mode that is closer to the mammalian brain. Compared with artificial neural networks, it has more powerful nonlinear computing, asynchronous event information processing and self-learning capabilities, and achieves more efficient human-like perception, cognition and decision-making capabilities. However, in response to data changes and unstable environments, the pulse neural network has the problem of needing to predict the network structure information and sample ratio distribution. At present, it cannot be well applied in the feedforward control of multi-rotor drone supervision. Summary of the invention

[0005] In view of this, the present invention aims to provide a multi-rotor UAV feedforward control method based on a dynamic cascade pulse neural network, so as to solve the problem that the pulse neural network cannot be well applied in the feedforward control of multi-rotor UAV supervision. The present invention combines the idea of ​​dynamic cascade structure to construct a new network framework, which overcomes the problem that network structure information, sample proportion distribution, etc. must be prior knowledge, thereby improving the adaptability and robustness of multi-rotor UAVs in complex flight environments.

[0006] To achieve the above object, the technical solution created by the present invention is implemented as follows: A multi-rotor UAV feedforward control method based on a dynamic cascade pulse neural network specifically comprises the following steps: S1: Obtain the pulse signal of the multi-rotor drone and build a network model of a dynamic cascade pulse neural network based on the pulse signal. The pulse signal includes an expected pulse signal and an actual pulse signal. S2: Constructing the cost function of dynamic cascade spiking neural network based on network model and pulse signal difference; S3: Solving the weights of the dynamic cascade spiking neural network based on the initial preset parameters and cost function of the dynamic cascade spiking neural network; S4: setting a preset similarity threshold, and determining a dynamic cascade structure learning rule based on the preset similarity threshold, wherein the dynamic cascade structure learning rule is used to dynamically adjust the number of hidden layer neurons, the number of hidden layer layers, and the weights of hidden layer neurons; S5: The output signal of the dynamic cascade pulse neural network is obtained according to the weight of the dynamic cascade pulse neural network and the dynamic cascade structure learning rule, and the feedforward control of the multi-rotor UAV is realized based on the output signal.

[0007] Further, in step S1, the network model of the dynamic cascade pulse neural network for: ; ; ; ; ; in, represents the resting potential, The transposed matrix representing the synaptic weights; , It is an intermediate parameter representing the network model and has no physical meaning. is the intermediate parameter representing the i-th synapse in the network model; represents the weight of the i-th synapse; represents the current kernel of the i-th synapse; represents the unit step function; represents the current moment in the training process; Indicates the latest pulse time emitted by the neuron; represents the pulse signal; n represents the number of synapses; , , and They represent the first synaptic weight, the second synaptic weight, and the nth synaptic weight respectively; represents the membrane capacitance; Represents the time constant of the neuronal membrane.

[0008] Furthermore, in step S2, the cost function is: ; ; ; in, Indicates the pulse signal difference between the expected pulse signal and the actual pulse signal, represents the expected membrane potential, represents the cost function; Represents a network model; represents the arrival time of the i-th expected pulse signal; represents the arrival time of the i-th actual pulse signal; Indicates the number of expected pulse signals; Indicates the number of actual pulse signals; Represents the total training time.

[0009] Furthermore, in step S3, the weight of the dynamic cascade spiking neural network is: ; ; ; ; ; ; in, represents the expected weight; Indicates the weight update increment; represents the cost function; express The approximate gradient of represents the error function between the expected pulse signal and the actual pulse signal, represents the update rate coefficient; represents the approximate impulse response signal; represents the membrane capacitance; Represents the current moment of the training process; represents the time estimation constant of the neuronal membrane; It is an intermediate parameter representing the approximate impulse response signal and has no physical meaning.

[0010] Furthermore, step S4 specifically includes the following steps: S41: Setting similarity threshold , the threshold of the number of neurons in a single hidden layer and hidden layer preset values; S42: Add neuron k to the hidden layer of the dynamic cascade pulse neural network, and calculate the similarity between the weight of neuron k and the weights of the original neurons in the current hidden layer: ; in, Represents the newly added neuron k and the original i-th neuron The weighted similarity of Represents the weight of the newly added neuron k; represents the weight of the original i-th neuron of the dynamic cascade pulse neural network; represents a constant; Indicates the number of neurons in the current hidden layer; S43: Determine similarity Is it less than the similarity threshold? , if yes, execute step S44, otherwise execute step S46; S44: Determine whether the number of neurons in the hidden layer is less than or equal to the neuron number threshold , if yes, then add a new neuron in the hidden layer to record the new information and execute step S47, otherwise execute step S45; S45: adding a new hidden layer in the dynamic cascade pulse neural network, adding new neurons to the new hidden layer, replacing the hidden layer of step S44 with the new hidden layer, and executing step S47; S46: merging the newly added neuron currently being calculated with the original neuron with the highest weight similarity, and updating the weight of the merged neuron; S47: Replace neuron k with the newly added neuron k+1, and repeat steps S42 to S46 until the number of hidden layers of the current dynamic cascade pulse neural network is greater than the preset value of the hidden layer, completing the dynamic adjustment of the number of hidden layer neurons, the number of hidden layer layers, and the weights of hidden layer neurons.

[0011] Furthermore, in step S46, the combined neuron weights are updated by the following formula: ; in, Indicates that neuron k is merged into the neuron with the highest similarity The weight after Indicates the number of neurons in the current hidden layer; represents the weight of neuron k.

[0012] Furthermore, in step S1, the expected pulse signal of the multi-rotor UAV is obtained based on the expected trajectory, and the actual pulse signal is obtained based on the actual flight trajectory of the multi-rotor UAV.

[0013] Further, in step S2, the pulse signal difference is the difference between the expected pulse signal and the actual pulse signal.

[0014] Furthermore, in step S3, the initial preset parameters of the dynamic cascade pulse neural network include an initial value of the number of hidden layer neurons, an initial value of the number of hidden layer layers, and an initial value of the hidden layer neuron weights, wherein the initial value of the number of hidden layer layers is set to 1.

[0015] Compared with the prior art, the invention can achieve the following beneficial effects: (1) The present invention creates a multi-rotor UAV feedforward control method based on a dynamic cascade pulse neural network. Combining the dynamic cascade concept, a new type of dynamic cascade pulse neural network is proposed, which can overcome the problem that the structural information and sample ratio distribution of the pulse neural network must be prior knowledge. The multi-rotor UAV desired trajectory and the feedback controller output are used as sample inputs for training to achieve dynamic cascade pulse neural network supervised feedforward control, which provides a new idea for multi-rotor UAV feedforward control.

[0016] (2) The present invention creates a multi-rotor UAV feedforward control method based on a dynamic cascade pulse neural network, designs a supervisory weight and a dynamic cascade structure learning mechanism, and dynamically and adaptively increases hidden layers and hidden layer neurons to extract more input external information center clusters, thereby ensuring that the multi-rotor UAV can adapt to the ever-changing unstable environment. When learning new information, catastrophic interference or forgetting will be avoided. Even if the input data changes continuously, it will not affect the normal operation of the dynamic cascade pulse neural network, thereby greatly improving the stability and robustness of the supervised feedforward control. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings constituting part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments and descriptions of the present invention are used to explain the present invention and do not constitute an improper limitation on the present invention. In the drawings: Figure 1 A schematic diagram of a flow chart of a multi-rotor UAV feedforward control method based on a dynamic cascade pulse neural network according to an embodiment of the present invention; Figure 2 A schematic diagram of a feedforward control structure of a multi-rotor UAV based on a dynamic cascade pulse neural network according to an embodiment of the present invention; Figure 3 A schematic diagram of the dynamic evolution process of the hidden layer described in the embodiment of the present invention; Figure 4 A schematic diagram of the dynamic evolution process of hidden layer neurons described in an embodiment of the present invention; Figure 5 This is a schematic diagram of longitude tracking of a twelve-rotor drone under outdoor level 4 wind disturbance; Figure 6 This is a schematic diagram of latitude tracking of a twelve-rotor drone under outdoor level 4 wind disturbance; Figure 7 This is a schematic diagram of the altitude tracking of a twelve-rotor drone under outdoor level 4 wind disturbance; Figure 8 This is a schematic diagram of the pitch angle tracking of a twelve-rotor UAV under level 4 wind disturbance outdoors; Fig. 9 The schematic diagram of the roll angle tracking of a twelve-rotor UAV under outdoor level 4 wind disturbance; Fig.10 Schematic diagram of yaw angle tracking of a twelve-rotor UAV under level 4 wind disturbance outdoors. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and do not constitute a limitation of the invention.

[0019] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0020] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0021] In the description of the invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the invention can be understood according to specific circumstances.

[0022] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0023] like Figure 1-Figure 2 As shown, the present invention proposes a multi-rotor UAV feedforward control method based on a dynamic cascade pulse neural network, which specifically includes the following steps: S1: obtaining a pulse signal of the multi-rotor UAV, and constructing a network model of the dynamic cascade pulse neural network based on the pulse signal, wherein the pulse signal includes an expected pulse signal and an actual pulse signal; S2: constructing a cost function of the dynamic cascade pulse neural network based on the network model and the pulse signal difference; S3: solving the weight of the dynamic cascade pulse neural network based on the initial preset parameters of the dynamic cascade pulse neural network and the cost function; S4: setting a preset similarity threshold, and determining a dynamic cascade structure learning rule based on the preset similarity threshold, the dynamic cascade structure learning rule is used to dynamically adjust the number of hidden layer neurons, the number of hidden layer layers and the weights of hidden layer neurons; S5: obtaining the output signal of the dynamic cascade pulse neural network according to the weight of the dynamic cascade pulse neural network and the dynamic cascade structure learning rule, and realizing feedforward control of the multi-rotor UAV based on the output signal.

[0024] The present invention first establishes a dynamic cascade pulse neural network model; determines the cost function of the dynamic cascade pulse neural network, and then designs a supervised weight learning rule (the weight of the dynamic cascade pulse neural network); determines the dynamic cascade structure learning rule, and finally realizes the dynamic cascade pulse neural network supervised feedforward control, thereby improving the dynamic control performance and anti-disturbance capability of the multi-rotor UAV in a complex flight environment. The multi-rotor UAV trajectory tracking supervised feedforward control method based on the dynamic cascade pulse neural network provides a new idea for the research on the trajectory tracking control of multi-rotor UAVs, and has certain theoretical reference significance and practical application value.

[0025] In some embodiments, in step S1, the network model of the dynamic cascade pulse neural network is: ; ; ; ; ; in, represents the resting potential, The transposed matrix representing the synaptic weights; , It is an intermediate parameter representing the network model and has no physical meaning. is the intermediate parameter representing the i-th synapse in the network model; represents the weight of the i-th synapse; represents the current kernel of the i-th synapse; represents the unit step function; represents the current moment in the training process; Indicates the latest pulse time emitted by the neuron; represents a pulse signal (if the input pulse is an expected pulse, the expected pulse signal is obtained; if the input pulse is an actual pulse, the actual pulse signal is obtained. There is no need to confirm the signal type here); n represents the number of synapses; , , and They represent the first synaptic weight, the second synaptic weight, and the nth synaptic weight respectively; represents the membrane capacitance; Represents the time constant of the neuronal membrane.

[0026] In some embodiments, in step S2, the cost function is: ; ; ; in, Indicates the pulse signal difference between the expected pulse signal and the actual pulse signal, represents the expected membrane potential, represents the cost function; Represents a network model; represents the arrival time of the i-th expected pulse signal; represents the arrival time of the i-th actual pulse signal; Indicates the number of expected pulse signals; Indicates the number of actual pulse signals; Represents the total training time.

[0027] In some embodiments, in step S3, the weights of the dynamic cascade spiking neural network are: ; ; ; ; ; ; in, represents the expected weight; Indicates the weight update increment; represents the cost function; express The approximate gradient of represents the error function between the expected pulse signal and the actual pulse signal, represents the update rate coefficient; represents the approximate impulse response signal; represents the membrane capacitance; Represents the current moment of the training process; represents the time estimation constant of the neuronal membrane; It is an intermediate parameter representing the approximate impulse response signal and has no physical meaning.

[0028] It should be noted that, due to the discontinuous jump in the weight space, for the sake of simplicity, the cost function is minimized at each moment.

[0029] In some embodiments, step S4 specifically includes the following steps: S41: Setting a similarity threshold , the threshold of the number of neurons in a single hidden layer and hidden layer preset values; S42: Add neuron k to the hidden layer of the dynamic cascade pulse neural network, and calculate the similarity between the weight of neuron k and the weights of the original neurons in the current hidden layer: ; in, Represents the newly added neuron k and the original i-th neuron The weighted similarity of Represents the weight of the newly added neuron k; represents the weight of the original i-th neuron of the dynamic cascade pulse neural network; represents a constant; Indicates the number of neurons in the current hidden layer; S43: Determine similarity Is it less than the similarity threshold? , if yes, execute step S44, otherwise execute step S46; S44: Determine whether the number of neurons in the hidden layer is less than or equal to the neuron number threshold , if yes, then add a new neuron in the hidden layer to record the new information and execute step S47, otherwise execute step S45; S45: adding a new hidden layer in the dynamic cascade pulse neural network, adding new neurons to the new hidden layer, replacing the hidden layer of step S44 with the new hidden layer, and executing step S47; S46: merging the newly added neuron currently being calculated with the original neuron with the highest weight similarity, and updating the weight of the merged neuron; S47: Replace neuron k with the newly added neuron k+1, and repeat steps S42 to S46 until the number of hidden layers of the current dynamic cascade pulse neural network is greater than the preset value of the hidden layer, completing the dynamic adjustment of the number of hidden layer neurons, the number of hidden layer layers, and the weights of hidden layer neurons.

[0030] It should be noted that, in step S44, if the similarity Less than the similarity threshold , and the number of neurons in the hidden layer of the current dynamic cascade pulse neural network is not greater than the threshold number of neurons , which means that there is a large difference between the information in the network and the input sample. New neurons are added to the hidden layer to record the new information, thereby determining the number of neurons in the hidden layer.

[0031] If the similarity Less than the similarity threshold , and the number of neurons in the hidden layer is greater than the threshold number of neurons , the newly added neurons are added to the new hidden layer. This is because the neurons in the same hidden layer have similar functions. Adding too many neurons will lead to repeated network training, which is not conducive to the stability of the network. If the number of dynamically cascaded hidden layers exceeds the preset value, the network learning optimization is terminated to avoid catastrophic network expansion, thereby determining the number of hidden layers.

[0032] Further, in step S45, when the input sample information is sufficiently similar to the information in the original neurons of the network, that is, the similarity is equal to or greater than , then the new neuron will be merged with the neuron with the highest weight similarity to determine the weight of the hidden layer neuron.

[0033] Furthermore, the similarity threshold , the threshold of the number of neurons in a single hidden layer The preset values ​​of the hidden layer are determined according to user requirements.

[0034] In some embodiments, in step S46, the merged neuron weights are updated by the following formula: ; in, Indicates that neuron k is merged into the neuron with the highest similarity The weight after Indicates the number of neurons in the current hidden layer; represents the weight of neuron k.

[0035] In some embodiments, in step S1, the expected pulse signal of the multi-rotor drone is obtained based on the expected trajectory, and the actual pulse signal is obtained based on the actual flight trajectory of the multi-rotor drone.

[0036] In some embodiments, in step S2 , the pulse signal difference is the difference between the expected pulse signal and the actual pulse signal.

[0037] In some embodiments, in step S3, the initial preset parameters of the dynamic cascade pulse neural network include an initial value of the number of hidden layer neurons, an initial value of the number of hidden layer layers, and an initial value of the hidden layer neuron weights, wherein the initial value of the number of hidden layer layers is set to 1.

[0038] Taking a twelve-rotor UAV as an example, the multi-rotor UAV feedforward control method based on a dynamic cascade pulse neural network provided by the present invention is described below: To verify the robustness of the present invention, a twelve-rotor UAV trajectory tracking flight experiment was conducted outdoors in a level 4 wind disturbance weather. The maximum instantaneous wind speed in the flight environment was measured using an American Omega anemometer and was 11 m / s. The code for the trajectory tracking supervised feedforward control method of the present invention was written and input into the twelve-rotor airborne control main chip to complete the trajectory tracking flight experiment of the twelve-rotor UAV. The dynamic cascade pulse neural network parameters were set as follows: , the initial value of the hidden layer is 1, the threshold of the hidden layer neurons .

[0039] like Figure 3-Figure 4 As shown, through the learning rule mechanism, the number of hidden layers finally converges to 4, and the hidden neurons in each hidden layer converge to 30, 30, 30 and 22. Figure 5-Figure 7 The tracking results of a twelve-rotor UAV in longitude, latitude and altitude based on a dynamic cascade spiking neural network are described. Figure 8-Figure 10 The attitude tracking results of the twelve-rotor UAV are described, which have good tracking performance. It can be seen that based on the feedforward control method provided by the present invention, the trajectory tracking errors of the twelve-rotor UAV in three directions are all within ±0.8 meters under level 4 wind disturbance. Experiments show that the method designed by the present invention can ensure that the twelve-rotor UAV has excellent trajectory tracking control performance, and can resist external level 4 wind disturbance, and has strong robustness.

[0040] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the disclosure of the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0041] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network, characterized in that: The specific steps include: S1: Obtain the pulse signal of the multi-rotor drone and build a network model of a dynamic cascade pulse neural network based on the pulse signal. The pulse signal includes an expected pulse signal and an actual pulse signal. S2: Constructing the cost function of dynamic cascade spiking neural network based on network model and pulse signal difference; S3: Solving the weights of the dynamic cascade spiking neural network based on the initial preset parameters and cost function of the dynamic cascade spiking neural network; S4: setting a preset similarity threshold, and determining a dynamic cascade structure learning rule based on the preset similarity threshold, wherein the dynamic cascade structure learning rule is used to dynamically adjust the number of hidden layer neurons, the number of hidden layer layers, and the weights of hidden layer neurons; S5: The output signal of the dynamic cascade pulse neural network is obtained according to the weight of the dynamic cascade pulse neural network and the dynamic cascade structure learning rule, and the feedforward control of the multi-rotor UAV is realized based on the output signal.

2. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 1 is characterized in that: In step S1, the network model of the dynamic cascade pulse neural network for: ； ； ； ； ； in, represents the resting potential, The transposed matrix representing the synaptic weights; , It is an intermediate parameter representing the network model and has no physical meaning. is the intermediate parameter representing the i-th synapse in the network model; represents the weight of the i-th synapse; represents the current kernel of the i-th synapse; represents the unit step function; represents the current moment in the training process; Indicates the latest pulse time emitted by the neuron; represents the pulse signal; n represents the number of synapses; , , and They represent the first synaptic weight, the second synaptic weight, and the nth synaptic weight respectively; represents the membrane capacitance; Represents the time constant of the neuronal membrane.

3. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 1 is characterized in that: In step S2, the cost function is: ； ； ； in, Indicates the pulse signal difference between the expected pulse signal and the actual pulse signal, represents the expected membrane potential, represents the cost function; Represents a network model; represents the arrival time of the i-th expected pulse signal; represents the arrival time of the i-th actual pulse signal; Indicates the number of expected pulse signals; Indicates the number of actual pulse signals; Represents the total training time.

4. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 3 is characterized in that: In step S3, the weight of the dynamic cascade spiking neural network is: ； ； ； ； ； ； in, represents the expected weight; Indicates the weight update increment; represents the cost function; express The approximate gradient of represents the error function between the expected pulse signal and the actual pulse signal, represents the update rate coefficient; represents the approximate impulse response signal; represents the membrane capacitance; Represents the current moment of the training process; represents the time estimation constant of the neuronal membrane; It is an intermediate parameter representing the approximate impulse response signal and has no physical meaning.

5. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 1 is characterized in that: Step S4 specifically includes the following steps: S41: Setting similarity threshold , the threshold of the number of neurons in a single hidden layer and hidden layer preset values; S42: Add neuron k to the hidden layer of the dynamic cascade pulse neural network, and calculate the similarity between the weight of neuron k and the weights of the original neurons in the current hidden layer: ； in, Represents the newly added neuron k and the original i-th neuron The weighted similarity of Represents the weight of the newly added neuron k; represents the weight of the original i-th neuron of the dynamic cascade pulse neural network; represents a constant; Indicates the number of neurons in the current hidden layer; S43: Determine similarity Is it less than the similarity threshold? , if yes, execute step S44, otherwise execute step S46; S44: Determine whether the number of neurons in the hidden layer is less than or equal to the neuron number threshold , if yes, then add a new neuron in the hidden layer to record the new information and execute step S47, otherwise execute step S45; S45: adding a new hidden layer in the dynamic cascade pulse neural network, adding new neurons to the new hidden layer, replacing the hidden layer of step S44 with the new hidden layer, and executing step S47; S46: merging the newly added neuron currently being calculated with the original neuron with the highest weight similarity, and updating the weight of the merged neuron; S47: Replace neuron k with the newly added neuron k+1, and repeat steps S42 to S46 until the number of hidden layers of the current dynamic cascade pulse neural network is greater than the preset value of the hidden layer, completing the dynamic adjustment of the number of hidden layer neurons, the number of hidden layer layers, and the weights of hidden layer neurons.

6. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 5 is characterized in that: In step S46, the combined neuron weights are updated by the following formula: ； in, Indicates that neuron k is merged into the neuron with the highest similarity The weight after Indicates the number of neurons in the current hidden layer; represents the weight of neuron k.

7. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 1 is characterized in that: In step S1, the expected pulse signal of the multi-rotor UAV is obtained based on the expected trajectory, and the actual pulse signal is obtained based on the actual flight trajectory of the multi-rotor UAV.

8. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 1 is characterized in that: In step S2, the pulse signal difference is the difference between the expected pulse signal and the actual pulse signal.

9. The multi-rotor UAV feedforward control method based on dynamic cascade pulse neural network according to claim 1 is characterized in that: In step S3, the initial preset parameters of the dynamic cascade pulse neural network include an initial value of the number of hidden layer neurons, an initial value of the number of hidden layer layers, and an initial value of the weights of the hidden layer neurons, wherein the initial value of the number of hidden layer layers is set to 1.

Images