An adaptive data uploading scheduling method for burst transient power events

Abstract

The application discloses a kind of adaptive data sending scheduling methods for burst transient power event, it is related to industrial automation control and industrial internet communication cross technical field, the steps of this method include: obtaining end effector operating data and extracting features to generate rigid collision state quantity;Further, according to this, the survival period of perception data packet is evaluated, the priority is established to establish preemption channel in medium access control layer reconstruction;While calculating spatial stress distortion scale, the trust arbitration of edge prediction model is carried out;When judging model distortion, cut off compression link, and trigger the dimensionality reduction direct transmission of original state message;Finally, the active release and adaptive deployment of network bandwidth are executed in combination with the running damping loss ratio of equipment.This application realizes the low-delay high-reliable return of burst resistance perception data under complex working conditions and the balanced scheduling of network resources.

Description

Technical Field

[0001] This invention relates to the field of technology, specifically to an adaptive data transmission and scheduling method for sudden transient power events. Background Technology

[0002] In unstructured environments, heavy-duty electric equipment is highly susceptible to encountering unknown, rigid physical constraints at its end-effectors, which can trigger sudden transient power events in the system. These transient events generate massive amounts of multi-source sensor data instantaneously, rapidly exhausting the buffers of traditional network communication and causing severe data queuing and jitter latency. This sudden communication congestion can directly consume the phase of the bidirectional closed-loop control system, leading to system oscillations and even damage to the underlying physical execution architecture.

[0003] To address the aforementioned shortcomings, existing technologies typically introduce predictive control algorithms and data compression classification mechanisms at the edge. The main approach involves deploying intelligent predictive models at edge nodes to anticipate environmental resistance and mask network dead time. Simultaneously, relying on static quality of service queues and data compression techniques, massive transient features are reduced in dimensionality and prioritized for transmission, aiming to maintain the flow of core services within limited communication bandwidth.

[0004] However, existing solutions still face several serious challenges when dealing with complex transient events: First, edge prediction models heavily rely on historical homogeneous samples, making them prone to model illusions and distorted predictions when faced with highly random and unstructured environments. This can introduce erroneous force feedback interference into the control loop. Second, static quality of service queues cannot perceive the dynamic urgency caused by sudden collisions, and conventional scheduling mechanisms can still lead to fatal delays in critical sensing data when facing transient congestion. Finally, conventional compression mechanisms are prone to losing high-frequency distortion components when subjected to abnormal forces. Furthermore, if the system rigidly maintains high-frequency transmission of underlying parameters when it enters a stagnant state with high damping dissipation, it will severely squeeze the bandwidth quota of environmental sensing terminals such as visual monitoring, causing misallocation of network resources and scheduling imbalance.

[0005] Therefore, the present invention provides an adaptive data transmission scheduling method for sudden transient power events. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive data transmission and scheduling method for sudden transient power events, so as to solve the existing problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an adaptive data transmission and scheduling method for sudden transient power events, comprising the following steps: S1. Obtain real-time multi-source operating data of the end effector, extract transient pressure derivative and displacement attenuation scale from the real-time multi-source operating data, and generate rigid collision state quantities based on the transient pressure derivative and displacement attenuation scale. S2. Based on the rigid collision state quantity, perform a lifetime assessment on the sensing data packet to generate a dynamic survival time. Combine the dynamic survival time with the priority reconstruction of the medium access control layer to establish a preemption channel. Transmit the resistance sensing message back through the preemption channel. S3. Perform multi-axis force analysis and synthesis of spatial resistance vector on the real-time multi-source operation data, calculate the spatial distortion scale within a short time window based on the spatial resistance vector, and use the spatial distortion scale to perform confidence arbitration on the edge prediction model to generate arbitration instructions; S4. Determine the distortion state of the model output based on the arbitration instruction. When the distortion state is determined, cut off the compressed link and trigger the generation of the direct transmission sequence of the original message for direct transmission and uploading configuration. S5. Synchronously collect the work integral of the drive end and the kinetic energy increment of the execution end to calculate the running damping loss ratio, and generate an adaptive scheduling strategy by actively releasing and allocating network bandwidth based on the running damping loss ratio.

[0008] A further improvement of this invention is that the generation process of the rigid collision state quantity includes: S11. Collect the working cylinder pressure sequence and piston displacement sequence of the end effector through a multi-axis pressure sensor and a position encoder; S12. Perform time-series alignment and noise reduction smoothing on the pressure sequence and displacement sequence; S13. For the aligned sequence, extract the transient pressure change gradient and the position offset attenuation ratio per unit time and perform coupled cross-operation. S14. When the transient pressure change gradient is greater than the set step rise threshold and the position offset attenuation ratio is less than the set stagnation lower limit, output the rigid impact evaluation parameter as the rigid collision state quantity.

[0009] A further improvement of this invention is that the step of performing lifetime assessment on the sensed data packet based on the rigid collision state quantity to generate a dynamic survival time, and combining the dynamic survival time to implement priority reconstruction of the media access control layer to establish a preemptive channel, includes: receiving the rigid collision state quantity and simultaneously obtaining the current queue backlog status of the communication media access control layer; extracting the regular service priority identifier of the currently to-be-sent sensed data packet; substituting the rigid collision state quantity into the lifetime assessment mapping model to calculate the dynamic survival time of the sensed data packet; when the dynamic survival time is less than a preset delay-sensitive control line, triggering a forced suspension instruction at the underlying network protocol layer to postpone the transmission of non-critical services, and generating a dynamic priority control instruction to establish the preemptive channel for the resistance sensed message.

[0010] A further improvement of this invention lies in performing multi-axis force analysis and synthesis of a spatial resistance vector on the real-time multi-source operating data, calculating a spatial distortion scale within a short-term window based on the spatial resistance vector, and using the spatial distortion scale to perform confidence arbitration on the edge prediction model to generate an arbitration instruction. This includes: retrieving the original sequence of multi-dimensional force sensors within the short-term window, and the predicted force vector synchronously output by the edge prediction model; synthesizing the original sequence of multi-dimensional force sensors into a three-dimensional spatial resistance vector; evaluating the directional divergence of the three-dimensional spatial resistance vector within the time window, calculating a spatial force distortion index describing the degree of force unevenness as the spatial distortion scale; cross-checking the spatial force distortion index with the prediction confidence output by the edge prediction model; and when the spatial force distortion index is detected to be higher than the divergence distortion threshold and the prediction confidence is greater than the lower confidence threshold, determining that the model output has a risk of distortion, and generating a downgraded confidence level scalar as the arbitration instruction.

[0011] A further improvement of this invention is that the process of generating the direct transmission configuration from the reduced-dimensional direct transmission sequence of the triggering original state message includes: receiving the trust level scalar in the arbitration instruction and monitoring the compressed link working mode of the current edge prediction model; parsing the trust level scalar and calling the corresponding frequency domain analysis algorithm package; when the trust level scalar is less than a preset reliability degradation threshold, cutting off the black-box compressed link of the edge prediction model; using the frequency domain analysis algorithm package to convert the original sequence of the force sensor and extract waveform components to form the direct transmission configuration, and directly transmitting the extracted waveform components in the original state format.

[0012] A further improvement of this invention is that the operating damping loss ratio is obtained by acquiring the integral work of the drive motor of the power system and the kinetic energy increment data corresponding to the actual displacement of the end effector; data unification and timing alignment processing is performed on the integral work and the kinetic energy increment data; the difference between the integral work and the kinetic energy increment data is calculated to obtain the operating damping loss ratio of the current operation process; when the operating damping loss ratio is greater than the high dissipation threshold for three consecutive times, an environment adaptive bandwidth allocation strategy is executed to actively extend the effective life cycle of the tactile sensing data packets and allocate the released network bandwidth resources to the environmental sensing terminal device.

[0013] A further improvement of the present invention is that the environment adaptive bandwidth allocation strategy includes: constructing a mapping relationship based on the operating damping loss ratio and the set baseline survival time, and calculating the dynamic survival time of the tactile sensing data packet; The sliding window step size of the edge gateway buffer queue is adjusted according to the dynamic survival time, and downsampling is performed on the original tactile data stream. The downsampled tactile sensing data packets are intercepted at the media access control layer, their quality of service labels are rewritten, and the tactile sensing data packets are downgraded from the high-priority queue to the weighted fair queue. Monitor the token consumption rate of the tactile data channel, extract the decrease margin of the token consumption rate, and dynamically inject the redundant tokens corresponding to the decrease margin into the token bucket of the visual data channel to expand the available bandwidth quota of the environmental perception terminal device.

[0014] A further improvement of the present invention is that the priority reconstruction includes configuring a regular scheduling mode and a priority preemption mode; wherein, the priority preemption mode is set to be triggered only when a hard collision occurs as represented by the rigid collision state quantity, and the dynamic survival time is greater than or equal to the control delay red line set by the system.

[0015] A further improvement of this invention is that the survival assessment mapping model calculates the instability acceleration of the control system based on the second derivative of the transient pressure change gradient, and the length of the dynamic survival time is negatively correlated with the instability acceleration of the control system.

[0016] A further improvement of the present invention is that the step of using the frequency domain analysis algorithm package to convert the original sequence of the force sensor and extract waveform components includes performing a fast Fourier transform on the original force waveform and retaining only the basic waveform and the waveform component data of the first N high-energy orders to form the direct transmission configuration.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first generates a rigid collision state quantity by extracting the transient pressure derivative and displacement attenuation scale on the execution side, and then uses this state quantity to dynamically evaluate the lifetime of the sensing data packet in order to establish a preemptive channel by reconstructing the priority at the media access control layer. This solves the problem of severe delay-induced oscillation and mechanical damage in the bidirectional closed-loop control system caused by network transmission delay in unstructured operation environments such as steel smelting and scrap steel grabbing. It also realizes low-latency backhaul of sudden transient resistance sensing data, effectively ensuring the operational safety of heavy machinery.

[0018] 2. By extracting spatial distortion scaling through multi-axis force analysis, the edge prediction model is arbitrated with confidence level. When the model is distorted, the original message is directly transmitted in a dimension reduction manner. At the same time, the network bandwidth is actively allocated by combining the operating damping loss ratio. This solves the problems of existing predictive control algorithms being prone to model illusions and providing erroneous force feedback in complex environments with unknown consolidation cold steel or hard slag distribution at the bottom of the furnace, as well as the network congestion caused by the full transmission of transient data. It realizes the effective removal of model error interference under harsh operating conditions and the dynamic adaptive balanced scheduling of multi-terminal sensing links under limited communication resources. Attached Figure Description

[0019] Figure 1 This is a flowchart of an adaptive data transmission and scheduling method for sudden transient power events according to the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0021] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0022] Example 1 Figure 1 This embodiment presents a flowchart of an adaptive data transmission and scheduling method for sudden transient power events, with the following steps: This embodiment applies to the data transmission and scheduling process of heavy-duty electric material grabbing equipment in a steel plant setting. In unstructured environments such as steel smelting, the end effector of a heavy-duty material grabbing machine often faces randomly stacked scrap steel and highly unknown hardened steel slag. Existing two-way closed-loop control systems are extremely sensitive to dead time. When the grab bucket touches solidified cold steel or hard slag at the bottom of the furnace, the huge resistance transmitted back to the operating lever through the network has a high time delay. The operator perceives the delay and instinctively applies counterforce. This delayed counter-operation triggers positive feedback in the control system, causing severe time delays in the robotic arm that induce system oscillations, which can easily lead to serious damage such as hydraulic pipeline rupture. Conventional solutions usually introduce predictive control algorithms at the edge to provide early resistance feedback. However, such calculations are prone to model illusions when facing unknown scrap steel mechanical distributions, resulting in distorted force feedback.

[0023] To address the above situation, this method, through step S1, acquires real-time multi-source operating data of the end effector, extracts transient pressure derivatives and displacement attenuation scales from the real-time multi-source operating data, and generates rigid collision state quantities based on the transient pressure derivatives and displacement attenuation scales. Conventional digital twin or predictive control algorithms introduced at the edge side to overcome network latency issues are prone to applying incorrect force feedback to the operator when simulating such unknown scrap steel structures and hard slag properties in a very short time. This embodiment, by collecting dynamic changes in pressure and displacement at the execution side, transforms sudden force changes into collision state quantities, thereby providing a triggering prerequisite for addressing the dead-time sensitivity and latency-induced oscillation problems caused by network transmission latency in bidirectional closed-loop control systems.

[0024] In some possible embodiments, generating the rigid collision state quantity based on the transient pressure derivative and displacement decay scale includes: acquiring the working cylinder pressure sequence and piston displacement sequence of the end effector through a multi-axis pressure sensor and a position encoder; performing time-series alignment and noise reduction smoothing on the pressure sequence and displacement sequence; extracting the transient pressure change gradient and position offset decay ratio per unit time for coupling and cross-operation on the aligned sequence; and outputting the rigid impact evaluation parameter as the rigid collision state quantity when the transient pressure change gradient is greater than a set step rise threshold and the position offset decay ratio is less than a set stagnation lower limit.

[0025] Multi-axis pressure sensors and position encoders are used to collect the pressure sequence and piston displacement sequence of the working cylinder of the end effector. Multi-axis pressure sensors are deployed at the inlet and outlet ports of the hydraulic cylinders of the heavy-duty electric excavator's boom, and high-resolution position encoders are mounted on the piston rod's movement guide side. During operation, the sampling frequency is set to 1000 Hz to ensure the capture of sudden transient changes in force. The working cylinder pressure sequence acquired by the multi-axis pressure sensors includes pressure sampling values ​​at multiple consecutive time stamps; the piston displacement sequence acquired by the position encoder includes the mechanical extension / retraction synchronized with the pressure sampling time.

[0026] The pressure and displacement sequences are time-aligned and denoised for smoothing. A Lagrange interpolation algorithm is used to resample the pressure and displacement sequences from different sampling periods to a unified system clock reference, eliminating data misalignment. For the resampled sequences, a moving average filtering algorithm is introduced for smoothing to remove high-frequency interference introduced by mechanical vibration. The smoothing coefficient can be calculated by analyzing the sensor's fundamental noise spectrum and the Nyquist frequency of the hydraulic cylinder's force response. In this embodiment, the smoothing coefficient is preferably 0.6. This value can filter out high-frequency noise caused by strong vibrations while properly preserving the steep pressure step edges when the cylinder experiences sudden force, preventing the loss of critical abrupt change information. If sensor frame loss occurs during acquisition, the quadratic polynomial extrapolation value of the first three valid samples is used for local filling to maintain sequence continuity.

[0027] For the aligned sequence, the transient pressure change gradient per unit time and the position offset attenuation ratio are extracted and coupled cross-operated. The sequence data within the current short-time window is read, and the transient pressure change gradient parameter is set to... The position offset attenuation ratio parameter is set as follows: The formula for calculating the transient pressure change gradient and the position offset attenuation ratio is constructed as follows: ,in, This represents the rigidity impact assessment value output by the computation. This represents the shape conversion gain, which can be calculated by extracting the bulk modulus change curves of standard hydraulic oil at different temperatures and fitting its thermal expansion compensation ratio. In this embodiment, The optimal value is 1.25. Under the typical 45-degree Celsius calibration conditions of a steel plant, this value can accurately correct the volume transfer deviation caused by internal friction and heat generation, and effectively map the pressure change to the mechanical deformation space. As a hyperparameter, the ratio of momentary stagnation and positional offset when the grab impacts the giant solidified scrap steel, causing the robotic arm to momentarily stop moving. When the denominator is zero, this hyperparameter can handle the calculation crash caused by the denominator returning to zero, maintaining the continuity of the control system's command output.

[0028] When the transient pressure change gradient exceeds the set step rise threshold and the position offset attenuation ratio is less than the set stagnation lower limit, the rigid impact assessment quantity is output as the rigid collision state quantity. The state of each variable within the current calculation window is judged in real time. The value of the step rise threshold can be obtained by performing trial excavation in a standard homogeneous soil layer, extracting the natural pressure fluctuation rise rate generated by normal cutting resistance, and calculating it as a multiple of the upper limit of its confidence interval. In this embodiment, the set step rise threshold is preferably 45 MPa per second. The reason is that the pressure rise rate of conventional loose scrap steel or surface soft slag is usually between 15 and 25 MPa per second; raising it to 45 MPa per second can effectively isolate the interference fluctuations generated by conventional grabbing operations and accurately lock abnormal collisions. Meanwhile, the value of the stagnation lower limit can be comprehensively calculated and extracted by measuring the mechanical clearance of the end-drive mechanism and the inherent elastic rebound of the hydraulic circuit. In this embodiment, the set stagnation lower limit is preferably 0.05. The reason is that this value can effectively eliminate minor residual displacement caused by mechanical transmission backlash or hydraulic rebound, indicating that the travel rate of the execution end has been hindered and reduced by more than 95%. When the system determines that the above dual threshold conditions are met simultaneously, it immediately outputs the rigid impact assessment value corresponding to that time point and saves it as the rigid collision state value.

[0029] After obtaining the rigid collision state variables, it is necessary to address the network congestion problem in transmitting resistance messages back to the remote control console. Conventional communication scheduling mechanisms mostly employ a first-come, first-served queuing transmission model. When faced with a sudden collision, the massive amount of environmental surveillance video streams often clogs the underlying communication channels, causing a backlog in critical tactile feedback data packets. Remote control commands are delayed due to latency, easily triggering severe positive feedback oscillations in the closed-loop control system. Specifically, bidirectional closed-loop control systems under classical control theory are extremely sensitive to dead time. When a rigid collision occurs in the robotic arm bucket, if the resistance sensing data is uploaded in a first-in, first-out queue according to conventional network protocols, network transmission latency and jitter will directly consume the phase margin of the control system. This causes the operator to perceive resistance with lag and instinctively apply counterforce. This delayed counter-operation triggers severe positive feedback in the control system, causing severe latency in the heavy excavator's robotic arm, inducing system oscillations and easily leading to hydraulic line rupture. Traditional proportional-integral-derivative (PID) control strategies are completely ineffective in the face of such random latency. Therefore, this embodiment dynamically evaluates the expected lifespan of data packets, combines the rigid collision state quantity to perform lifespan assessment on the sensed data packets to generate a dynamic lifespan time, and combines the dynamic lifespan time to implement priority reconstruction of the media access control layer to establish a preemption channel, and transmits resistance sensing messages back through the preemption channel.

[0030] In some possible embodiments, the step of combining the rigid collision state quantity to perform lifetime assessment on the sensed data packet to generate a dynamic survival time includes: receiving the rigid collision state quantity and simultaneously acquiring the current queue backlog status of the communication medium access control layer; extracting the regular service priority identifier of the currently to-be-sent sensed data packet. The excavator vehicle-mounted communication gateway receives the rigid collision state quantity output in step S1 and simultaneously reads the transmission buffer of the underlying network chip to obtain the queue backlog status parameters. The data packet's lifetime must be reassessed. Simultaneously, the regular service priority identifier of the current resistance-aware message is parsed. Since regular background traffic labels cannot reflect the severity of sudden transient events, the packet's lifetime must be reassessed.

[0031] Substituting the rigid collision state variables into the survival time assessment mapping model, the dynamic survival time of the sensing data packet is calculated. The survival time assessment mapping model first extracts the second derivative of the transient pressure change gradient in the rigid collision state variables, and then calculates the instability acceleration of the control system. Subsequently, the dynamic lifespan of the sensed data packets was deduced. The calculation formula is expressed as: ,in, For hyperparameters, This represents the system inertial attenuation coefficient. The system inertial attenuation coefficient can be extracted by performing a frequency sweep excitation during the excavator's no-load start-up phase, collecting the inherent resonant response period of the robotic arm, and fitting it to the viscosity-temperature curve of the current hydraulic oil temperature. In this embodiment, The preferred value is 2.5 because this value can adapt to the viscosity changes of metallurgical flame-retardant hydraulic oil under different high temperature difference environments, and accurately characterize the inertial hysteresis effect of metallurgical heavy robotic arms during action switching. This is the congestion penalty factor. This congestion penalty factor can be dynamically calculated by statistically analyzing the product of the average packet loss rate and retransmission delay of the vehicular gateway over a short time window. In this embodiment, The preferred value of 0.02 is because this value closely reflects the basic communication conditions under the complex electromagnetic and dust interference in the steel plant during the shift, especially when the queue is congested. When the value increases, it can exponentially compress the expected lifespan of data packets, thereby accelerating the triggering of preemptive intervention.

[0032] When the dynamic survival time is less than a preset delay-sensitive control threshold, a forced suspension command is triggered at the underlying network protocol layer to temporarily suspend the transmission of non-critical services, and a dynamic priority control command is generated to establish the preemption channel for the resistance-aware message. The calculated dynamic survival time... Perform threshold comparison. Set the delay-sensitive control line to... This delay-sensitive control line can be calculated by real-time acquisition of the current excavator main pump output power and the robotic arm spatial extension posture matrix, deducing the mechanical resonant frequency period under the current posture, taking one-quarter of this period and deducting the inherent hardware processing overhead of the vehicle communication module. In this embodiment, A value of 35ms is preferable because when the information feedback delay of the control system approaches a quarter of the resonant period, it is most likely to induce positive feedback oscillations in the system. The control line obtained using this dynamic calculation ensures that the excavator accurately intercepts timeout risks under oscillating conditions such as extreme extension or heavy-load cutting. Once determined... Less than The system directly issues a forced suspension command to the media access control layer, freezing the transmission queues of high-throughput non-critical services such as video surveillance and ambient temperature monitoring. Simultaneously, it generates a dynamic priority control command, rewriting the scheduling label of resistance-aware packets to the highest preemption level and establishing a dedicated preemption channel. This breaks the fair competition mechanism of conventional network protocols, ensuring the ultra-fast return of rigid collision data.

[0033] While the aforementioned preemptive scheduling strategy effectively addresses the latency oscillation problem, in actual remote operations, to further overcome the inherent limitations of total network bandwidth in remote areas, edge computing nodes typically deploy AI prediction models to pre-simulate environmental resistance and compress data. However, conventional edge prediction models primarily utilize homogeneous environmental interaction data for pre-training. Once the working equipment enters a random, unstructured environment with a mixture of rock and soft soil, the model is highly susceptible to cognitive biases, outputting severely distorted and overly smoothed prediction data, thereby misleading remote operators.

[0034] Specifically, multi-axis force analysis is performed on real-time multi-source operational data to synthesize a spatial resistance vector. Based on this spatial resistance vector, a spatial distortion scale within a short time window is calculated. This spatial distortion scale is then used to perform confidence arbitration on the edge prediction model, generating arbitration commands. In unstructured environments such as steel smelting, predictive control algorithms are typically deployed at the edge to provide early resistance feedback in order to overcome network transmission latency. However, no edge computing model can accurately simulate the highly random mechanical properties of scrap steel entanglement in an extremely short time. When the grab bucket cuts into unknown consolidated cold steel, if the model still outputs the estimated force of loose furnace material, it will apply incorrect force feedback to the operator, creating a model illusion. This embodiment deeply analyzes the degree of distortion in the force direction within the underlying sensor data and cross-checks it with the model's confidence level. Once a model distortion risk is detected, an arbitration mechanism is immediately triggered for downgrading, cutting off the interference of erroneous prediction commands on the bidirectional closed-loop control system. The original sequence of multi-dimensional force sensors within the short time window, as well as the predicted force vector synchronously output by the edge prediction model, are retrieved.

[0035] The edge prediction model employs a deep network architecture composed of multi-layer temporal convolutional networks and long short-term memory units. Its training process involves collecting historical multi-source operational parameters of heavy excavating equipment operating in conventional homogeneous soil as a sample set. This sample set includes rotational angular velocity sequences and synchronously sampled real three-dimensional spatial resistance vector sequences. The sample set is divided into training and validation sets, with an initial learning rate of 0.001. An adaptive moment estimation algorithm is used to perform gradient descent, and the mean squared error is used as the loss function to calculate the output bias and backpropagate to update the connection weights of network nodes. Convergence is considered complete when the model's loss value on the validation set fluctuates by less than 0.5% for ten consecutive training epochs. In unstructured scenario applications, the edge prediction model's input ports continuously receive the end effector's attitude matrix and historical readings from multi-axis pressure sensors within the current short time window. The temporal convolutional layers within the model are responsible for extracting the temporal evolution of the excavation force state, while the long short-term memory units are responsible for retaining long-term mechanical motion inertial memory. The model's output port generates the predicted force vector for the next control cycle, and simultaneously calls the normalized exponential function layer at the end of the model to output the maximum class probability value as the prediction confidence, which is then used for subsequent comparison and verification operations.

[0036] The duration parameter of the short time window is set as follows: This duration parameter can be calculated by mapping the product of the average rotational angular velocity of the end effector and the basic sampling period of the multidimensional force sensor, combined with the system communication throughput. In this embodiment... The optimal sampling interval is 50 milliseconds. This is chosen because, in heavy excavation operations, this interval contains enough sampling points to filter out conventional mechanical vibration clutter, while also ensuring timely capture of abrupt changes in force during rigid collisions, preventing data accumulation and resulting in judgment lag. The original sequences from the multi-dimensional force sensors are synthesized into a three-dimensional spatial resistance vector; that is, through matrix transformation, the original sequences from each independent dimension are combined into a three-dimensional spatial resistance vector in a unified coordinate system. The directional divergence of the three-dimensional spatial drag vector within this time window is evaluated, and a spatial force distortion index describing the degree of force unevenness is calculated as a spatial distortion scale. The calculation relationship of this distortion index is constructed as follows: , This represents the spatial force distortion index output by the calculation, where N is the total number of sampling points within the short time window. The mean vector of the three-dimensional spatial resistance vectors within this window, the variable The hyperparameter is chosen because when the excavator is suspended and unloaded, the magnitude of the three-dimensional spatial resistance vector approaches zero. This small positive number can prevent division-by-zero crashes in the algorithm's underlying layers. This calculation relationship transforms the complex force state into a scalar describing the degree of distortion and divergence in the force direction, intuitively reflecting the severity of the external unstructured environment. The spatial force distortion index is cross-checked with the prediction confidence output of the edge prediction model. When the spatial force distortion index is detected to be higher than the divergence distortion threshold and the prediction confidence is greater than the lower confidence threshold, the model output is deemed to have a risk of distortion, and a downgraded confidence level scalar is generated as an arbitration instruction. The divergence distortion threshold is set to... The divergence distortion threshold can be obtained by performing calibration tests in a standard homogeneous soil layer, extracting the high-frequency statistical distribution of the three-dimensional resistance vector divergence during the test, and then weighting it by taking its upper quartiles. In this embodiment... The preferred value is 0.45. This is chosen because the material distribution in scrap yards is extremely uneven. When one side of the grab collidees with a large steel block, the force direction will undergo a violent deflection or distortion within a very short time. This value precisely exceeds the extreme value of force deflection during the stripping of conventional loose furnace charge, thus pinpointing the actual force distortion caused by asymmetrical collisions. The prediction confidence level is set to [value missing]. The confidence threshold is set to The confidence threshold can be calculated by backtracking the historical output data of the edge prediction model in the structured environment validation set and extracting the lower boundary of the confidence probability density curve when the model is in a stable convergence state. In this embodiment... The preferred value is 0.85. This is because when the edge prediction model outputs a high confidence level greater than this value but the underlying sensor detects severe stress distortion, it means that the model is trapped in a logical loop of misclassifying hard cold steel as loose soft material. This extreme conflict between high confidence and high distortion is a reliable signal that the algorithm will distort when it faces unknown unstructured material properties, thereby forcibly triggering a confidence level downgrade and generating an arbitration instruction.

[0037] Specifically, the model output distortion state is determined according to the arbitration instruction. When a distortion state is determined, the compressed link is cut off, and the generation of the dimensionality reduction direct transmission sequence of the original message is triggered for direct transmission and uploading configuration. When encountering sudden events such as the bucket cutting into an unknown reef, if the edge prediction model is determined to have a high risk of distortion, continuing to use the existing compressed link to transmit the estimated force data will send erroneous force feedback with model illusions to the remote control station. The existing conventional solution is to switch back to full original data transmission, but the full data from the multi-axis force sensors of heavy excavators at a kilohertz sampling rate will instantly crowd out the limited wireless communication bandwidth, thereby inducing system oscillations. This embodiment adopts a mechanism of cutting off the compressed link and performing dimensionality reduction direct transmission of the original message, which reduces the transmission load while discarding model error illusions. First, the trust level scalar in the arbitration instruction is received, and the current compressed link working mode of the edge prediction model is monitored. Then, the trust level scalar is parsed, and the corresponding frequency domain parsing algorithm package is called. When the trust level scalar is less than the preset reliability degradation threshold, the black-box compressed link of the edge prediction model is cut off. The reliability degradation threshold is set to The reliability degradation threshold can be calculated by cross-mapping the false alarm rate distribution of the statistical edge model in the structured test set and combining it with the overflow trigger probability of the hydraulic cylinder safety valve. In this embodiment... The optimal value is 0.65. This is chosen because when the confidence level is below this value, the risk of erroneous force feedback causing the robotic arm to overload and trigger the safety valve overflow increases exponentially. In this case, the transmission of black-box prediction data must be forcibly terminated. After disconnecting the link, the original force sensor sequence is transformed and waveform components are extracted using a frequency domain analysis algorithm package. This includes performing a Fast Fourier Transform on the original force waveform, retaining only the basic waveform and the preceding waveform. The waveform component data of the second-highest energy order are used to form a direct transmission configuration, and the extracted waveform components are directly transmitted in their original format. The time window length for the Fast Fourier Transform is set to... The length of this time window can be estimated by measuring the tactile nerve conduction delay of the operator's hand and the refresh cycle of the excavator's main control loop. In this embodiment... The optimal value is 20ms. This duration is chosen because it is shorter than the perceptible drag hysteresis blind zone, ensuring that the dimensionality reduction analysis will not introduce new destructive delays into the bidirectional closed-loop control system. Number of high-energy orders In this embodiment, the minimum harmonic order required for the cumulative energy to occupy a specific proportion of the total energy can be calculated by collecting typical energy spectrum diagrams of excavators cutting homogeneous hard rock. The preferred value is 5. The reason for this choice is that the core energy of the rigid collision of the consolidated cold steel is concentrated in the fundamental frequency and the first five harmonic components. Retaining only the first five orders not only removes the sensor's thermal noise and mechanical loosening noise, but also reduces the amount of data sent by more than 80%, avoiding network congestion caused by sudden direct transmission. This ensures the operator's real tactile perception while stabilizing the communication link.

[0038] Specifically, the system synchronously collects the integral of the work done at the drive end and the increment of the kinetic energy at the execution end to calculate the operating damping loss ratio. Based on this ratio, it actively releases and allocates network bandwidth to generate an adaptive scheduling strategy. When the heavy-duty material handling machine's robotic arm forcibly strips slab-formed scrap steel or hard slag from the furnace bottom for extended periods, a large amount of drive work is converted into internal energy and frictional heat to damage the slab structure, causing a sharp increase in the operating damping of the mechanical system. Under such high-dissipation and slow-displacement conditions, maintaining a high frequency of uploading underlying tactile sensor data would not only result in a significant waste of communication bandwidth but also clog the transmission channels of visual and other environmental perception terminal devices, preventing operators from clearly understanding the on-site conditions.

[0039] This embodiment proactively reduces the transmission frequency of low-value data by evaluating the energy loss ratio on the execution side, thus allocating network resources to more critical visual monitoring equipment. The operating damping loss ratio is obtained by acquiring the integral work of the drive motor in the power system and the kinetic energy increment data corresponding to the actual displacement of the end effector. The on-board control unit synchronously collects the terminal voltage and stator current sequences of the drive motor, calculates the integral work through time-domain integration, and simultaneously collects the speed variable of the end effector to calculate the kinetic energy increment data. The integral work and kinetic energy increment data are processed for unified data and time-series alignment, both converted to Joules, and Kalman filtering is used to eliminate sampling phase differences. The difference between the integral work and the kinetic energy increment data is calculated to obtain the operating damping loss ratio of the current operation process. The calculation relationship of this ratio is constructed as follows: ,in For the operating damping loss ratio, the variable For the integral of work done, the variable is... For kinetic energy increment data, variables Compensation for changes in gravitational potential energy, variable To prevent zero-valued positive values, this constant can be extracted by measuring the microampere-level leakage current noise floor of the electronic control system. When the operating damping loss ratio exceeds the high dissipation threshold three times consecutively, an environmental adaptive bandwidth allocation strategy is implemented to proactively extend the effective lifespan of tactile sensing data packets to reduce transmission frequency, and allocate the released network bandwidth resources to the environmental sensing terminal equipment. The high dissipation threshold is set to... The high dissipation threshold can be calculated by statistically analyzing the average steady-state energy dissipation distribution when an excavator cuts and crushes high-hardness granite in simulated experimental fields with rock layers of different hardness. In this embodiment... A value of 0.85 is preferred because this value can accurately distinguish between the grasping of conventional loose materials and the high-resistance, hard pulling of solidified cold steel. When energy loss exceeds 85%, it means that the end mechanism has stalled, and the high-frequency updates of tactile data have lost their real-time control significance. The environmental adaptive bandwidth allocation strategy includes constructing a mapping relationship based on the operating damping loss ratio and the set baseline survival time to calculate the dynamic survival time of the tactile sensing data packets. The baseline survival time is set to... The baseline survival time can be extracted by analyzing the operator's visual-tactile fusion response latency, in this embodiment. The preferred value is 100ms. The calculation relationship for dynamic survival time is as follows: ,in The delay stretching factor can be calculated by evaluating the maximum available bandwidth margin of the network link. In this embodiment, a value of 2.0 is preferred because it can extend the lifespan of tactile data by up to two times under high dissipation conditions. The sliding window step size of the edge gateway buffer queue is adjusted according to the dynamic lifespan, and downsampling is performed on the original tactile data stream to reduce the packet generation frequency per unit time. The sliding window step size is set to... The sliding window step size can be calculated and extracted by dividing the dynamic survival time by the hardware interrupt triggering cycle of the underlying sensor. In this embodiment... The optimal value is 10, chosen because it reduces the frequency of tactile data generation to one-tenth, significantly decreasing message backlog. At the media access control layer, downsampled tactile data packets are intercepted, their quality of service labels are rewritten, and the tactile data packets are downgraded from the high-priority queue to the weighted fair queue. The token consumption rate of the tactile data channel is monitored, and the remaining margin of the token consumption rate is extracted. Redundant tokens corresponding to this remaining margin are dynamically injected into the token bucket of the visual data channel, expanding the available bandwidth quota of the environmental perception terminal device. The transfer ratio of token injection can be dynamically calculated by monitoring the frame rate jitter and encoding compression ratio of the current visual channel. In this embodiment, the preferred transfer ratio is 90%, chosen because it allows most of the bandwidth saved from the tactile channel to be smoothly transferred to visual terminals such as high-definition cameras. This helps operators judge the construction site with clear, uninterrupted real-time images when the robotic arm is stuck in a high-resistance situation, solving the blind spot caused by the failure of a single tactile sensor.

[0040] The threshold and weight settings involved in this embodiment can be set by default according to the present invention, or can be set by those skilled in the art.

[0041] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0042] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0043] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0044] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0045] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. An adaptive data transmission and scheduling method for sudden transient power events, characterized in that: Includes the following steps: S1. Obtain real-time multi-source operating data of the end effector, extract transient pressure derivative and displacement attenuation scale from the real-time multi-source operating data, and generate rigid collision state quantities based on the transient pressure derivative and displacement attenuation scale. S2. Based on the rigid collision state quantity, perform a lifetime assessment on the sensing data packet to generate a dynamic survival time. Combine the dynamic survival time with the priority reconstruction of the medium access control layer to establish a preemption channel. Transmit the resistance sensing message back through the preemption channel. S3. Perform multi-axis force analysis and synthesis of spatial resistance vector on the real-time multi-source operation data, calculate the spatial distortion scale within a short time window based on the spatial resistance vector, and use the spatial distortion scale to perform confidence arbitration on the edge prediction model to generate arbitration instructions; S4. Determine the distortion state of the model output based on the arbitration instruction. When the distortion state is determined, cut off the compressed link and trigger the generation of the direct transmission sequence of the original message for direct transmission and uploading configuration. S5. Synchronously collect the work integral of the drive end and the kinetic energy increment of the execution end to calculate the running damping loss ratio, and generate an adaptive scheduling strategy by actively releasing and allocating network bandwidth based on the running damping loss ratio.

2. The adaptive data transmission and scheduling method for sudden transient power events according to claim 1, characterized in that: The process of generating the rigid collision state quantity includes: S11. Collect the working cylinder pressure sequence and piston displacement sequence of the end effector through a multi-axis pressure sensor and a position encoder; S12. Perform time-series alignment and noise reduction smoothing on the pressure sequence and displacement sequence; S13. For the aligned sequence, extract the transient pressure change gradient and the position offset attenuation ratio per unit time and perform coupled cross-operation. S14. When the transient pressure change gradient is greater than the set step rise threshold and the position offset attenuation ratio is less than the set stagnation lower limit, output the rigid impact evaluation parameter as the rigid collision state quantity.

3. The adaptive data transmission and scheduling method for sudden transient power events according to claim 2, characterized in that: The step of performing lifetime assessment on the sensed data packet based on the rigid collision state quantity to generate a dynamic lifetime, and combining the dynamic lifetime to implement priority reconstruction of the media access control layer and establish a preemptive channel, includes: receiving the rigid collision state quantity and synchronously obtaining the current queue backlog status of the communication media access control layer; extracting the regular service priority identifier of the sensed data packet to be sent; substituting the rigid collision state quantity into the lifetime assessment mapping model to calculate the dynamic lifetime of the sensed data packet; When the dynamic survival time is less than the preset delay-sensitive control line, a forced suspension instruction is triggered at the underlying network protocol layer to temporarily suspend the transmission of non-critical services, and a dynamic priority control instruction is generated to establish the preemption channel for the resistance-aware message.

4. The adaptive data transmission and scheduling method for sudden transient power events according to claim 3, characterized in that: Multi-axis force analysis is performed on the real-time multi-source operating data to synthesize a spatial resistance vector. A spatial distortion scale within a short time window is calculated based on the spatial resistance vector. The spatial distortion scale is then used to perform confidence arbitration on the edge prediction model to generate an arbitration instruction. This process includes: retrieving the original sequence of multi-dimensional force sensors within the short time window, and the predicted force vector synchronously output by the edge prediction model; synthesizing the original sequence of multi-dimensional force sensors into a three-dimensional spatial resistance vector; evaluating the directional divergence of the three-dimensional spatial resistance vector within the time window, and calculating a spatial force distortion index describing the degree of force unevenness as the spatial distortion scale; cross-checking the spatial force distortion index with the prediction confidence output by the edge prediction model. When the spatial force distortion index is detected to be higher than the divergence distortion threshold and the prediction confidence is detected to be greater than the lower confidence threshold, it is determined that the model output has a risk of distortion, and a downgraded confidence level scalar is generated as the arbitration instruction.

5. The adaptive data transmission and scheduling method for sudden transient power events according to claim 4, characterized in that: The process of generating the direct transmission configuration from the reduced-dimensional direct transmission sequence of the triggering original message includes: receiving the trust level scalar in the arbitration instruction and monitoring the compressed link working mode of the current edge prediction model; parsing the trust level scalar and calling the corresponding frequency domain analysis algorithm package; when the trust level scalar is less than the preset reliability degradation threshold, cutting off the black-box compressed link of the edge prediction model; using the frequency domain analysis algorithm package to convert the original sequence of the force sensor and extract waveform components to form the direct transmission configuration, and directly transmitting the extracted waveform components in the original format.

6. The adaptive data transmission and scheduling method for sudden transient power events according to claim 5, characterized in that: The operating damping loss ratio is obtained by acquiring the integral of the work done by the drive motor of the power system and the kinetic energy increment data corresponding to the actual displacement of the end effector. Perform data unification and timing alignment processing on the work integral and the kinetic energy increment data; The difference between the integral of work done and the incremental kinetic energy data is calculated to obtain the operating damping loss ratio of the current operation process. When the operating damping loss ratio is greater than the high dissipation threshold for three consecutive times, an environment adaptive bandwidth allocation strategy is executed to actively extend the effective life cycle of the tactile sensing data packets and allocate the released network bandwidth resources to the environmental sensing terminal device.

7. The adaptive data transmission and scheduling method for sudden transient power events according to claim 6, characterized in that: The environment adaptive bandwidth allocation strategy includes: constructing a mapping relationship between the operating damping loss ratio and the set baseline survival time, and calculating the dynamic survival time of the tactile sensing data packet; The sliding window step size of the edge gateway buffer queue is adjusted according to the dynamic survival time, and downsampling is performed on the original tactile data stream. The downsampled tactile sensing data packets are intercepted at the media access control layer, their quality of service labels are rewritten, and the tactile sensing data packets are downgraded from the high-priority queue to the weighted fair queue. Monitor the token consumption rate of the tactile data channel, extract the decrease margin of the token consumption rate, and dynamically inject the redundant tokens corresponding to the decrease margin into the token bucket of the visual data channel to expand the available bandwidth quota of the environmental perception terminal device.

8. The adaptive data transmission and scheduling method for sudden transient power events according to claim 3, characterized in that: The priority reconfiguration includes configuring a regular scheduling mode and a priority preemption mode; wherein, the priority preemption mode is set to be triggered only when a hard collision occurs as represented by the rigid collision state quantity, and the dynamic survival time is greater than or equal to the control delay red line set by the system.

9. The adaptive data transmission and scheduling method for sudden transient power events according to claim 3, characterized in that: The survival assessment mapping model calculates the instability acceleration of the control system based on the second derivative of the transient pressure change gradient, and the length of the dynamic survival time is negatively correlated with the instability acceleration of the control system.

10. The adaptive data transmission and scheduling method for sudden transient power events according to claim 5, characterized in that: The process of converting and extracting waveform components from the original force sensor sequence using the frequency domain analysis algorithm package includes performing a fast Fourier transform on the original force waveform and retaining only the basic waveform and the waveform component data of the first N high-energy orders to form the direct transmission configuration.

Images