Cantilever beam forming machine multi-cylinder synchronous hydraulic intelligent control method and system

Abstract

The present application relates to the technical field of construction intelligent control, and particularly relates to a multiple-oil-cylinder synchronous hydraulic intelligent control method and system for a cantilever beam forming machine; the present application calculates the physical residual characteristic of each jacking oil cylinder of the cantilever beam forming machine, calculates the cross-cylinder coupling correlation factor between different jacking oil cylinders, calculates the neighborhood maximum coupling strength according to the cross-cylinder coupling correlation factor between different jacking oil cylinders, calculates the coupling directionality factor of different oil cylinders, and further constructs a multi-dimensional time sequence state vector; that is, the cross-cylinder coupling correlation factor based on cross-correlation entropy is introduced, which significantly solves the synchronous control lag problem caused by the strong nonlinear coupling of the multiple oil cylinders of the cantilever beam forming machine, compared with the traditional linear correlation coefficient, the cross-correlation entropy can effectively filter out the high-frequency chatter and pulse noise of the hydraulic system, accurately quantize the implicit interference strength between the oil cylinders, so that the control system can distinguish between normal synchronous action and abnormal coupling interference, and avoid false adjustment caused by structural rigid force transmission.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for construction, and in particular to a method and system for synchronous hydraulic control of multiple oil cylinders in a cantilever beam-making machine. Background Technology

[0002] In existing synchronous control systems for cantilever beam-building machines, state vectors are typically constructed using direct data splicing. This involves directly combining raw readings from sensors such as displacement, pressure, and flow rate into a vector. However, this combined vector cannot determine whether these values ​​conform to the fundamental laws of hydraulic fluid dynamics. When oil temperature rises, causing changes in the oil's elastic modulus, or when a slight internal leak occurs in the cylinder, sensor data exhibits non-physical drift, leading to misleading subsequent models. Furthermore, in multi-cylinder synchronous systems for cantilever beam-building machines, the outriggers do not move independently but are coupled together through a highly rigid steel box girder structure. Existing technologies generally assume that the influence between cylinders is uniform and linear. However, in actual operating conditions, due to asymmetric eccentric loads generated during concrete pouring, wind loads, and pressure pulsations in the hydraulic lines, strong nonlinear and asymmetric coupling interference exists between the cylinders.

[0003] Meanwhile, in existing multi-cylinder synchronous control technology, even if a model is introduced to predict the future state of each lifting cylinder, the subsequent decision-making layer usually adopts a global unified control method, that is, using the same set of control parameters for all cylinders. However, when only cylinder No. 1 experiences internal leakage, causing lag, the global unified control will attempt to adjust the valves of all cylinders. This not only wastes the adjustment resources of normal cylinders, but may also cause new system oscillations due to the frequent actions of normal cylinders. Due to different manufacturing tolerances and wear, the physical characteristics of multiple cylinders are actually different, and the global unified strategy cannot perform targeted and precise compensation control. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a multi-cylinder synchronous hydraulic intelligent control method and system for cantilever beam-making machines, which solves the problems existing in the prior art.

[0005] This invention provides a multi-cylinder synchronous hydraulic intelligent control method for cantilever beam construction machines, the method comprising the following steps: S1: Collect status data of each lifting cylinder of the cantilever beam-making machine; S2: Perform data preprocessing on the state data; S3: Construct a multi-dimensional temporal state vector based on the normalized state data; Specifically, S3 consists of: S3.1: Calculating the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine; S3.2: Calculating the cross-cylinder coupling correlation factor between different lifting cylinders; S3.3: Calculating the maximum coupling strength in the neighborhood based on the cross-cylinder coupling correlation factor between different lifting cylinders; S3.4: Calculating the coupling directionality factor of different cylinders; S3.5: Constructing a multi-dimensional temporal state vector. S4: Input the multidimensional temporal state vector into the long short-term memory network model to predict the state of each lifting cylinder within a preset number of control cycles in the future; S5: Perform synchronous hydraulic intelligent control of multiple cylinders of the cantilever beam-making machine based on the state of each lifting cylinder within a preset number of control cycles in the future; S5 specifically involves: within the control cycle, calculating the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine based on the predicted state of each lifting cylinder within a preset number of control cycles in the future, classifying each lifting cylinder according to the physical residual characteristics, and formulating different control strategies according to the roles of each lifting cylinder.

[0006] Preferably, in step S3.1, the corrected theoretical displacement of the i-th lifting cylinder is calculated, and the expression is: ; In the formula, Let be the corrected theoretical displacement of the i-th lifting cylinder at time t. This refers to the total flow rate at the main oil outlet of the pumping station. The internal leakage coefficient of the hydraulic cylinder. Let be the cylinder pressure of the i-th lifting cylinder at time t. Let be the effective area of ​​the piston of the i-th lifting cylinder. Let be the gravitational load borne by the i-th lifting cylinder. This represents the equivalent stiffness of the local beam. Physical residual characteristics are constructed based on the corrected theoretical displacement of the lifting cylinder and the measured displacement of the displacement sensing unit. The expression is: ; In the formula, Let be the physical residual of the i-th lifting cylinder at time t. Let be the corrected theoretical displacement of the i-th lifting cylinder at time t. Let t be the displacement of the i-th lifting cylinder measured by the displacement sensing unit at time t.

[0007] Preferably, in S3.2, calculating the cross-cylinder coupling correlation factor between different lifting cylinders specifically involves: The dynamic feature vectors of the two cylinders within the sliding time window are extracted, representing the flow rate change rate of the i-th cylinder and the displacement change rate of the j-th cylinder, respectively. The generalized cross-correlation entropy of the two cylinders is then calculated as the cross-cylinder coupling correlation factor, expressed as: ; In the formula, Let the generalized cross-correlation entropy between cylinder i and cylinder j be used as the cross-cylinder coupling correlation factor between cylinder i and cylinder k, and N be the calculation window length. Let σ be the Gaussian kernel function, and σ be the kernel bandwidth parameter. Let be the rate of change of flow rate of cylinder i at time k. Let be the rate of change of flow rate of cylinder j at time k, where k is the summation index.

[0008] Preferably, N = any integer value in the range of 5-10 sampling periods, and σ = 0.5.

[0009] Preferably, in step S3.4, a coupling directionality factor is defined to distinguish between active interference and passive response between hydraulic cylinders. ; The expression is: ; In the formula, This is the coupling direction identifier between the i-th and j-th cylinders at time t, with a value of +1 or... 1, Let be the pressure change rate of the i-th cylinder. Let J be the pressure change rate of the j-th cylinder. It is a symbolic function.

[0010] Preferably, in step S3.5, a multidimensional temporal state vector is constructed based on the normalized state data, physical residual characteristics, maximum coupling strength in the neighborhood, and coupling directionality factor, expressed as: ; In the formula, Let be the displacement of the i-th lifting cylinder. Let be the pressure of the i-th lifting cylinder. Let be the total flow rate of the pump station's total outlet for the i-th lifting cylinder. Let be the physical residual characteristic of the i-th lifting cylinder at time t. The maximum coupling strength in the neighborhood of adjacent cylinders. This is the coupling direction identifier for the i-th and j-th cylinders at time t.

[0011] Preferably, in step S4, the long short-term memory network model includes an input layer, a physically guided attention mechanism, a stacked LSTM layer, a fully connected regression layer, and a task prediction head.

[0012] Preferably, the input layer receives the multidimensional temporal state vector; Following the input layer, a physically guided attention layer is introduced, expressed as: ; In the formula, Let be the attention weight for the i-th cylinder feature. and All are learnable parameters; Following the physically guided attention layer, a stacked LSTM layer is introduced. The stacked LSTM layer adopts a two-layer LSTM stacked structure. The first LSTM layer is a feature extraction layer, which includes 64 LSTM units and is responsible for extracting the basic dynamic features in the multi-dimensional temporal state vector. The second layer is a coupled inference layer, which includes 32 LSTM units and is responsible for inferring higher-order features. After the stacked LSTM layers, a fully connected regression layer is introduced. The hidden state is input into the fully connected regression layer, and the predicted sequence is output. Following the fully connected regression layer, a task prediction head is also included, which is a future physical residual feature prediction head.

[0013] Preferably, the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine are calculated based on the predicted state of each lifting cylinder within a preset number of control cycles in the future. The role of each lifting cylinder is then assigned based on these physical residual characteristics, specifically by calculating the navigation priority score for each lifting cylinder using the following formula: ; In the formula, The navigation priority score is given to the i-th lifting cylinder. Let represent the predicted physical residual characteristics of the i-th lifting cylinder of the cantilever beam-making machine at time t+1. This represents the average residual of all cylinders at the current moment. This is a local minimum value, used to prevent the denominator from being 0. The time interval between the last time this lifting cylinder was selected as the pilot cylinder. The time decay constant; The lifting cylinder with the highest priority score is selected as the lead cylinder, and the rest are follow cylinders.

[0014] Preferably, The value is 5s.

[0015] Preferably, different control strategies are formulated according to the roles of each lifting cylinder as follows: the PPO network architecture of the pilot cylinder includes a policy network and a value network, using two independent multilayer perceptron architectures. The input of the policy network is the multidimensional temporal state vector of the pilot cylinder, including an input layer, a first hidden layer, a second hidden layer, and an output layer. The output of the output layer is the proportional valve control voltage increment of the pilot cylinder, and the activation function is Tanh. The network architecture of the value network is the same as that of the policy network. The objective function of the PPO algorithm for the pilot cylinder is: ; In the formula, θ represents the parameters of the new policy network that is currently being updated and optimized. The importance sampling ratio, This is the estimated value of the dominance function. For the expectation operator, This is a pruning function used to forcibly limit the importance sampling ratio to [1]. ,1+ Within the interval, The cropping area; The expression for the clipping range of the PPO algorithm is: ; In the formula, The current clipping factor for the navigator cylinder. Here, α is the base value for the clipping factor, and α is the gain factor. The physical residual characteristics of the pilot cylinder at time t are predicted. This is a mapping function used to map physical residual characteristics to the (0.1) interval, preventing the hydraulic cylinder from... Excessive size leads to strategy collapse; In each control cycle t, the pilot cylinder adjusts its position according to the current state s. t Develop a strategy to output action a t That is, the proportional valve voltage increment, and observe the reward r. t and the next state s t+1 ;Will The data is stored in the high-priority buffer dedicated to the pilot cylinder, and then the policy gradient is calculated using the objective function described above. Then, the Adam optimizer is updated, and KL divergence constraints are used to finally output the final proportional valve control voltage increment, thereby realizing the pilot cylinder control.

[0016] Preferably, Set it to 0.2, and α to 0.15.

[0017] Preferably, for the following cylinder, the control strategy is as follows: the control parameters are updated once every K cycles, and the pruning coefficient of the objective function of the PPO algorithm is set to a fixed small value of 0.05.

[0018] Preferably, in step S1, a multi-source heterogeneous sensing unit is arranged on each lifting cylinder support point of the cantilever beam-making machine. The multi-source heterogeneous sensing unit includes a displacement sensing unit, a pressure sensing unit, and a flow sensing unit. The displacement sensing unit specifically comprises: a magnetostrictive displacement sensor installed on the piston rod side of each lifting cylinder, wherein the resolution of the magnetostrictive displacement sensor is not less than 0.01 mm; The pressure sensing unit specifically comprises: high-frequency dynamic pressure sensors installed in the rodless chamber and rod chamber of each lifting cylinder to measure the pressure data of the rodless chamber and the pressure data of the rod chamber, which are used to characterize load changes and hydraulic stiffness. The flow sensing unit specifically refers to: installing turbine flow meters or ultrasonic flow meters at the main oil outlet and key branch oil lines of the pump station to collect flow data in real time.

[0019] According to another aspect of the present invention, a multi-cylinder synchronous hydraulic intelligent control system for a cantilever beam-building machine is provided. The system employs the aforementioned multi-cylinder synchronous hydraulic intelligent control method for a cantilever beam-building machine, and the system includes: The data acquisition module is used to collect the status data of each lifting cylinder of the cantilever beam-making machine; The data preprocessing module is used to perform data preprocessing operations on the state data; The feature construction module is used to construct multi-dimensional temporal state vectors based on normalized state data. The state prediction module is used to input the multi-dimensional temporal state vector into the long short-term memory network model to predict the state of each lifting cylinder within a preset number of control cycles in the future. The control module is used to perform synchronous hydraulic intelligent control of multiple cylinders of the cantilever beam-making machine based on the status of each lifting cylinder within a preset number of control cycles in the future.

[0020] The embodiments of the present invention have the following technical effects: This invention significantly solves the synchronization control lag problem caused by strong nonlinear coupling of multiple cylinders in cantilever bridge construction machines by introducing a cross-cylinder coupling correlation factor based on cross-correlation entropy. Compared with the traditional linear correlation coefficient, cross-correlation entropy can effectively filter out high-frequency chatter and pulse noise in the hydraulic system, accurately quantify the intensity of implicit interference between cylinders, and enable the control system to distinguish between normal synchronous actions and abnormal coupling interference, avoiding misadjustment caused by rigid force transmission of the structure. Secondly, by integrating physical residual characteristics and cross-cylinder coupling correlation factors, a high-dimensional state space driven by data and physics is constructed, enabling it to identify the interference propagation path and compensate in advance at the initial stage of load change. Furthermore, the directional factor clarifies the primary and secondary relationships of load transmission, effectively suppressing overshoot and oscillation. Finally, this scheme enhances the system's robustness to individual cylinder differences and oil temperature changes. Even when some cylinders experience early internal leakage, the physical residual can still maintain high-precision synchronization of the system, greatly improving the safety of cantilever construction of long-span bridges and the quality of the completed bridge's alignment.

[0021] This invention significantly improves the dynamic response capability and training stability of the PPO algorithm under extreme operating conditions of cantilever beam-building machines by introducing a shear coefficient calculation scheme based on physical residuals. When a serious internal leakage or sudden load change in a certain cylinder is predicted, leading to an increase in physical residuals, the algorithm automatically widens the shear range, allowing the strategy network to update parameters with larger step sizes. This breaks through the conservative limitation of traditional fixed shear coefficients, achieving rapid and powerful compensation for hydraulic valve openings and effectively curbing the divergence of synchronization errors. Conversely, under steady-state or small-disturbance operating conditions, the shear range automatically narrows, strictly limiting the magnitude of strategy updates and avoiding hydraulic shocks and system oscillations caused by over-adjustment. This update mechanism ensures both rapid correction capability under fault conditions and stable operation under normal conditions, significantly improving the safety margin and robustness of multi-cylinder synchronous control. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines provided in this embodiment of the invention; Figure 2 This is a flowchart of constructing a multidimensional temporal state vector based on normalized state data, provided in an embodiment of the present invention. Figure 3 This is a diagram illustrating the effect of dynamically defining the shear range based on physical residuals, as provided in an embodiment of the present invention.

[0024] Figure 4 This is a comparison chart of the reward convergence curves of the traditional fixed shearing method and the method of the present invention during the PPO training process provided in the embodiments of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0026] Example 1, Figure 1 A flowchart of a multi-cylinder synchronous hydraulic intelligent control method for a cantilever beam-building machine is shown, such as... Figure 1As shown, the multi-cylinder synchronous hydraulic intelligent control method for cantilever beam construction machines includes the following steps: S1: Collect status data of each lifting cylinder of the cantilever beam-making machine; Multi-source heterogeneous sensing units are installed on the support points of each lifting cylinder of the cantilever beam-making machine. The multi-source heterogeneous sensing units include displacement sensing units, pressure sensing units, and flow sensing units.

[0027] The displacement sensing unit specifically comprises a magnetostrictive displacement sensor installed on the piston rod side of each lifting cylinder. The resolution of the magnetostrictive displacement sensor is not less than 0.01 mm, which is used to measure the real-time stroke data of each cylinder and to calculate the synchronization error and predict the trend.

[0028] The pressure sensing unit specifically comprises: high-frequency dynamic pressure sensors installed in the rodless and rod-side chambers of each lifting cylinder to measure the pressure data of the rodless and rod-side chambers, which are used to characterize load changes and hydraulic stiffness.

[0029] The flow sensing unit specifically refers to: installing turbine flow meters or ultrasonic flow meters at the main oil outlet and key branch oil lines of the pump station to collect flow data in real time.

[0030] S2: Perform data preprocessing on the state data; To ensure that the data input to the model has a high signal-to-noise ratio, high consistency, and physical interpretability, this step performs signal conditioning and preprocessing procedures before the sensor signals enter the model. Specifically, the preprocessing operations include: digital filtering and smoothing, outlier detection and fault tolerance repair, and normalization.

[0031] Specifically, the digital filtering and smoothing process involves: applying an adjustable-window-length moving average filter to displacement and flow data; establishing a first-order Kalman filter model for pressure data, treating the hydraulic system as a linear stochastic process, dynamically estimating the true pressure value, and filtering out instantaneous spikes caused by valve core vibration. The outlier detection and fault tolerance repair process involves: using the sliding window Z-score method to identify outliers; for single-sampling-point outliers, linear interpolation before and after the point in time is used for repair; for continuous multi-point outliers... The normalization and feature scaling process involves: applying the max-min normalization method to all variables, uniformly mapping them to the [0,1] interval.

[0032] S3: Construct a multi-dimensional temporal state vector based on the normalized state data; In existing synchronous control systems for cantilever beam-building machines, state vectors are typically constructed using direct data splicing, where raw readings from sensors such as displacement, pressure, and flow are directly combined into vectors. However, these vectors cannot accurately reflect whether the values ​​conform to fundamental laws of hydraulic fluid dynamics. When oil temperature rises, causing changes in the oil's elastic modulus, or when a cylinder experiences slight internal leakage, sensor data exhibits non-physical drift, leading to misleading subsequent models. Furthermore, in multi-cylinder synchronous systems of cantilever beam-building machines, the outriggers do not move independently but are coupled together through a highly rigid steel box girder structure. Existing technologies generally assume that the influence between cylinders is uniform and linear. However, in actual operating conditions, due to asymmetric eccentric loads generated during concrete pouring, wind loads, and pressure pulsations in the hydraulic lines, strong nonlinear and asymmetric coupling interference exists between cylinders. Therefore, this paper proposes introducing hydraulic residual characteristics based on a physical model and cross-cylinder coupling correlation factors when constructing multi-dimensional time-series state vectors to encompass the physical laws governing the beam-building machine's operation.

[0033] Specifically, such as Figure 2 As shown, S3 specifically includes: S3.1: Calculate the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine; Specifically: Based on fluid mechanics and Hooke's law, calculate the corrected theoretical displacement of the i-th lifting cylinder, expressed as: ; In the formula, Let be the corrected theoretical displacement of the i-th lifting cylinder at time t. This refers to the total flow rate at the main oil outlet of the pumping station. The internal leakage coefficient of the hydraulic cylinder. Let be the cylinder pressure of the i-th lifting cylinder at time t. Let be the effective area of ​​the piston of the i-th lifting cylinder. Let be the gravitational load borne by the i-th lifting cylinder. The equivalent stiffness of the local beam can be calculated by converting data from fiber optic grating (FBG) strain sensors placed on the beam surface.

[0034] Physical residual characteristics are constructed based on the corrected theoretical displacement of the lifting cylinder and the measured displacement of the displacement sensing unit. The expression is: ; In the formula, Let be the physical residual of the i-th lifting cylinder at time t. Let be the corrected theoretical displacement of the i-th lifting cylinder at time t. Let t be the displacement of the i-th lifting cylinder measured by the displacement sensing unit at time t.

[0035] S3.2: Calculate the cross-cylinder coupling correlation factor between different lifting cylinders; In the multi-cylinder synchronization system of a cantilever beam-building machine, the outriggers do not move independently but are coupled together through a rigid steel box girder structure. Existing technologies typically assume that the influence between cylinders is uniform and linear. However, in actual working conditions, due to asymmetric eccentric loads generated during concrete pouring, wind loads, and pressure pulsations in the hydraulic lines, strong nonlinear and asymmetric coupling interference exists between the cylinders. For example, when cylinder 1 decelerates due to a sudden increase in load, the massive box girder structure instantly transmits this resistance to cylinders 2 and 4, causing irregular fluctuations in their flow rate and pressure. Traditional control algorithms do not consider the mutual influence between different cylinders, resulting in synchronization errors failing to converge quickly after disturbances occur. This embodiment introduces a cross-cylinder coupling correlation factor based on cross-correlation entropy spectral density, using cross-correlation entropy to measure the nonlinear, non-Gaussian coupling strength between the signals of two cylinders. This factor can effectively filter out pulse noise in the hydraulic system and accurately capture implicit interference caused by force transmission through the mechanical structure.

[0036] Specifically, the calculation of the cross-cylinder coupling correlation factor between different lifting cylinders is as follows: The dynamic feature vectors of the two cylinders within the sliding time window are extracted, representing the flow rate change rate of the i-th cylinder and the displacement change rate of the j-th cylinder, respectively. The generalized cross-correlation entropy of the two cylinders is then calculated as the cross-cylinder coupling correlation factor, expressed as: ; In the formula, Let be the generalized cross-correlation entropy between the i-th and j-th cylinders. The range of this value is [0,1]. A value closer to 1 indicates stronger dynamic coupling between the two cylinders; a value closer to 0 indicates independent motion between the two cylinders. N is the calculation window length, preferably N = 5-10 sampling periods, used to smooth instantaneous noise. σ is a Gaussian kernel function, where σ is the kernel bandwidth parameter used to control the sensitivity to outliers; preferably, σ = 0.5. Let be the rate of change of flow rate of cylinder i at time k. Let be the rate of change of flow rate of cylinder j at time k, where k is the summation index.

[0037] S3.3: Calculate the maximum coupling strength in the neighborhood based on the cross-cylinder coupling correlation factor between different lifting cylinders; The maximum coupling strength in the neighborhood is calculated by selecting spatially adjacent cylinders to eliminate weak coupling interference from distant locations and reduce computational dimensionality. For example, in a four-cylinder system, the neighbors of cylinder 1 are typically cylinders 2 and 4. The cross-cylinder coupling correlation factors between cylinders 1 and 2, and between cylinders 1 and 4, are calculated separately. The largest cross-cylinder coupling correlation factor is then determined as the maximum coupling strength in the neighborhood. The expression for the maximum coupling strength in the neighborhood is as follows: .

[0038] S3.4: Calculate the coupling directionality factor for different hydraulic cylinders; To distinguish between active interference and passive response between hydraulic cylinders, a coupling directionality factor is defined. ; The expression is: ; In the formula, This is the coupling direction identifier between the i-th and j-th cylinders at time t, with a value of +1 or... 1, Let be the pressure change rate of the i-th cylinder. Let J be the pressure change rate of the j-th cylinder. If it is a sign function, then =+1 indicates that cylinder j accelerates when the pressure in cylinder i increases. This usually means that cylinder i is the load source and is compressing cylinder j. =-1 indicates that cylinder j decelerates when the pressure in cylinder i increases, which usually means a braking effect caused by a rigid structural connection.

[0039] S3.5: Construct a multidimensional temporal state vector; A multidimensional temporal state vector is constructed based on the normalized state data, physical residual characteristics, maximum coupling strength in the neighborhood, and coupling direction factor. The expression is: ; In the formula, Let be the displacement of the i-th lifting cylinder. Let be the pressure of the i-th lifting cylinder. The total flow rate at the pump station's main outlet for the i-th lifting cylinder is the data after preprocessing as described above. Let be the physical residual characteristic of the i-th lifting cylinder at time t. The maximum coupling strength in the neighborhood of adjacent cylinders. This is the coupling direction identifier for the i-th and j-th cylinders at time t.

[0040] This embodiment significantly solves the synchronization control lag problem caused by strong nonlinear coupling of multiple cylinders in a cantilever bridge construction machine by introducing a cross-cylinder coupling correlation factor based on cross-correlation entropy. Compared with the traditional linear correlation coefficient, cross-correlation entropy can effectively filter out high-frequency chatter and pulse noise in the hydraulic system, accurately quantify the intensity of implicit interference between cylinders, and enable the control system to distinguish between normal synchronous actions and abnormal coupling interference, avoiding erroneous adjustments caused by rigid force transmission. Secondly, by integrating physical residual characteristics and cross-cylinder coupling correlation factors, a high-dimensional state space driven by both data and physical processes is constructed, enabling it to identify interference propagation paths and compensate in advance during the initial stage of load mutation. Furthermore, the directional factor clarifies the primary and secondary relationships of load transmission, effectively suppressing overshoot and oscillation. Finally, this scheme enhances the system's robustness to individual cylinder differences and oil temperature changes. Even when some cylinders experience early internal leakage, the physical residual can still maintain high-precision synchronization of the system, significantly improving the safety and bridge alignment quality of long-span cantilever bridge construction.

[0041] S4: Input the multidimensional temporal state vector into the long short-term memory network model to predict the state of each lifting cylinder within a preset number of control cycles in the future; The hydraulic system of a cantilever beam-building machine exhibits strong nonlinearity, multivariable coupling, and significant hysteresis characteristics, making traditional prediction methods based on mathematical analytical models difficult to apply. This step employs a Long Short-Term Memory (LSTM) network to construct a state prediction model. Based on the historical multidimensional time-series state tensor constructed in the previous steps, it predicts the system state evolution within a preset number of control cycles.

[0042] To fully explore the physical laws contained in the state vector, this step designs a dual-channel feature fusion long short-term memory network model, including an input layer, a physically guided attention mechanism, a stacked LSTM layer, a fully connected regression layer, and a task prediction head.

[0043] The input layer receives the multidimensional temporal state vector.

[0044] Following the input layer, a physically guided attention layer is introduced. This layer does not blindly focus on all features, but instead assigns weights based on the magnitude of the physical residual features, as expressed in the following expression: ; In the formula, Let be the attention weight for the i-th cylinder feature. and All parameters are learnable. When the physical residual of a certain cylinder is abnormally large, this layer automatically increases its attention to this cylinder, forcing the long short-term memory network model to focus on learning the dynamic characteristics under fault or abnormal conditions.

[0045] Following the physically guided attention layer, a stacked LSTM layer is introduced. The stacked LSTM layer adopts a two-layer LSTM stacked structure. The first LSTM layer is a feature extraction layer, which includes 64 LSTM units and is responsible for extracting basic dynamic features such as velocity and acceleration from the multi-dimensional temporal state vector. The second layer is a coupled inference layer, which includes 32 LSTM units and is responsible for inferring higher-order features, especially the interaction logic between cylinders.

[0046] Following the stacked LSTM layers, a fully connected regression layer is introduced. The hidden states are input into the fully connected regression layer, and the predicted sequence is output. After the fully connected regression layer, a task prediction head is also included. The task prediction head designed in this model is a future physical residual feature prediction head, that is, the state of each lifting cylinder is its future physical residual.

[0047] S5: Synchronous hydraulic intelligent control of multiple cylinders of the cantilever beam-making machine based on the status of each lifting cylinder within a preset control cycle in the future. In existing multi-cylinder synchronous control technology, even if a model is introduced to predict the future state of each lifting cylinder, the subsequent decision-making layer usually adopts a global unified control method, that is, using the same set of control parameters for all cylinders. However, when only cylinder No. 1 experiences internal leakage, causing lag, the global unified control will attempt to adjust the valves of all cylinders. This not only wastes the adjustment resources of normal cylinders, but may also cause new system oscillations due to the frequent actions of normal cylinders. At the same time, due to different manufacturing tolerances and wear, the physical characteristics of multiple cylinders are actually different, and the global unified strategy cannot perform targeted and precise compensation control.

[0048] Based on this, this embodiment provides a control method based on physical residual classification, which divides the lifting cylinder into a pilot cylinder and a follower cylinder, and applies different control methods to different types of cylinders. Specifically, S5 is as follows: Within the control cycle, the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine are calculated based on the predicted state of each lifting cylinder within the preset control cycle in the future, and the roles of each lifting cylinder are divided according to the physical residual characteristics. Specifically, the navigation priority score for each lifting cylinder is calculated using the following formula: ; In the formula, The navigation priority score is given to the i-th lifting cylinder. Let represent the predicted physical residual characteristics of the i-th lifting cylinder of the cantilever beam-making machine at time t+1. This represents the average residual of all cylinders at the current moment. This is a local minimum value, used to prevent the denominator from being 0. The time interval between the last time this lifting cylinder was selected as the pilot cylinder. The time decay constant is 5s, which is preferred in this embodiment to prevent a certain lifting cylinder from occupying the navigation position for a long time, thus causing the algorithm to become unbalanced.

[0049] The lifting cylinder with the highest priority score is selected as the lead cylinder, and the rest are follow cylinders.

[0050] Different control strategies are formulated based on the role of each lifting cylinder; When a lifting cylinder is identified as the pilot cylinder, it means that the physical residual of that lifting cylinder deviates significantly from the normal value, usually caused by internal leakage, jamming, or local extreme off-center loading. The main challenge here is to quickly eliminate the residual to prevent structural distortion. For the pilot cylinder, the control objective is to quickly eliminate the residual. The shearing coefficient used in the traditional PPO algorithm is too conservative when facing sudden faults, easily leading to error accumulation. Therefore, this step designs an update strategy to reduce error accumulation for the pilot cylinder.

[0051] Specifically, the PPO network architecture of the pilot cylinder includes a policy network and a value network, employing two independent multilayer perceptron architectures to avoid value assessment interfering with action decision-making. The policy network takes as input the multidimensional temporal state vector of the pilot cylinder and includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The first hidden layer comprises 256 neurons with ReLU activation, and the second hidden layer comprises 128 neurons with ReLU activation. The output of the output layer is the proportional valve control voltage increment of the pilot cylinder, with Tanh activation. The value network has the same architecture as the policy network. Notably, the value network takes as input the proportional valve control voltage increment of the pilot cylinder and outputs the estimated reward for the current state of the pilot cylinder.

[0052] The core of the PPO algorithm for the pilot cylinder lies in limiting the magnitude of policy updates to ensure training stability. Its objective function is: ; In the formula, θ represents the parameters of the new policy network that is currently being updated and optimized. The importance sampling ratio, This is the estimated value of the dominance function. For the expectation operator, This is a pruning function used to forcibly limit the importance sampling ratio to [1]. ,1+ Within the interval, This refers to the cropping area.

[0053] Compared to the existing PPO algorithm, this embodiment improves the clipping range of the PPO algorithm, as expressed in the following expression: ; In the formula, The current clipping factor for the navigator cylinder. The basic value for the clipping factor is set to 0.2 in this embodiment, and α is the gain factor, preferably 0.15. The physical residual characteristics of the pilot cylinder at time t are predicted. This is a mapping function used to map physical residual characteristics to the (0,1) interval, preventing hydraulic cylinders from... Too large an area can cause the strategy to collapse.

[0054] In each control cycle t, the pilot cylinder adjusts its position according to the current state s. t Develop a strategy to output action a t That is, the proportional valve voltage increment, and observe the reward r. t and the next state s t+1 ;Will The data is stored in the high-priority buffer dedicated to the pilot cylinder, and then the policy gradient is calculated using the objective function described above. Then, the Adam optimizer is updated, and KL divergence constraints are used to finally output the final proportional valve control voltage increment, thereby realizing the pilot cylinder control.

[0055] For the following cylinder, the control objective is to maintain the status quo and make only minor adjustments to keep it synchronized. The control strategy is to update it once every K cycles. The pruning coefficient of the objective function of the PPO algorithm is set to a fixed small value, which is 0.05 in this embodiment, to limit the range of strategy changes and prevent oscillations caused by over-adjustment.

[0056] Appendix Figure 3 The diagram shows the effect of dynamically defining the shear range based on physical residuals, such as... Figure 3 As shown, the horizontal axis represents the normalized physical residual |r|, and the vertical axis represents the shear range ε. The blue solid line represents the shear range ε ​​of this invention, exhibiting a smooth S-shaped growth characteristic: when the physical residual is small, the shear range ε ​​automatically narrows; when the physical residual increases significantly, the shear range ε ​​automatically widens. The gray dashed line represents the traditional fixed shear range. Through the above content, it is intuitively demonstrated that this embodiment can dynamically adjust the shear range according to the system state, taking into account both the speed of fault correction and the stability of normal operation.

[0057] Appendix Figure 4The diagram shows a comparison of the reward convergence curves of the traditional fixed-shear method and the method of this invention during PPO training. The horizontal axis represents the number of training rounds, and the vertical axis represents the average round reward. The red dashed line represents the traditional fixed-shear method, and the blue solid line represents the method of this invention. It can be seen that compared to the traditional method, the reward of this invention rises more steeply in the early stages of training, and the reward curve is smoother and more stable at a higher level in the mid-to-late stages. In contrast, the traditional method exhibits greater fluctuations and converges to the high-reward region more slowly, demonstrating the superiority of the method of this invention.

[0058] This embodiment significantly improves the dynamic response capability and training stability of the PPO algorithm under extreme conditions of cantilever beam-building machines by introducing a shear coefficient calculation scheme based on physical residuals. When a serious internal leakage or sudden load change in a certain cylinder is predicted, leading to an increase in physical residuals, the algorithm automatically widens the shear range, allowing the strategy network to update parameters with larger step sizes. This breaks through the conservative limitation of traditional fixed shear coefficients, achieving rapid and powerful compensation for hydraulic valve openings and effectively curbing the divergence of synchronization errors. Conversely, under steady-state or small-disturbance conditions, the shear range automatically narrows, strictly limiting the magnitude of strategy updates and avoiding hydraulic shocks and system oscillations caused by over-adjustment. This update mechanism ensures both rapid correction capability under fault conditions and stable operation under normal conditions, significantly improving the safety margin and robustness of multi-cylinder synchronous control.

[0059] Example 2: The present invention also provides a multi-cylinder synchronous hydraulic intelligent control system for a cantilever beam-building machine. The system adopts the multi-cylinder synchronous hydraulic intelligent control method for a cantilever beam-building machine described in Example 1. The system includes: The data acquisition module is used to collect the status data of each lifting cylinder of the cantilever beam-making machine; The data preprocessing module is used to perform data preprocessing operations on the state data; The feature construction module is used to construct multi-dimensional temporal state vectors based on normalized state data. The state prediction module is used to input the multi-dimensional temporal state vector into the long short-term memory network model to predict the state of each lifting cylinder within a preset number of control cycles in the future. The control module is used to perform synchronous hydraulic intelligent control of multiple cylinders of the cantilever beam-making machine based on the status of each lifting cylinder within a preset number of control cycles in the future.

[0060] Example 3: The present invention also provides an electronic device, including one or more processors and a memory.

[0061] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0062] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-building machines described above in any embodiment of this application, and / or other desired functions. Various contents such as initial extrinsic parameters and thresholds may also be stored in the computer-readable storage medium.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent control of multi-cylinder synchronization hydraulic system of a cantilever beam forming machine, characterized in that, The method includes the following steps: S1: Collect status data of each lifting cylinder of the cantilever beam-making machine; S2: Perform data preprocessing on the state data; S3: Construct a multi-dimensional temporal state vector based on the normalized state data; Specifically, S3 consists of: S3.1: Calculating the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine; S3.2: Calculating the cross-cylinder coupling correlation factor between different lifting cylinders. Specifically, in S3.2, calculating the cross-cylinder coupling correlation factor between different lifting cylinders involves: The dynamic feature vectors of the two cylinders within the sliding time window are extracted, representing the flow rate change rate of the i-th cylinder and the displacement change rate of the j-th cylinder, respectively. The generalized cross-correlation entropy of the two cylinders is then calculated as the cross-cylinder coupling correlation factor, expressed as: ； In the formula, Let the generalized cross-correlation entropy between cylinder i and cylinder j be used as the cross-cylinder coupling correlation factor between cylinder i and cylinder k, and N be the calculation window length. Let σ be the Gaussian kernel function, and σ be the kernel bandwidth parameter. Let be the rate of change of flow rate of cylinder i at time k. Let be the rate of change of flow rate of cylinder j at time k, where k is the summation index; S3.3: Calculate the maximum coupling strength in the neighborhood based on the cross-cylinder coupling correlation factor between different lifting cylinders. In S3.3, the largest cross-cylinder coupling correlation factor is determined as the maximum coupling strength in the neighborhood. The expression for the maximum coupling strength in the neighborhood is as follows: S3.4: Calculate the coupling directionality factor for different cylinders. In S3.4, to distinguish between active interference and passive response between cylinders, a coupling directionality factor is defined. ; The expression is: ; In the formula, This is the coupling direction identifier between the i-th and j-th cylinders at time t, with a value of +1 or... 1, Let be the pressure change rate of the i-th cylinder. Let J be the pressure change rate of the j-th cylinder. S3.5: Construct a multidimensional temporal state vector. In S3.5, a multidimensional temporal state vector is constructed based on the normalized state data, physical residual characteristics, maximum coupling strength in the neighborhood, and coupling directionality factor. The expression is: ； In the formula, Let be the displacement of the i-th lifting cylinder. Let be the pressure of the i-th lifting cylinder. Let be the total flow rate of the pump station's total outlet for the i-th lifting cylinder. Let be the physical residual characteristic of the i-th lifting cylinder at time t. The maximum coupling strength in the neighborhood of adjacent cylinders. This indicates the coupling direction between the i-th and j-th cylinders at time t. S4: Input the multidimensional temporal state vector into the long short-term memory network model to predict the state of each lifting cylinder within a preset number of control cycles in the future; S5: Perform synchronous hydraulic intelligent control of multiple cylinders of the cantilever beam-making machine based on the state of each lifting cylinder within a preset number of control cycles in the future; S5 specifically involves: within the control cycle, calculating the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine based on the predicted state of each lifting cylinder within a preset number of control cycles in the future, classifying each lifting cylinder according to the physical residual characteristics, and formulating different control strategies according to the roles of each lifting cylinder.

2. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 1, characterized in that, In step S3.1, the corrected theoretical displacement of the i-th lifting cylinder is calculated, and the expression is: ； In the formula, Let be the corrected theoretical displacement of the i-th lifting cylinder at time t. This refers to the total flow rate at the main oil outlet of the pumping station. The internal leakage coefficient of the hydraulic cylinder. Let be the cylinder pressure of the i-th lifting cylinder at time t. Let be the effective area of ​​the piston of the i-th lifting cylinder. Let be the gravitational load borne by the i-th lifting cylinder. This represents the equivalent stiffness of the local beam. Physical residual characteristics are constructed based on the corrected theoretical displacement of the lifting cylinder and the measured displacement of the displacement sensing unit. The expression is: ; In the formula, Let be the physical residual of the i-th lifting cylinder at time t. Let be the corrected theoretical displacement of the i-th lifting cylinder at time t. Let t be the displacement of the i-th lifting cylinder measured by the displacement sensing unit at time t.

3. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 1, characterized in that, N is any integer value among 5-10 sampling periods, and σ=0.

5.

4. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 1, characterized in that, In S4, the Long Short-Term Memory network model includes an input layer, a physically guided attention mechanism, a stacked LSTM layer, a fully connected regression layer, and a task prediction head.

5. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 4, characterized in that, The input layer receives the multidimensional temporal state vector; Following the input layer, a physically guided attention layer is introduced, expressed as: ； In the formula, Let be the attention weight for the i-th cylinder feature. and All are learnable parameters; Following the physically guided attention layer, a stacked LSTM layer is introduced. The stacked LSTM layer adopts a two-layer LSTM stacked structure. The first LSTM layer is a feature extraction layer, which includes 64 LSTM units and is responsible for extracting the basic dynamic features in the multi-dimensional temporal state vector. The second layer is a coupled inference layer, which includes 32 LSTM units and is responsible for inferring higher-order features. After the stacked LSTM layers, a fully connected regression layer is introduced. The hidden state is input into the fully connected regression layer, and the predicted sequence is output. Following the fully connected regression layer, a task prediction head is also included, which is a future physical residual feature prediction head.

6. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 5, characterized in that, Based on the predicted state of each lifting cylinder within a preset control cycle, the physical residual characteristics of each lifting cylinder of the cantilever beam-making machine are calculated. Based on these physical residual characteristics, each lifting cylinder is assigned a role: specifically, the navigation priority score for each lifting cylinder is calculated using the following formula: ； In the formula, The navigation priority score is given to the i-th lifting cylinder. Let be the physical residual characteristics of the i-th lifting cylinder of the cantilever beam-making machine at time t+1. This represents the average residual of all cylinders at the current moment. This is a local minimum value, used to prevent the denominator from being 0. The time interval between the last time this lifting cylinder was selected as the pilot cylinder. The time decay constant; The lifting cylinder with the highest priority score is selected as the lead cylinder, and the rest are follow cylinders.

7. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 6, characterized in that, The value is 5s.

8. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 7, characterized in that, Based on the role of each lifting cylinder, different control strategies are formulated as follows: The PPO network architecture of the pilot cylinder includes a policy network and a value network, employing two independent multilayer perceptron architectures. The input of the policy network is the multidimensional temporal state vector of the pilot cylinder, including an input layer, a first hidden layer, a second hidden layer, and an output layer. The output of the output layer is the proportional valve control voltage increment of the pilot cylinder, with Tanh as the activation function. The network architecture of the value network is the same as that of the policy network. The objective function of the PPO algorithm for the pilot cylinder is: ； In the formula, θ represents the parameters of the new policy network that is currently being updated and optimized. The importance sampling ratio, This is the estimated value of the dominance function. For the expectation operator, This is a pruning function used to forcibly limit the importance sampling ratio to [1]. ,1+ Within the interval, The cropping area; The expression for the clipping range of the PPO algorithm is: ； In the formula, The current clipping factor for the navigator cylinder. Here, α is the base value for the clipping factor, and α is the gain factor. The physical residual characteristics of the pilot cylinder at time t are predicted. This is a mapping function used to map physical residual features to the (0,1) interval, preventing... Excessive size leads to strategy collapse; In each control cycle t, the pilot cylinder adjusts its position according to the current state s. t Develop a strategy to output action a t That is, the proportional valve voltage increment, and observe the reward r. t and the next state s t+1 ;Will The data is stored in the high-priority buffer dedicated to the pilot cylinder, and then the policy gradient is calculated using the objective function described above. Then, the Adam optimizer is updated, and KL divergence constraints are used to finally output the final proportional valve control voltage increment, thereby realizing the pilot cylinder control.

9. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 8, characterized in that, Set it to 0.2, and α to 0.

15.

10. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 9, characterized in that, For the following cylinder, the control strategy is to update the control parameters once every K cycles, and the pruning coefficient of the objective function of the PPO algorithm is set to a fixed small value of 0.

05.

11. The multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines according to claim 1, characterized in that, In S1, multi-source heterogeneous sensing units are arranged on the support points of each lifting cylinder of the cantilever beam-making machine. The multi-source heterogeneous sensing units include displacement sensing units, pressure sensing units and flow sensing units. The displacement sensing unit specifically comprises: a magnetostrictive displacement sensor installed on the piston rod side of each lifting cylinder, wherein the resolution of the magnetostrictive displacement sensor is not less than 0.01 mm; The pressure sensing unit specifically comprises: high-frequency dynamic pressure sensors installed in the rodless chamber and rod chamber of each lifting cylinder to measure the pressure data of the rodless chamber and the pressure data of the rod chamber, which are used to characterize load changes and hydraulic stiffness. The flow sensing unit specifically refers to: installing turbine flow meters or ultrasonic flow meters at the main oil outlet and key branch oil lines of the pump station to collect flow data in real time.

12. A multi-cylinder synchronous hydraulic intelligent control system for a cantilever beam-making machine, characterized in that, The system employs the multi-cylinder synchronous hydraulic intelligent control method for cantilever beam-making machines as described in any one of claims 1-11. The system includes: The data acquisition module is used to collect the status data of each lifting cylinder of the cantilever beam-making machine; The data preprocessing module is used to perform data preprocessing operations on the state data; The feature construction module is used to construct multi-dimensional temporal state vectors based on normalized state data. The state prediction module is used to input the multi-dimensional temporal state vector into the long short-term memory network model to predict the state of each lifting cylinder within a preset number of control cycles in the future. The control module is used to perform synchronous hydraulic intelligent control of multiple cylinders of the cantilever beam-making machine based on the status of each lifting cylinder within a preset number of control cycles in the future.

Images