Weighing adaptive vibration control system and method

Abstract

The invention discloses a weighing self-adaptive vibration control system and method, and belongs to the technical field of weighing. Vibrator state information, weighing sensor output signals, sliding mode face values, errors and derivatives of the errors are collected. And then determining a sliding mode surface form and a control law according to a system dynamics model and a control target. Meanwhile, a neural network multi-layer perceptron structure is adopted, an input layer receives a sliding mode surface value, an error and a derivative signal thereof, a hidden layer processes input information through nonlinear transformation and feature extraction, and an output layer outputs switching gain and a sliding mode surface coefficient. And finally, training the neural network through an online learning algorithm. In the system operation process, according to the output error of the weighing sensor, the weight and bias of the neural network are adjusted based on an optimization algorithm, and the parameters of the sliding mode controller are continuously optimized. And the sliding mode controller generates an accurate control signal according to the optimized parameters, vibration of the vibrator is accurately controlled, efficient and stable operation of the weighing process under vibration interference is achieved, and the weighing performance is improved.

Description

Technical Field

[0001] The invention relates to a weighing self-adaptive vibration control system and method, belonging to the technical field of weighing. Background Art

[0002] During the material conveying process for weighing anode copper sample scraps, vibrators are often used to assist material flow, but the mechanical vibrations they generate are transmitted to the weighing sensor, causing fluctuations in the sensor output signal and inaccurate weighing results. Traditional control methods have difficulty adapting to nonlinear, time-varying vibration interference environments, and fixed parameters result in insufficient robustness. In existing technologies, sliding mode control is used for interference resistance due to its strong robustness, but the fixed setting of its switching gain is prone to high-frequency jitter problems; and although neural networks have adaptive capabilities, they lack deep integration with the controller. Therefore, a system that combines sliding mode control with online parameter adjustment of neural networks is needed to achieve high-precision weighing under vibration interference. Summary of the Invention

[0003] In order to overcome the problems in the background technology, the present invention provides a weighing adaptive vibration control system and method based on the problems in the weighing process of anode copper sample chips.

[0004] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0005] A weighing adaptive vibration control method comprises the following steps:

[0006] (1) Collect data from the vibrator and weighing sensor and calculate the sliding surface value s(t);

[0007] (2) Construct a neural network online learning model. First, the sliding mode surface value, error and its derivative are input into the forward propagation output sliding mode controller required switching gain K sw and sliding surface coefficient ε; then calculate the error function J according to the output error of the weighing sensor; finally, the error signal δ of the hidden layer is obtained by back propagation based on the error function J h , and according to the error signal δ of the hidden layer h , using an optimization algorithm to update the weights and biases of the neural network;

[0008] (3) Determine the sliding mode control rate based on the system's dynamic model and control objectives.

[0009] More preferably, the data of the vibrator includes state information of the vibrator; the state information of the vibrator includes: displacement x(t) and velocity of the vibrator The data of the weighing sensor includes an output signal of the weighing sensor.

[0010] More preferably, the sliding surface value s(t) is

[0011] The error e(t) is the difference between the actual output of the weighing sensor and the actual material weight; the error derivative The first-order difference method is used, that is, Where Δt is the sampling time interval, t is the sampling time, and ε is the sliding surface coefficient.

[0012] More preferably, the forward propagation specifically includes the following steps:

[0013] Input layer processing: The input layer receives the sliding surface value s(t), error e(t) and its derivative As the input signal, the output vector of the input layer is output; the input signal needs to be normalized and its value range is mapped to the [0,1] interval to improve the training efficiency and stability of the neural network.

[0014] Hidden layer processing: The hidden layer uses the ReLU activation function for nonlinear transformation and feature extraction: the output vector of the input layer is multiplied by the weight matrix of the hidden layer, and then the bias vector of the hidden layer is added to obtain the linear input z of the hidden layer. h ; Then, the linear input z of the hidden layer h Apply the ReLU activation function for nonlinear transformation, i.e. a h =ReLU(z h ), through this nonlinear transformation, the neural network can learn the complex features in the data.

[0015] Output layer processing: output vector a of the hidden layer h Multiplying it with the weight matrix of the output layer and adding the bias vector of the output layer, we get the linear input z of the output layer. o ; In this system, the output layer does not use additional activation functions, z o is the final output, corresponding to the switching gain K of the sliding mode controller sw and the sliding surface coefficient ε is z o1 is the linear input of the first neuron in the output layer, z o2 is the linear input of the second neuron in the output layer, z o1 and z o2 They are at the same level and together constitute the input of the output layer.

[0016] More preferably, the error function J is:

[0017] Among them, y i is the actual output of the weighing sensor, is the expected output (i.e., the true weight), and N is the number of sampling points. In actual calculations, as the system runs, the number of sampling points N is continuously updated, and the value of the error function J is calculated in real time.

[0018] More preferably, the back propagation specifically includes:

[0019] Output layer error signal calculation: According to the error function J, the output layer linear input z o The error signal δ of the output layer is calculated by taking the partial derivative of o ; Since the output layer uses a linear activation function, δ o Expressed as

[0020] Hidden layer error signal calculation: According to the error signal δ of the output layer o And the weight matrix W from the hidden layer to the output layer o , calculate the error signal δ of the hidden layer h is δ h =W o δ o ⊙ReLU′(z h ), where ⊙ represents element-wise multiplication, ReLU′(z h ) is the derivative of the ReLU activation function.

[0021] Weight and bias update: Use an optimization algorithm such as gradient descent to update the weights and biases of the neural network. This algorithm combines the principles of momentum and adaptive learning rates to automatically adjust the learning rate during training. The weight and bias update formula is as follows:

[0022] Weight update:

[0023] Bias update:

[0024] Where α is the learning rate, and are the gradients of the error function J with respect to weights and biases, respectively.

[0025] More preferably, the sliding mode control rate is calculated based on the sliding mode surface value s(t) = K sw ·sign(s(t))+K1x(t);

[0026] where K sw is the switching gain; K1 is the state feedback gain matrix; sgn(s(t)) is the sign function, which is defined as:

[0027]

[0028] The present invention also claims protection for the weighing adaptive vibration control system, comprising: a vibrator, a weighing sensor, a sliding mode controller, a neural network online learning module, a data acquisition module and a calculation module;

[0029] The vibrator is used to assist in material transportation; the weighing sensor is used to measure the weight of the material and convert the weight information into an electrical signal output;

[0030] The data acquisition module is responsible for collecting the status information of the vibrator, the output signal of the weighing sensor, the sliding mode surface value, the error and its derivative data in real time, and transmitting these data to the neural network online learning module and the sliding mode controller.

[0031] The neural network online learning module receives the data from the data acquisition module and uses the neural network. The input layer receives a sliding mode surface value, error and its derivative as input signals, and the output layer outputs the parameters K required by the sliding mode controller. sw and ε; then, through the online learning algorithm, the weights and bias of the neural network are adjusted in real time according to the output error of the weighing sensor to optimize the parameters of the sliding mode controller.

[0032] The sliding mode controller generates a control signal based on the control parameters output by the neural network online learning module to accurately control the vibration of the vibrator. The sliding mode controller is based on the control law designed on the sliding surface, which can make the system state reach the sliding surface within a finite time and maintain movement on the sliding surface, thereby effectively suppressing vibration interference.

[0033] The calculation module is used for sliding surface calculation, switching gain calculation, and data output to the neural network online learning module.

[0034] Beneficial effects of the present invention: The present invention realizes effective suppression of vibration interference of the vibrator through the collaborative work of multiple modules, thereby improving the accuracy and stability of weighing. First, the data acquisition module is used to collect the status information of the vibrator and the output signal of the weighing sensor in real time, and the calculation module is used to calculate the sliding surface value, error and its derivative, and the collected data is pre-processed by filtering and amplification. Secondly, an adaptive control system based on sliding mode control is constructed. The sliding mode controller is designed to determine the form and control law of the sliding surface according to the dynamic model and control objectives of the system. At the same time, a neural network online learning module is built. The module adopts a multi-layer perceptron structure. The input layer receives the sliding surface value, error and its derivative as input signals, the hidden layer processes the input information through nonlinear transformation and feature extraction, and the output layer outputs the switching gain and sliding surface coefficient required by the sliding mode controller. Finally, the neural network is trained by an online learning algorithm. During the operation of the system, the weights and biases of the neural network are adjusted according to the output error of the weighing sensor using an algorithm based on gradient descent, so that the neural network can continuously optimize the parameters of the sliding mode controller. The sliding mode controller can generate accurate control signals based on the optimized parameters, accurately control the vibration of the vibrator, and ultimately achieve efficient and stable operation of the weighing process under vibration interference, effectively improving the weighing performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of the model structure of the weighing adaptive vibration control system of the present invention.

[0036] Figure 2 The figure is a flow chart of the weighing adaptive vibration control according to the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described in detail below with reference to specific embodiments, but the protection scope of the present invention is not limited thereto.

[0038] Example 1

[0039] A weighing adaptive vibration control system, comprising: a vibrator, a weighing sensor, a sliding mode controller, a neural network online learning module, a data acquisition module and a calculation module;

[0040] The vibrator is used to assist in material transportation, but the vibration generated during operation may interfere with the weighing sensor.

[0041] Load cells are used to measure the weight of materials and convert the weight information into electrical signals for output.

[0042] The data acquisition module is responsible for collecting the status information of the vibrator, the output signal of the weighing sensor, the sliding mode surface value, the error and its derivative data in real time, and transmitting these data to the neural network online learning module and the sliding mode controller.

[0043] The neural network online learning module receives the data from the data acquisition module and uses the neural network. The input layer receives a sliding mode surface value, error and its derivative as input signals, and the output layer outputs the parameters K required by the sliding mode controller. sw and ε; then, through the online learning algorithm, the weights and bias of the neural network are adjusted in real time according to the output error function of the weighing sensor to optimize the parameters of the sliding mode controller.

[0044] The sliding mode controller is based on the control parameter z output by the neural network online learning module. o1 and z o2 , generate control signals, and accurately control the vibration of the vibrator. The sliding mode controller designs the control law based on the sliding surface, which can make the system state reach the sliding surface within a finite time and keep moving on the sliding surface, thereby effectively suppressing vibration interference.

[0045] The calculation module is used for sliding surface calculation, switching gain calculation, and data output to the neural network online learning module.

[0046] Example 2

[0047] like Figure 1-2 As shown, a weighing adaptive vibration control system based on a neural network optimized sliding mode control designed in Example 1 performs weighing adaptive vibration control, including the following steps:

[0048] S1: Status information collection

[0049] The status information of the vibrator and weighing sensor is collected by the data acquisition module.

[0050] The data acquisition module is equipped with a displacement sensor and a velocity sensor. The displacement sensor is typically a laser displacement sensor, which utilizes the reflection properties of laser light to accurately calculate the vibrator's displacement by measuring the time difference between laser emission and reception. The velocity sensor is a Hall effect velocity sensor. Based on the Hall effect, when the vibrator moves, it causes a change in the magnetic field. The Hall element converts this magnetic field change into an electrical signal, thereby measuring the vibrator's velocity. These sensors continuously collect the vibrator's displacement and velocity data at a high sampling rate.

[0051] The load cell uses a strain gauge load cell. Its operating principle is that when the weight of the material is applied, the elastic body deforms, causing the resistance of the strain gauge attached to the elastic body to change. By measuring this resistance change, an electrical signal proportional to the material weight is generated. The data acquisition module converts this signal into digital form to obtain accurate weighing data.

[0052] In order to describe the system in detail, the acquisition process is presented through the model equation:

[0053] Establishment of system dynamic model:

[0054] The dynamic equation of the vibrator is:

[0055]

[0056] Where M is the equivalent mass of the relevant vibrating component, C is the damping coefficient, k is the stiffness coefficient, F(t) is the control signal output by the sliding mode controller, and x(t) is the displacement of the vibrator. is the speed of the vibrator, is the acceleration of the vibrator.

[0057] Load cell model:

[0058] The output of the weighing sensor is expressed as y(t), and the actual weight of the material is W real , plus the error e(t) caused by vibration interference, that is: y(t) = W real +e(t).

[0059] The state space representation of the load cell:

[0060] Define the state variable x1=x, The state equation of the weighing sensor can be expressed as:

[0061]

[0062] Where x1 is the displacement of the vibrator, is the rate of change of the vibrator displacement over time, x2 is the speed of the vibrator, is the rate of change of the vibrator speed, and k is the stiffness coefficient.

[0063] Written in matrix form:

[0064]

[0065] where X(t) = [x1(t), x2(t)] T ,

[0066] The output equation is: y(t) = Dx(t) + W real

[0067] Where D = [0, 1], X(t) is the state space representation of the system, x1(t) is the displacement of the vibrator at time t, and x2(t) is the velocity of the vibrator at time t.

[0068] S2: Parameter calculation

[0069] Rely on the calculation module to calculate the sliding surface value s(t): According to the pre-designed sliding surface formula You need to calculate the error e(t) and its derivative first The error e(t) is the difference between the actual output of the load cell and the preset material weight.

[0070] Error derivative The first-order difference method is used, that is, Where Δt is the sampling time interval, t is the sampling time, and ε is the sliding surface coefficient.

[0071] The collected and calculated data is transmitted to the neural network online learning module and sliding mode controller via an industrial-grade communication bus. The communication bus is highly reliable and has strong anti-interference capabilities, ensuring that data is not lost or errors during transmission.

[0072] S3: Online Learning of Neural Networks

[0073] (1) Design of system neural network:

[0074] 1. Neural network structure:

[0075] Input layer: 3 nodes: sliding surface value s(t), error e(t) and its derivative

[0076] Hidden layer: There are two hidden layers, and the activation function is ReLU, which is used to perform nonlinear transformation and feature extraction on the information collected from the data.

[0077] Output layer: 2 nodes (sliding mode control parameters: switching gain K sw , sliding surface coefficient ε).

[0078] 2. Data conversion from input layer to hidden layer:

[0079] Linear combination: The output of the input layer is multiplied by the weight matrix of the hidden layer and then added with the bias vector to achieve a linear combination.

[0080] The output vector of the input layer is The weight matrix of the hidden layer is W h , whose dimension is 3×10, the element w in the weight matrix ij represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer; the bias vector of the hidden layer is b h =[b h1 ,b h2 ,…,b h10 ] T .

[0081] Then the linear input z of the jth neuron in the hidden layer is hjExpressed as:

[0082] where x i (i=1,2,3) represents the input of the i-th neuron in the hidden layer, b hj is the bias vector of the jth neuron in the hidden layer.

[0083] Expressed in matrix form:

[0084] in is the linear input vector of all neurons in the hidden layer, x is a weight matrix, and x is the output vector of the input layer.

[0085] Nonlinear transformation: hidden layer input z obtained by linear combination h Nonlinear transformation is performed through the ReLU activation function.

[0086] The mathematical expression of the ReLU function is: f(z) = max(0,z), where z is the variable of the function and z is the hidden layer input. h , max(0,z) is the operation of taking the maximum value.

[0087] The linear input z to each neuron in the hidden layer hj Apply the ReLU activation function to get the output a of the jth neuron in the hidden layer hj :

[0088] a hj =ReLU(z hj )=max(0,z hj ).

[0089] Expressed in vector form: a h =ReLU(z h );

[0090] where a h =[a h1 ,a h2 ,…,a h10 ] T is the output vector of all neurons in the hidden layer.

[0091] Data conversion from hidden layer to output layer:

[0092] Linear combination: the output of the hidden layer a h and the weight matrix W of the output layer o Perform multiplication and then add the bias vector b of the output layer o The weight matrix W of the output layer o The dimension is 10×2, the bias vector b o =[bo1 ,b o2 ] T The linear input z of the kth neuron in the output layer is ok Expressed as:

[0093] where w jk is the connection weight from the jth neuron in the input layer to the kth neuron in the hidden layer, b ok is the bias of the kth neuron in the output layer.

[0094] Expressed in matrix form:

[0095] where z o =[z o1 ,z o2 ] T is the linear input vector of all neurons in the output layer, z o1 is the linear input of the first neuron in the output layer, z o2 is the linear input of the second neuron in the output layer, is a weight matrix.

[0096] Output:

[0097] z o is the final output, corresponding to the switching gain K of the sliding mode controller sw And the sliding surface coefficient ε, that is:

[0098]

[0099] 3. Training goal: The training goal is to minimize the fluctuation of the weighing sensor output signal, that is, to minimize the error function J:

[0100]

[0101] where y i is the actual output of the weighing sensor, is the desired output (i.e. the true weight W real ), where N is the number of sampling points. In actual calculations, as the system runs, the number of sampling points N is continuously updated, and the value of the error function J is calculated in real time.

[0102] (2) Online learning algorithm

[0103] An online learning algorithm based on gradient descent is used. The specific steps are as follows:

[0104] 1. Data Collection

[0105] During the operation of the weighing adaptive vibration control system, the data acquisition module is responsible for collecting various necessary data in real time, including:

[0106] A sliding surface value s(t): reflects the distance between the current state of the system and the sliding surface, and is an important indicator in sliding mode control.

[0107] Error e(t): The difference between the actual output y(t) of the weighing sensor and the expected output (i.e. the real weight W real ) is calculated, that is, e(t)=y(t)-W real The error reflects the difference between the control effect of the system and the expected target.

[0108] Error derivative The rate of change of the error over time can provide information about the dynamic changes of the system and help to adjust the control parameters more accurately.

[0109] The output y(t) of the weighing sensor is used to calculate the error and evaluate the control performance of the system.

[0110] The collected data will serve as the input of the neural network, providing a basis for subsequent calculations and parameter adjustments.

[0111] Forward propagation: The collected sliding surface value, error, and its derivative are used as input signals and fed into the neural network. These signals are first normalized and their value range is mapped to the interval [0, 1] to improve the training efficiency and stability of the neural network. The forward propagation process is the calculation process of data starting from the input layer, passing through the hidden layers, and finally reaching the output layer. The details are as follows:

[0112] Input layer to hidden layer: output vector of input layer and the weight matrix W of the hidden layer h Perform multiplication and then add the bias vector b of the hidden layer h , get the linear input of the hidden layer Next, for z h Apply the ReLU activation function for nonlinear transformation to obtain the output a of the hidden layer h =ReLU(z h ). Through this nonlinear transformation, neural networks are able to learn complex features in the data.

[0113] Hidden layer to output layer: output a of the hidden layer h and the weight matrix W of the output layer o Perform multiplication and add the bias vector b of the output layer o , and get the linear input of the output layer In this system, no additional activation function is used in the output layer, z o is the final output, corresponding to the switching gain W of the sliding mode controller realand sliding surface coefficient ε.

[0114] Control input calculation: switching gain K according to the neural network output sw And the sliding surface coefficient ε, calculate the output of the sliding mode controller, that is, the sliding mode control rate F(t) of the sliding mode controller. The output expression of the sliding mode controller is F(t) = K sw sign(s(t))+εX(t), where is the state feedback gain matrix K sw , ε is the state vector of the system, and sign(s(t)) is the sign function. The output F(t) of the sliding mode controller is used to control the vibration of the vibrator.

[0115] Back propagation: Back propagation is the process of calculating the error signal of each layer of the neural network based on the error signal e(t) and updating the weights and biases of the neural network. The specific steps are as follows:

[0116] Calculate the error signal of the output layer: the error signal δ of the output layer o The output layer can be linearly input z according to the error function J o The partial derivative of δ is calculated. o It can be expressed as In this system, since the output layer adopts a linear activation function, δ o Related to the e(t) error.

[0117] Calculate the error signal of the hidden layer: According to the error signal δ of the output layer o and the weight matrix z from the hidden layer to the output layer o , calculate the error signal δ of the hidden layer h The error signal of the hidden layer needs to consider the derivative of the ReLU activation function. The specific calculation formula is δ h =W o δ o ⊙ReLU′(z h ), where ⊙ represents element-wise multiplication, ReLU′(z h ) is the derivative of the ReLU activation function.

[0118] Update weights and biases: Based on the error signals of each layer, the gradient descent algorithm is used to update the weights and biases of the neural network. Combining the ideas of momentum method and adaptive learning rate, the learning rate is automatically adjusted during training to improve training efficiency and stability. The weight and bias update formula is as follows:

[0119] Weight update:

[0120] Bias update:

[0121] Where α is the learning rate, and are the gradients of the error function J with respect to weights and biases, respectively.

[0122] S4: Sliding mode controller control

[0123] The sliding mode controller receives the switching gain K output by the neural network online learning module through the communication bus sw and sliding surface coefficient ε.

[0124] 1. Calculation of sliding mode control law:

[0125] State vector determination: The state vector X(t) of the system usually includes the displacement and velocity of the vibrator, that is,

[0126] Sliding surface definition:

[0127] According to the error e(t) = y(t) - W real Design sliding surface:

[0128] Where ε>0, ε is the sliding surface coefficient.

[0129] According to the received parameters (switch gain K sw and sliding surface coefficient ε) and the pre-designed sliding surface, the sliding mode control law is calculated as:

[0130] F(t)=K sw ·sign(s(t))+K1x(t);

[0131] where K sw is the switching gain, and ε is the sliding surface coefficient, which needs to be dynamically adjusted through the neural network.

[0132] in, is the state error vector, reflecting the difference between the current system state and the desired state. In this system, the state error can be calculated based on the actual displacement and velocity of the vibrator compared to the desired displacement and velocity. ε is the sliding surface coefficient, a positive number whose value has a significant impact on the dynamic performance of the system. A larger ε value will cause the system state to approach the sliding surface more quickly, but may result in an overly drastic system response. A smaller ε value will slow the system state's approach to the sliding surface, but the system response will be smoother.

[0133] K1 is the state feedback gain matrix, which is pre-designed based on the system's dynamic model and control requirements. Its function is to adjust the control input based on the current state of the system, so that the system state changes in the desired direction.

[0134] sgn(s(t)) is the sign function, which is defined as follows according to the positive or negative value of s(t):

[0135]

[0136] The function of the switching term is to ensure that the system state can reach the sliding surface within a limited time and keep moving on the sliding surface. When s(t)<0, the switching term outputs K sw ; When s(t)>0, the switching item outputs -K sw Through this switching mechanism, the system state is continuously adjusted to approach and remain on the sliding surface.

[0137] 2. Control signal generation and output

[0138] According to the calculated control law F(t) = K sw The sliding mode controller converts the signal sign(s(t))+K1x(t) into a suitable control signal. The control signal is a voltage signal, which is amplified by the power amplifier to a sufficient power to drive the motor to change the vibration amplitude and frequency of the vibrator.

[0139] S5: Stability analysis:

[0140] 1. In order to ensure the stability of the sliding mode controller, Lyapunov stability theory is used for analysis. The Lyapunov function V(s) is selected as:

[0141] The function is non-negative, and when s(t) = 0, V(s) = 0, by analyzing the time derivative of V(s) The sign of can be used to determine the stability of the system.

[0142] 2. Take the time derivative of the Lyapunov function V(s) and get Let the sliding surface F(t) = K sw ·Substitute sign(s(t))+K1x(t) The expression of , combined with the state equation of the system, can be deduced to obtain The specific expression of Through analysis The stability of the system is determined by the sign of . The sliding surface coefficient ε is optimized through online learning of the neural network to determine whether it can be satisfy s(t)≠0, if the sliding surface coefficient and switching gain optimized by the neural network can make s(t)≠0, satisfying the Lyapunov function V(s) is monotonically decreasing. According to Lyapunov stability theory, the weighing sensor system is stable, that is, the system state will reach the sliding surface within a finite time and maintain stable motion on the sliding surface.

[0143] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A weighing adaptive vibration control method, characterized in that: The steps include: (1) Collect data from the vibrator and weighing sensor and calculate the sliding surface value s(t); (2) Construct a neural network online learning model. First, the sliding mode surface value, error and its derivative are input into the forward propagation output sliding mode controller required switching gain K sw and sliding surface coefficient ε; then calculate the error function J according to the output error of the weighing sensor; finally, the error signal δ of the hidden layer is obtained by back propagation based on the error function J h , and according to the error signal δ of the hidden layer h , using an optimization algorithm to update the weights and biases of the neural network; (3) Determine the sliding mode control rate based on the system's dynamic model and control objectives.

2. The weighing adaptive vibration control method according to claim 1, characterized in that: The data of the vibrator includes the displacement x(t) and velocity of the vibrator The data of the weighing sensor includes an output signal of the weighing sensor.

3. The weighing adaptive vibration control method according to claim 1, characterized in that: The sliding surface value s(t) is The error e(t) is the difference between the actual output of the weighing sensor and the actual material weight; the error derivative The first-order difference method is used, that is, Where Δt is the sampling time interval, t is the sampling time, and ε is the sliding surface coefficient.

4. The weighing adaptive vibration control method according to claim 3, characterized in that: The forward propagation specifically includes the following steps: Input layer processing: The input layer receives the sliding surface value s(t), error e(t) and its derivative As the input signal, it is then normalized and its value range is mapped to the interval [0,1], and finally the output vector of the input layer is obtained; Hidden layer processing: The hidden layer uses the ReLU activation function for nonlinear transformation and feature extraction: the output vector of the input layer is multiplied by the weight matrix of the hidden layer, and then the bias vector of the hidden layer is added to obtain the linear input z of the hidden layer. h ; Then, the linear input z of the hidden layer h Apply the ReLU activation function for nonlinear transformation to obtain the output vector a of the hidden layer h ;= Output layer processing: output vector a of the hidden layer h Multiplying it with the weight matrix of the output layer and adding the bias vector of the output layer, we get the linear input z of the output layer. o ;z o Corresponding to the switching gain K of the sliding mode controller sw and the sliding surface coefficient ε is z o1 is the linear input of the first neuron in the output layer, z o2 is the linear input of the second neuron in the output layer.

5. The weighing adaptive vibration control method according to claim 4, characterized in that: The error function J is: Among them, y i is the actual output of the weighing sensor, is the expected load cell output and N is the number of sampling points.

6. The weighing adaptive vibration control method according to claim 5, characterized in that: The back propagation specifically includes: Output layer error signal calculation: According to the error function J, the output layer linear input z o The error signal δ of the output layer is calculated by taking the partial derivative of o , δ o Expressed as Hidden layer error signal calculation: According to the error signal δ of the output layer o And the weight matrix W from the hidden layer to the output layer o , calculate the error signal δ of the hidden layer h is δ h =W o δ o ⊙ReLU′(z h ), where ⊙ represents element-wise multiplication, ReLU′(z h ) is the derivative of the ReLU activation function, z h is the linear input of the forward propagation hidden layer; Weight and bias update: Use the optimization algorithm to update the weights and biases of the neural network. The update formulas for weights and biases are as follows: Weight update: W new =W old -α▽ W J; Bias update: b new =b old -α▽ b J; Where α is the learning rate, ▽ W J and ▽ b J are the gradients of the error function J with respect to weights and biases, respectively, and W new is the weight of the new neural network, W old is the weight of the original neural network, b new is the new neural network bias; b old - is the original neural network bias.

7. The weighing adaptive vibration control method according to claim 3, characterized in that: The sliding mode control rate F(t) is calculated based on the sliding mode surface value s(t) as F(t) = K sw ·sign(s(t))+K1x(t); where K sw is the switching gain; K1 is the state feedback gain matrix; x(t) is the displacement of the oscillator, and sgn(s(t)) is the sign function, defined as:

8. A weighing adaptive vibration control system, characterized by: include: Vibrator, weighing sensor, sliding mode controller, neural network online learning module, data acquisition module and calculation module; The vibrator is used to assist in material transportation; the weighing sensor is used to measure the weight of the material and convert the weight information into an electrical signal output; The data acquisition module is responsible for collecting the status information of the vibrator, the output signal of the weighing sensor, the sliding mode surface value, the error and its derivative data in real time, and transmitting these data to the neural network online learning module and the sliding mode controller; The neural network online learning module receives the data from the data acquisition module and uses the neural network. The input layer receives a sliding mode surface value, error and its derivative as input signals, and the output layer outputs the parameters K required by the sliding mode controller. sw and ε; then, through the online learning algorithm, the weights and biases of the neural network are adjusted in real time according to the output error of the weighing sensor to optimize the parameters of the sliding mode controller; The sliding mode controller generates a control signal based on the control parameters output by the neural network online learning module to accurately control the vibration of the vibrator. The sliding mode controller designs a control law based on the sliding mode surface so that the system state reaches the sliding mode surface within a finite time and maintains movement on the sliding mode surface; The calculation module is used for sliding surface calculation, switching gain calculation, and data output to the neural network online learning module.

Images