Industrial robot adaptive control system fusing situation prediction mechanism

Abstract

The invention discloses an industrial robot adaptive control system fused with a situation prediction mechanism, and particularly relates to the field of situation prediction based on a feedforward network and fuzzy logic, comprising the following steps: in the execution process of an industrial robot, obtaining state input data through continuous sampling; and inputting the state input data into a feedforward neural network with a time sequence feature mapping capability to obtain a prediction output result of the task trend of the robot. A dynamic entropy deviation cooperation mechanism between a feedforward neural network and a fuzzy logic reasoning model is constructed to realize modal attribution conflict identification and control right coordination of a critical state of a multi-modal task, so that continuity of a task response path and stable operation of a control system in a fuzzy transition interval are ensured; therefore, the problems of modal dislocation and out-of-control execution caused by lack of entropy weight coupling of feedforward prediction output and fuzzy reasoning judgment in a control right decision domain are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of situation prediction based on feedforward network and fuzzy logic, and more particularly, to an industrial robot adaptive control system fusing situation prediction mechanism. BACKGROUND

[0002] In the multi-modal task execution scenario, the current industrial robot system generally relies on the feedforward network to construct the trend mapping of the state sequence to predict the evolution direction of the future task in advance, perform the so-called situation prediction operation, and thus realize the preloading of the instruction response and the forward scheduling of the control rhythm; however, the confidence output of the situation prediction is highly dependent on the prior distribution assumption and the input structure stability during training, and in the non-static or boundary fuzzy working condition, the prediction mechanism cannot provide reliable judgment for the "state attribution uncertain region", at this time the control system often introduces the fuzzy logic mechanism, and through constructing the fuzzy membership function, the attribution possibility of the state of the robot between multiple control modes is determined, so as to realize the smooth transition between control strategies and the flexible connection of the modal evolution path.

[0003] However, when the industrial robot faces the critical transition interval in the multi-modal control conversion process, the following problem chain often occurs:

[0004] The first layer problem is the structural conflict of modal attribution judgment:

[0005] The current fuzzy logic system often outputs the fuzzy state with high membership degree overlap near the control mode conversion boundary, that is, a state belongs to multiple modal rule systems (such as task A and task B) at the same time, such a state is essentially a "control fuzzy band", and the feedforward network may output highly concentrated prediction results based on the trend inference characteristics of the fixed structure within a short time, directly determine that the state should belong to a completely different modal task (such as task C), and thus structural divergence occurs between the prediction results and the fuzzy determination;

[0006] The second layer problem is that the situation prediction output and the fuzzy reasoning output lack a unified decision domain:

[0007] Since the fuzzy logic control structure and the feedforward network prediction mechanism are independent of each other in the information flow structure and the objective function design, the reasoning results of the same state by the two often cannot be effectively negotiated at the control right management layer, resulting in the system generating multiple mutually exclusive control instructions for the same state, and the fragmentation of the decision domain makes the control system have to reserve multiple control paths at the same time and perform competitive scheduling of resources;

[0008] The third layer problem is the loss of control right responsibility and the loss of focus of response:

[0009] The above multi-path divergence will eventually lead to the blurring of the control focus, and the system will mistakenly regard the uncertain state as a high-confidence prediction result without establishing a unified entropy weight consensus, leading to unstable multi-scattering state of the control behavior, and in severe cases, the system will be frozen, the strategy rollback will fail, and the task response will be chaotic.

[0010] Based on the above, the problems existing in the prior art can be summarized as follows:

[0011] During the evolution of the multi-mode task situation, the industrial robot control system lacks an entropy weight coordination mechanism between the one-way prediction output of the feedforward network and the multi-state membership reasoning of the fuzzy logic, leading to structural misplacement of state criticality judgment, and further causing the risk of systematic collapse of control decision domain splitting and task response responsibility migration out of control.

[0012] This problem reflects the fundamental fault in the situation prediction and fuzzy logic coordination control mechanism of the prior art, and is the most potentially destructive problem in the current adaptive control system, but it is easily overlooked. SUMMARY

[0013] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an industrial robot adaptive control system integrating situation prediction mechanism, which builds a dynamic entropy deviation coordination mechanism between the feedforward neural network and the fuzzy logic reasoning model, to realize the mode attribution conflict recognition and control right coordination of the multi-mode task critical state, and further ensure the continuity of the task response path and the stable operation of the control system in the fuzzy transition interval, thereby solving the mode misplacement and execution out of control problems caused by the lack of entropy weight coupling of the feedforward prediction output and the fuzzy reasoning judgment in the control right decision domain.

[0014] To achieve the above object, the present application provides the following technical scheme: an industrial robot adaptive control system integrating situation prediction mechanism, comprising a situation prediction module, a judgment attribution module, a comparison module, an identification misplacement module, a buffer module, and a coordination module.

[0015] The situation prediction module is used to acquire state input data by continuous sampling during the execution of the industrial robot, and input the state input data into a feedforward neural network with time sequence feature mapping capability to obtain the prediction output result of the robot task trend.

[0016] The judgment attribution module is used to input the state input data in the situation prediction module into a fuzzy logic reasoning model constructed based on a fuzzy membership function, to perform fuzzy membership reasoning on the attribution probability of the current state of the industrial robot between multiple control modes, and obtain the fuzzy membership value of the current state under each control mode.

[0017] The comparison module is configured to build a collaborative discriminant model, and to perform collaborative comparison between the prediction output obtained by the situation prediction module and the fuzzy membership value obtained by the judgment attribution module, so as to extract difference information between the two in the modal judgment result.

[0018] The recognition misplacement module is configured to calculate a control decision entropy deviation index value of the difference information, and to determine whether the current state of the industrial robot is in a modal critical misplacement state. When the entropy deviation index value exceeds a preset threshold, the conflict risk state is identified.

[0019] In a preferred embodiment, the buffer module is configured to start a fuzzy priority control mechanism when the conflict risk state is identified, to build a neutral task buffer control band by locking the modal direction of the lower limit of the entropy gradient in the fuzzy membership value output by the fuzzy logic inference model, and to shield the immediate control signal output of the feedforward neural network in the situation prediction module.

[0020] The coordination module is configured to observe and feed back the evolution process of the real-time state input data of the industrial robot within a preset time window in the neutral task buffer control band, to extract a dynamic response feature representing the state evolution trajectory of the industrial robot, and to determine the trend of the entropy deviation index value returning to stability based on the extracted dynamic response feature. When the trend of returning to stability recovers to a preset fluctuation range, the immediate control signal output of the situation prediction module is restored, forming a coordinated control between the judgment attribution module and the situation prediction module.

[0021] In a preferred embodiment, the situation prediction module is configured to number the original state input data obtained by continuous sampling in time sequence during the execution of the industrial robot, and to build an embedded state sequence matrix with a preset sliding window length. Each row in the state sequence matrix represents the distribution trajectory of the state parameters at adjacent time steps, and each column corresponds to the change dimension of a type of state parameters.

[0022] The time-coupled coding operation is performed on the state sequence matrix, and the partial derivative calculation is performed on the change direction between the state parameters at each time step and the adjacent time steps before and after it by using a position offset nesting function, so as to build a difference mapping tensor representing the state evolution trend.

[0023] A feedforward neural network with at least three hidden layers is built, and a response regulation unit is embedded in each hidden layer. Each response regulation unit includes a group of response gating functions constructed based on the gradient directionality of the state parameters. The response gating functions are configured to receive the difference mapping tensor as input, and to generate a compressed and encoded state trend signal at the output end of the response gating function.

[0024] The feedforward neural network is trained, and during the training process, a trend residual function is used as a loss indicator, the trend residual function is used to evaluate the trend deviation between the state trend signal and the trajectory of the industrial robot formed in the real task path; further comprising using a back propagation algorithm to optimize the connection weight parameters between each layer in the feedforward neural network;

[0025] After the training is completed, the state input data obtained by real-time sampling is converted into a state sequence matrix within the current time window, and a difference mapping tensor is generated according to the same steps, and the difference mapping tensor is input into the trained feedforward neural network, and a prediction output result including a task trend direction vector, a modal shift probability distribution and a path transition rate is output;

[0026] The prediction output result is input into a decision attribution module to provide state trend-based task evolution path prediction support for real-time attribution judgment of critical control states by a fuzzy logic reasoning model.

[0027] In a preferred embodiment, in the decision attribution module, for each control mode, a fuzzy membership function with an adjustable shape parameter is constructed, the fuzzy membership function is constructed in a piecewise nonlinear mapping form, and the membership starting point, the membership peak point and the decay inflection point are set according to the state boundary conditions in the industrial robot task, to describe the membership transition characteristics of the state parameters in the modal boundary interval;

[0028] Based on the state membership expression requirements under multiple control modes, a rule base structure of the fuzzy logic reasoning model is designed, and each fuzzy rule is expressed as a conditional association mapping between the input state parameter interval and the corresponding modal label.

[0029] In a preferred embodiment, in the decision execution phase of the decision attribution module, the current state input data is arranged according to the preset input dimension order in the fuzzy rule base structure, and fuzzy coding operation is performed on the state parameters of each input dimension; the fuzzy coding result is sequentially subjected to rule matching operation, and all fuzzy rules satisfying the premise condition are marked; the activation degree value of each activated fuzzy rule is calculated, and the control mode label corresponding to the activation degree value is bound; the activation degree values of all activated rules under the same control mode are subjected to weighted accumulation to generate the modal response coefficient corresponding to each control mode, and the vector set composed of the modal response coefficients is an intermediate representation tensor for multi-modal fusion reasoning;

[0030] The intermediate representation tensor composed of the modal response coefficients is subjected to normalization processing operation; according to the result of the normalization processing operation, a fuzzy membership value set containing all control modes is output, and the fuzzy membership value set is used as reference data for deciding the attribution distribution of the current state of the industrial robot among the control modes.

[0031] In a preferred embodiment, the task trend prediction result output by the situation prediction module and the fuzzy membership value output by the judgment attribution module are mapped into a joint modal representation vector one by one corresponding to the control mode by the comparison module to construct a cooperative discrimination model, the vector cosine value and the normalized offset between the prediction channel of the situation prediction module and the fuzzy channel of the judgment attribution module are calculated, and the difference vector between the two in the control mode attribution dimension is extracted as the difference information of modal discrimination.

[0032] In a preferred embodiment, when the identification misplacement module calculates the control decision entropy deviation index value of the difference information, the entropy deviation index value is constructed based on the deviation degree of each modal dimension in the difference vector to form an information entropy distribution, and the standard deviation entropy value of the information entropy distribution is calculated.

[0033] The deviation degree includes an absolute difference calculation operation between the prediction output result and the fuzzy membership value in each modal dimension, the difference between the prediction probability of the prediction output result and the corresponding fuzzy membership value in each control mode is calculated, and after the difference values of all modal dimensions are normalized, the deviation weight distribution is obtained, which is used as the probability proportion basis of each modal dimension in the control decision entropy deviation index value.

[0034] Technical effects and advantages of the present application:

[0035] 1. The present application realizes the identification of state critical misplacement by constructing a cooperative discrimination model of situation prediction and fuzzy reasoning and introducing an entropy deviation index, and solves the control confusion problem caused by the lack of entropy weight cooperation of prediction and fuzzy mechanism.

[0036] 2. The state sequence matrix is constructed by using a sliding window and the difference mapping tensor is extracted, and the response gating function in the feedforward neural network is combined to improve the continuity of task trend prediction and the forward response ability in complex dynamic scenarios.

[0037] 3. The modal attribution distribution is constructed by fuzzy rule matching and response coefficient calculation, and the fuzzy membership value set is output by normalization operation, which enhances the distinguishability between multi-modal control modes and the compatibility of modal attribution expression.

[0038] 4. The control decision entropy deviation index is constructed based on the difference of modal discrimination results, the system can identify the risk state before the conflict, enter the buffer control band in time, reduce the task execution interruption risk and improve the overall stability.

[0039] 5. The system can adaptively restore the feedforward control path by continuously feeding back the state evolution trajectory and detecting the entropy deviation value back to the stable trend, realizing the closed-loop connection between the fuzzy and prediction mechanisms. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 A situation prediction flowchart for the system in the present application.

[0041] Figure 2 A determination attribution flowchart for the system in the present application.

[0042] Figure 3 A modal comparison and misplacement recognition flowchart for the system in the present application.

[0043] Figure 4 A buffer control band construction flowchart for the system in the present application.

[0044] Figure 5 A coordination control recovery flowchart for the system in the present application.

[0045] Figure 6 A system module schematic diagram of the present application. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the protection scope of the present application.

[0047] With reference to the drawings in the description Figures 1-6 , an industrial robot adaptive control system fusing a situation prediction mechanism according to an embodiment of the present application comprises a situation prediction module, a determination attribution module, a comparison module, a misplacement recognition module, a buffer module, and a coordination module.

[0048] The situation prediction module is configured to acquire state input data by continuous sampling during the execution of the industrial robot, and input the state input data into a feedforward neural network with a time sequence feature mapping capability, so as to obtain a prediction output result of the robot task trend.

[0049] The determination attribution module is configured to input the state input data in the situation prediction module into a fuzzy logic reasoning model constructed based on a fuzzy membership function, to perform fuzzy membership reasoning on the attribution probability of the current state of the industrial robot between multiple control modes, and obtain fuzzy membership values of the current state under each control mode.

[0050] The comparison module is configured to construct a collaborative discrimination model, to collaboratively compare the prediction output result obtained in the situation prediction module with the fuzzy membership values obtained in the determination attribution module, and extract difference information between the two in the modal determination result.

[0051] The recognition misplacement module is configured to calculate a control decision entropy deviation index value of the difference information, and determine whether a current state of the industrial robot is in a modal critical misplacement state. When the entropy deviation index value exceeds a preset threshold, the recognition misplacement module identifies a conflict risk state.

[0052] The buffer module is configured to start a fuzzy priority control mechanism when the conflict risk state is identified, lock a modal direction of an entropy gradient lower limit in a fuzzy membership value output by a fuzzy logic inference model, and construct a neutral task buffer control band to shield an immediate control signal output of a feedforward neural network in the situation prediction module.

[0053] The coordination module is configured to observe and feed back an evolution process of real-time state input data of the industrial robot within a preset time window in the neutral task buffer control band, extract a dynamic response feature representing a state evolution trajectory of the industrial robot, and determine a return-to-stable trend of the entropy deviation index value based on the extracted dynamic response feature. When the return-to-stable trend is restored to a preset fluctuation range, the coordination module restores the immediate control signal output of the situation prediction module, forms a coordinated control between the determination attribution module and the situation prediction module, and ensures continuity and stability of a task response process.

[0054] The coordination module is configured to perform the following operations in the neutral task buffer control band:

[0055] 1) State evolution observation: Obtain continuous state input data of the industrial robot within a current preset time window, and construct a state evolution trajectory of the industrial robot within the window.

[0056] 2) Dynamic response feature extraction: Based on the state evolution trajectory, extract a dynamic response feature reflecting a change trend, a response directionality, and a disturbance stability degree.

[0057] 3) Entropy deviation return-to-stable judgment: Use the extracted dynamic response feature to calculate an evolution trend of the control decision entropy deviation index value, and determine whether the trend returns to a system preset fluctuation stable range.

[0058] 4) Control signal restoration: When the return-to-stable trend of the entropy deviation index value meets a restoration condition, the immediate control signal output of the situation prediction module is re-enabled, a coordinated control path with the determination attribution module is formed, and task response continuity and stability are ensured.

[0059] The state input data of the robot within a time window [t0, t0+T] is defined as:

[0060] S(t)={s1(t),s2(t),…,s n (t)},t∈[t0,t0+T]

[0061] The dynamic response feature vector of the state evolution trajectory is constructed as:

[0062]

[0063] Based on the dynamic response feature vector, the control decision entropy deviation index value recovery trend function is calculated as:

[0064]

[0065] When the following conditions are met, the output control signal recovery marker is output:

[0066] Ξ(t0)≤ε th

[0067] Where S(t) is the multi-dimensional state input data sequence of the industrial robot at time t, including position, speed, attitude, force and other parameters in actual application; n is the total number of state input data dimensions; T is the time window period; Φ(·) is a dynamic response feature construction function, the input of the dynamic response feature construction function includes first derivative (speed), second derivative (acceleration) and attitude transformation gradient Δθ(t), and the output is used to measure the current state evolution intensity and directionality; F(t0) represents the dynamic response feature vector extracted at the beginning of the time window t0; P mod (t) represents the modal attribution difference vector constructed between the feedforward prediction result and the fuzzy membership value at each time t (the probability basis for entropy deviation calculation); Ω(F(t)) is a response weighting function, which is used to adjust the influence intensity of the entropy value at each time point on the overall trend according to the dynamic response feature; Λ(·) is a recovery trend aggregation function, which is used to perform integrated evaluation on the weighted entropy evolution in the entire time window; Ξ(t0) represents the control decision entropy deviation index value trend function of the current time window, which is used to judge whether the recovery condition is reached; ε th is the fluctuation stability upper limit threshold of the entropy deviation trend, which is preset by the system.

[0068] The situation prediction module is used to number the original state input data obtained by continuous sampling in time sequence during the execution of the industrial robot, and to construct a state sequence matrix in an embedded form according to a preset sliding window length, each row of the state sequence matrix representing the distribution trajectory of the state parameters of adjacent time steps, and each column corresponding to the change dimension of a type of state parameters;

[0069] The time-coupled coding operation is performed on the state sequence matrix, and the position offset nesting function is used to calculate the partial derivative of the change direction between each time step state parameter and its adjacent time steps before and after it, so as to construct a difference mapping tensor representing the state evolution trend;

[0070] constructing a feedforward neural network with at least three layers of hidden layers, wherein each of the hidden layers is embedded with a response regulation unit, each of the response regulation units comprises a set of response gating functions constructed based on the gradient directionality of the state parameters, the response gating functions are configured to receive a differential mapping tensor as an input, and generate a compressed encoded state trend signal at the output end of the response gating functions;

[0071] training the feedforward neural network, wherein a trend residual function is used as a loss indicator during the training process, the trend residual function is configured to evaluate the trend deviation between the state trend signal and the trajectory trend formed by the industrial robot in the real task path, and the connection weight parameters between the layers of the feedforward neural network are optimized by using a back propagation algorithm;

[0072] after the training is completed, the state input data obtained by real-time sampling is converted into a state sequence matrix within a current time window, and a differential mapping tensor is generated according to the same steps, the differential mapping tensor is input into the trained feedforward neural network, and a prediction output result including a task trend direction vector, a modal shift probability distribution and a path transition rate is output;

[0073] the prediction output result is input into a decision attribution module to provide state trend-based task evolution path prediction support for real-time attribution judgment of critical control states by a fuzzy logic reasoning model;

[0074] wherein the differential mapping tensor generated according to the same steps represents the manner of constructing an embedded form state sequence matrix according to the preset sliding window length, the manner of performing time coupling coding operation on the state sequence matrix, and the manner of using a position offset nested function to calculate the partial derivative, and the differential mapping tensor representing the current state evolution trend is generated.

[0075] in the decision attribution module, for each control mode, a fuzzy membership function with adjustable shape parameters is constructed, the fuzzy membership function is constructed in a piecewise nonlinear mapping form, the membership starting point, the membership peak point and the decay inflection point are set according to the common state boundary conditions in the industrial robot task, to describe the membership transition characteristics of the state parameters in the modal boundary interval; the common state boundary conditions include robot position limit, velocity mutation point, attitude stability boundary, load critical value, environmental disturbance threshold and control mode switching region;

[0076] based on the state membership expression requirements under multiple control modes, a rule base structure of the fuzzy logic reasoning model is designed, each fuzzy rule is represented as a conditional association mapping between the input state parameter interval and the corresponding modal label, and a modal conflict constraint coefficient is introduced to adjust the reasoning priority weight of the conflict modal when multiple rules are activated simultaneously.

[0077] In the decision execution phase of the decision attribution module, the current state input data is arranged in the order of the preset input dimension in the fuzzy rule base structure, and the state parameters of each input dimension are subjected to fuzzy coding operation; the fuzzy coding results are subjected to rule matching operation in sequence, and all fuzzy rules meeting the premise conditions are marked; the activation degree value of each activated fuzzy rule is calculated, and the control mode label corresponding to the activation degree value is bound; the activation degree values of all activated rules under the same control mode are subjected to weighted accumulation to generate the modal response coefficient corresponding to each control mode, and the vector set composed of the modal response coefficients is the intermediate representation tensor of multi-modal fusion reasoning;

[0078] The intermediate representation tensor composed of the modal response coefficients is subjected to normalization processing operation, and a distribution consistency constraint factor is introduced in the normalization process to maintain the response value interval separation degree between the control modes; according to the result of the normalization processing operation, a fuzzy membership value set containing all control modes is output, which is used as reference data for determining the attribution distribution of the current state of the industrial robot among the control modes.

[0079] The collaborative discrimination model is constructed through the comparison module, the task trend prediction result output by the situation prediction module is one-to-one mapped to the joint modal representation vector with the fuzzy membership value output by the decision attribution module according to the control mode, the vector cosine value and the normalized offset between the joint modal representation vector in the prediction channel of the situation prediction module and the fuzzy channel of the decision attribution module are calculated, and the difference vector between the two in the control mode attribution dimension is extracted as the difference information of modal discrimination.

[0080] When the control decision entropy deviation index value of the difference information is calculated in the misposition recognition module, the entropy deviation index value is constructed based on the deviation degree of each modal dimension in the difference vector to construct an information entropy distribution, and the standard deviation entropy value of the information entropy distribution is calculated.

[0081] The deviation degree includes an absolute difference calculation operation between the prediction output result and the fuzzy membership value in each modal dimension, the difference between the prediction probability of the prediction output result under each control mode and the corresponding fuzzy membership value is calculated, and after the difference values of all modal dimensions are normalized, the deviation weight distribution is obtained, which is used to construct the probability proportion of each modal dimension in the control decision entropy deviation index value.

[0082] To be whole, this scheme revolves around the core problem of "industrial robots in the process of executing multi-modal tasks, due to the lack of dynamic entropy coordination mechanism between situation prediction mechanism and fuzzy logic reasoning mechanism, leading to control split and response defocus", constructs an industrial robot adaptive control system integrated with situation prediction mechanism, and its workflow is sequentially developed based on six clear functional modules, namely: situation prediction module, judgment attribution module, comparison module, identification misplacement module, buffer module and coordination module.

[0083] In the process of industrial robot executing tasks, the system continuously samples the state input data through the situation prediction module, and encodes it into a state sequence matrix with time sequence relationship; by setting a sliding window structure, this matrix can retain the historical evolution trajectory of state parameters, and on this basis, introduce a time coupling coding mechanism, calculate the gradient direction of each time step state on the front and rear time nodes by means of position offset nesting function, and obtain the difference mapping tensor representing the state evolution trend; then a multi-layer feedforward neural network embedded with response control unit is constructed, and the trend signal is compressed and encoded through the response gating function; the whole training process takes the trend residual function as the loss index, and the connection weight is optimized through the back propagation algorithm to improve the prediction accuracy.

[0084] The judgment attribution module, on the other hand, processes the same state input data in parallel from another path; the system constructs a fuzzy rule base based on the fuzzy logic reasoning mechanism, and maps the state parameters to the membership space corresponding to each control mode through the fuzzy membership function; specifically, the judgment attribution module performs fuzzy coding operation, fuzzy processes the input data, and then outputs the modal response coefficient matrix in the control mode through rule matching and activation degree calculation, forming the integrated expression of the reasoning intermediate tensor; then perform normalization operation with distribution consistency constraint to generate a set of fuzzy membership values that meet the requirements of modal distinguishability and compatibility, to represent the attribution probability distribution of the current state among the control modes;

[0085] The comparison module compares the state trend prediction results output by the situation prediction module with the fuzzy membership values generated by the judgment attribution module, and calculates the difference values in the modal dimension; the identification misplacement module further models the decision entropy of these difference values, and generates a control decision entropy deviation index value according to the modal offset degree between the prediction output and the fuzzy reasoning output; this entropy deviation index value reflects whether the current state is in a modal critical misplacement state, i.e. there is an uncoordinated conflict between the prediction trend and the fuzzy attribution; when the entropy deviation index value exceeds the preset threshold, the system will identify the state as a conflict risk state and activate the buffer mechanism;

[0086] In the buffer module, the system automatically suspends the direct control output of the feedforward network to the state, and instead establishes a neutral task buffer control band according to the lower limit direction of the entropy gradient output in the fuzzy logic reasoning model. The buffer control band can be regarded as a fuzzy priority scheduling interval, temporarily maintaining the robot running in the intermediate state between modes, thereby preventing premature or incorrect mode switching triggered by mispositioning judgment;

[0087] The coordination module continuously monitors the dynamic response changes of the state input data within the set time window in the buffer control band, extracts the state evolution trajectory, and judges whether the control decision entropy deviation index value is back to stable based on the trend change. When the index value falls within the preset fluctuation range, the system considers that the mode consistency is restored, that is, the feedforward prediction and the fuzzy reasoning reach a mode consensus, at which time the feedforward neural network control output is reactivated, and the control right is transferred back from the fuzzy logic mechanism to the feedforward prediction mechanism, forming a smooth connection of task response, ensuring the response stability and task continuity of the entire control system;

[0088] The reason for adopting the above structure design is based on the defect that the existing industrial robot control system cannot identify the mode critical mispositioning state in the situation prediction mechanism, and the fuzzy logic reasoning model can maintain control continuity in the fuzzy boundary, but cannot provide trend prediction ability. Therefore, the system solves the fault problem between the feedforward neural network and the fuzzy reasoning by heterogeneously cooperating the information sources of the two and establishing a transition judgment mechanism between the decision domains through the entropy index, improves the robustness and intelligence of the control system in the multi-mode dynamic switching task.

[0089] The above only describes the preferred embodiments of the present application and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An industrial robot adaptive control system integrating a situation prediction mechanism, comprising a situation prediction module, an attribution determination module, a comparison module, a misalignment identification module, a buffer module, and a coordination module, characterized by: The situation prediction module is used to obtain state input data through continuous sampling during the execution of the industrial robot, and input the state input data into a feedforward neural network with time series feature mapping capability to obtain the prediction output result of the robot task trend; The determination module is used to input the state input data in the situation prediction module into the fuzzy logic reasoning model built based on the fuzzy membership function, perform fuzzy membership reasoning on the attribution probability of the current state of the industrial robot between multiple control modes, and obtain the fuzzy membership value of the current state in each control mode; The comparison module is used to build a collaborative discrimination model, which compares the prediction output results obtained in the situation prediction module with the fuzzy membership values ​​obtained in the determination and attribution module, and extracts the difference information between the two in the modal determination results; The misalignment identification module is used to calculate the control decision entropy deviation index value of the difference information, determine whether the current state of the industrial robot is in a modal critical misalignment state, and when the entropy deviation index value exceeds a preset threshold, it is marked as a conflict risk state.

2. The industrial robot adaptive control system integrating a situation prediction mechanism according to claim 1, characterized in that: The buffer module is used to activate the fuzzy priority control mechanism when a conflict risk state is identified. By locking the modal direction of the lower limit of the entropy gradient in the fuzzy membership value output by the fuzzy logic reasoning model, it constructs a neutral task buffer control zone and shields the immediate control signal output of the feedforward neural network in the situation prediction module. The coordination module is used to observe and feedback the evolution process of the real-time state input data of the industrial robot within a preset time window in the neutral task buffer control band, so as to extract the dynamic response characteristics that characterize the state evolution trajectory of the industrial robot, and judge the stabilization trend of the entropy deviation index value based on the extracted dynamic response characteristics. When the stabilization trend returns to within the preset fluctuation range, the immediate control signal output of the situation prediction module is restored to form coordinated control between the determination attribution module and the situation prediction module.

3. The industrial robot adaptive control system integrating a situation prediction mechanism according to claim 2, characterized in that: The state prediction module is used to sequentially number the raw state input data obtained by continuous sampling during the execution of the industrial robot, and to construct an embedded state sequence matrix according to a preset sliding window length. Each row in the state sequence matrix represents the distribution trajectory of the state parameters of adjacent time steps, and each column corresponds to the change dimension of a type of state parameter. Performing a time-coupled encoding operation on the state sequence matrix, and using a position offset nested function to calculate partial derivatives of the state parameters of each time step and the direction of change between the state parameters of the previous and next adjacent time steps, so as to construct a differential mapping tensor that characterizes the state evolution trend; Constructing a feedforward neural network having at least three hidden layers, wherein the hidden layers are embedded with response control units, each response control unit including a set of response gating functions constructed based on the gradient directionality of a state parameter, the response gating function being configured to receive a differential mapping tensor as an input and generate a compressed encoded state trend signal at an output of the response gating function; Training the feedforward neural network, using a trend residual function as a loss indicator during the training process, wherein the trend residual function is used to evaluate the trend deviation between the state trend signal and the trajectory direction formed by the industrial robot in the actual task path; and also including optimizing the connection weight parameters between each layer in the feedforward neural network using a backpropagation algorithm; After training is completed, the state input data obtained by real-time sampling is converted into a state sequence matrix within the current time window, and the difference mapping tensor is generated according to the same steps. The difference mapping tensor is input into the trained feedforward neural network, and the output includes the task movement direction vector, modal offset probability distribution, and path transfer rate prediction output results; The prediction output result is input into the determination and attribution module to provide support for the prediction of the task evolution path based on the state trend, and support the fuzzy logic reasoning model to make real-time attribution judgment on the critical control state.

4. The industrial robot adaptive control system integrating a situation prediction mechanism according to claim 3 is characterized in that: In the determination and attribution module, for each control mode, a fuzzy membership function with adjustable shape parameters is constructed. The fuzzy membership function is constructed in the form of piecewise nonlinear mapping. The membership starting point, membership peak point, and attenuation inflection point are set according to the state boundary conditions in the industrial robot task to describe the attribution transition characteristics of the state parameters in the modal boundary interval; Based on the state membership expression requirements under multiple control modes, the rule base structure of the fuzzy logic reasoning model is designed, and each fuzzy rule is expressed as a conditional association mapping between the input state parameter interval and the corresponding modal label.

5. The industrial robot adaptive control system integrating a situation prediction mechanism according to claim 4, characterized in that: In the judgment execution phase of the attribution judgment module, the current state input data is arranged in the order of the input dimensions preset in the fuzzy rule library structure, and the state parameters of each input dimension are fuzzy-encoded. The rule matching operation is performed on the fuzzy encoding results in sequence, marking all fuzzy rules that meet the prerequisites. The activation value of each activated fuzzy rule is calculated, and its corresponding control mode label is bound to the activation value. By performing weighted accumulation on the activation values ​​of all activated rules under the same control mode, the modal response coefficient corresponding to each control mode is generated. The vector set composed of the modal response coefficients is the intermediate representation tensor of multimodal fusion reasoning. A normalization operation is performed on the intermediate representation tensor composed of the modal response coefficients; based on the result of the normalization operation, a fuzzy membership value set containing all control modes is output, and the fuzzy membership value set serves as reference data for determining the distribution of the current state of the industrial robot among the control modes.

6. The industrial robot adaptive control system integrating a situation prediction mechanism according to claim 5, characterized in that: A collaborative discrimination model is constructed through a comparison module. The task trend prediction results output by the situation prediction module and the fuzzy membership values ​​output by the determination and attribution module are mapped one-to-one into a joint modal representation vector according to the control mode. The cosine value and normalized offset of the vector angle between the prediction channel of the situation prediction module and the fuzzy channel of the determination and attribution module of the joint modal representation vector are calculated, and the difference vector between the two in the control mode attribution dimension is extracted as the difference information for modal judgment.

7. The industrial robot adaptive control system integrating a situation prediction mechanism according to claim 6, characterized in that: When the misalignment identification module calculates the control decision entropy deviation index value of the difference information, the entropy deviation index value constructs an information entropy distribution based on the deviation degree of each modal dimension in the difference vector, and calculates the standard deviation entropy value of the information entropy distribution; The degree of deviation includes an absolute difference calculation operation between the predicted output result and the fuzzy membership value in each modal dimension, by taking the numerical difference between the predicted probability of the predicted output result and the corresponding fuzzy membership value under each control mode, and then normalizing the difference of all modal dimensions as the offset weight distribution, which is used to construct the probability proportion basis of each modal dimension in the control decision entropy deviation index value.

Images